{"tstamp": 1722305246.8396, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "6cf9d4db20f8436c8c7f2ccaca29d2fa", "0_model_name": "sentence-transformers/all-MiniLM-L6-v2", "0_prompt": "what is the difference between cluster point and limit point?", "0_output": [["what is the difference between cluster point and limit point?", "Askubuntu Stackexchange\n\nQ: How to enable user sharing per instructions? I have this entry in .xsession-errors log\n\"Nautilus-Share-Message: Called \"net usershare info\" but it failed: 'net usershare' returned \nerror 255: net usershare: cannot open usershare directory /var/lib/samba/usershares.         \nError No such file or directory\nPlease ask your system administrator to enable user sharing.\n\nThis is in relation to my applets having to be reloaded after every boot. Just wondering how I would enable user sharing, and how it affects my applets ??\n\nA: I don't think this effects your applets at all, that's probably a different bug. You could just make the directory with sudo:\nsudo mkdir -p /var/lib/samba/usershares\n\nBut I don't really know what it will do.\n"]], "0_corpus": "stackexchange", "1_conv_id": "38308d8cc72f4a24b8ddc6bafec79a96", "1_model_name": "BAAI/bge-large-en-v1.5", "1_prompt": "what is the difference between cluster point and limit point?", "1_output": [["what is the difference between cluster point and limit point?", "Math Stackexchange\n\nQ: Limit and limit points What is the basic difference between limit and limit points, and if a sequence has one unique limit how it can have a lot of limit points \n\nA: A limit point is a generalization of a limit (each limit is a limit point but not vice versa). You can see this in the definition:\nLimit: $a$ is a limit of $(a_n)$, iff in each neighborhood of $a$ are almost all elements of $(a_n)$.\nLimit point: $a$ is a limit of $(a_n)$, iff in each neighborhood of $a$ are infinite elements of $(a_n)$.\nI highlighted the difference between both definitions.\nBecause only infinite elements need to be in any neighborhood of $a$. There can be infinite elements outside this neighborhood which can clustered to another limit point. This is the reason why there might be many limit points for a sequence. If $a$ is a limit only finite elements can be outside any given neighborhood which is not enough to go to another limit.\n"]], "1_corpus": "stackexchange"}
{"tstamp": 1722306411.5045, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "5993407c26d04c54981a49217a38518d", "0_model_name": "Alibaba-NLP/gte-Qwen2-7B-instruct", "0_prompt": "when did scotland last qualify for world cup", "0_output": [["when did scotland last qualify for world cup", "Scotland national football team\n\nScotland lost just two of their first 43 international matches. It was not until a 2–0 home defeat by Ireland in 1903 that Scotland lost a match to a team other than England. This run of success meant that Scotland would have regularly topped the Elo ratings, which were calculated in 1998, between 1876 and 1904. Scotland won the British Home Championship outright on 24 occasions, and shared the title 17 times with at least one other team. A noteworthy victory for Scotland before the Second World War was the 5–1 victory over England in 1928, which led to that Scotland side being known as the \"Wembley Wizards\". Scotland played their first match outside the British Isles in 1929, beating Norway 7–3 in Bergen. Scotland continued to contest regular friendly matches against European opposition and enjoyed wins against Germany and France before losing to the Austrian \"Wunderteam\" and Italy in 1931.\nScotland, like the other Home Nations, did not enter the three FIFA World Cups held during the 1930s. This was because the four associations had been excluded from FIFA due to a disagreement regarding the status of amateur players. The four associations, including Scotland, returned to the FIFA fold after the Second World War. A match between a United Kingdom team and a \"Rest of the World\" team was played at Hampden Park in 1947 to celebrate this reconciliation.\n1950s: Entering the World Cup"]], "0_corpus": "wikipedia", "1_conv_id": "940948f56e9c4f5297cb17f69935bd8b", "1_model_name": "Salesforce/SFR-Embedding-2_R", "1_prompt": "when did scotland last qualify for world cup", "1_output": [["when did scotland last qualify for world cup", "Scotland at the FIFA World Cup\n\nThe World Cup consists of two parts, the qualification phase and the final phase (officially called the World Cup Finals). The qualification phase, which currently take place over the three years preceding the Finals, is used to determine which teams qualify for the Finals. The current format of the Finals involves 32 teams competing at venues within the host nation (or nations) over a period of about a month. The World Cup Finals is the most widely viewed sporting event in the world, with an estimated 715.1 million people watching the 2006 tournament final.\nHistory\n1930s\nScotland did not compete in the first three World Cup competitions, in 1930, 1934 and 1938. Because of a dispute with FIFA over \"broken-time\" payments to players, the SFA, with the football associations of England, Ireland and Wales, withdrew from FIFA in 1928, and did not rejoin as a permanent member until 1946.\n1950s\nThe readmission of the Scottish Football Association to FIFA in 1946 meant that Scotland were now eligible to enter the 1950 FIFA World Cup. FIFA advised that places would be awarded to the top two teams in the 1949–50 British Home Championship, but the SFA announced that Scotland would attend the finals only if Scotland won the competition. Scotland won their first two matches, but a 1–0 home defeat by England meant that the Scots finished as runners-up. The SFA stood by this proclamation, despite pleas to the contrary by the Scotland players, supported by England captain Billy Wright and the other England players. Sportswriter Brian Glanville later described the SFA as showing \"baffling insularity and pique\" in their refusal to send a team to the World Cup."]], "1_corpus": "wikipedia"}
{"tstamp": 1722307573.5271, "task_type": "retrieval", "type": "tievote", "models": ["", ""], "ip": "", "0_conv_id": "9deb3465ffd3458081570489c3810813", "0_model_name": "mixedbread-ai/mxbai-embed-large-v1", "0_prompt": "are opened or are open?", "0_output": [["are opened or are open?", "Ell Stackexchange\n\nQ: Are they the same in meaning, the open door and the opened door? What is the difference between \"the open door\" and \"the opened door\"? Are they the same in meaning? Please explain it. Thanks a lot.\n\nA: Open is an adjective. \nOpened is a past participle used as a modifier. \nThe difference in meaning is that open is in contrast to not open. The door  could have been open for ten years or a century. It could always be open. \nOpened expands to The \"having been opened\" door...\nThis refers to the action of the door having been opened. The door used to be shut, now it is open. It has been opened. It has not always been open. It could have been opened ten years ago or two  seconds ago, but at some point it was closed. \n"]], "0_corpus": "stackexchange", "1_conv_id": "9913e8b8325c4550a0bfb3f5298189d2", "1_model_name": "BM25", "1_prompt": "are opened or are open?", "1_output": [["are opened or are open?", "Ell Stackexchange\n\nQ: Are they the same in meaning, the open door and the opened door? What is the difference between \"the open door\" and \"the opened door\"? Are they the same in meaning? Please explain it. Thanks a lot.\n\nA: Open is an adjective. \nOpened is a past participle used as a modifier. \nThe difference in meaning is that open is in contrast to not open. The door  could have been open for ten years or a century. It could always be open. \nOpened expands to The \"having been opened\" door...\nThis refers to the action of the door having been opened. The door used to be shut, now it is open. It has been opened. It has not always been open. It could have been opened ten years ago or two  seconds ago, but at some point it was closed. \n"]], "1_corpus": "stackexchange"}
{"tstamp": 1722309466.1126, "task_type": "retrieval", "type": "leftvote", "models": ["", ""], "ip": "", "0_conv_id": "bc8a08f1718846e58aa907a16ce9620b", "0_model_name": "GritLM/GritLM-7B", "0_prompt": "where did german immigrants settled in the 1800s", "0_output": [["where did german immigrants settled in the 1800s", "German Americans\n\nA particularly attractive destination was Milwaukee, which came to be known as \"the German Athens\". Radical Germans trained in politics in the old country dominated the city's Socialists. Skilled workers dominated many crafts, while entrepreneurs created the brewing industry; the most famous brands included Pabst, Schlitz, Miller, and Blatz.\nWhereas half of German immigrants settled in cities, the other half established farms in the Midwest. From Ohio to the Plains states, a heavy presence persists in rural areas into the 21st century.\nDeep South\nFew German immigrants settled in the Deep South, apart from New Orleans, the German Coast, and Texas.\nTexas\nTexas attracted many Germans who entered through Galveston and Indianola, both those who came to farm, and later immigrants who more rapidly took industrial jobs in cities such as Houston. As in Milwaukee, Germans in Houston built the brewing industry. By the 1920s, the first generation of college-educated German Americans were moving into the chemical and oil industries.\nTexas had about 20,000 German Americans in the 1850s. They did not form a uniform bloc, but were highly diverse and drew from geographic areas and all sectors of European society, except that very few aristocrats or upper middle class businessmen arrived. In this regard, Texas Germania was a microcosm of the Germania nationwide."]], "0_corpus": "wikipedia", "1_conv_id": "b7cfab5df92f40cb97e749a1057688d3", "1_model_name": "sentence-transformers/all-MiniLM-L6-v2", "1_prompt": "where did german immigrants settled in the 1800s", "1_output": [["where did german immigrants settled in the 1800s", "Immigration to Germany\n\nImmigration to Germany, both in the country's modern borders and the many political entities that preceded it, has occurred throughout the country's history. Today, Germany is one of the most popular destinations for immigrants in the world, with well over 1 million people moving there each year since 2013. As of 2019, around 13.7 million people living in Germany, or about 17% of the population, are first-generation immigrants."]], "1_corpus": "wikipedia"}
{"tstamp": 1722310078.9673, "task_type": "retrieval", "type": "tievote", "models": ["", ""], "ip": "", "0_conv_id": "723e06bd28704a6b90a00a505ecb7438", "0_model_name": "intfloat/multilingual-e5-large-instruct", "0_prompt": "A new method called ATTEMPT utilizing attentional mixtures of soft prompts for knowledge transfer.", "0_output": [["A new method called ATTEMPT utilizing attentional mixtures of soft prompts for knowledge transfer.", "Title: ATTEMPT: Parameter-Efficient Multi-task Tuning via Attentional Mixtures of Soft Prompts\n\nAbstract: This work introduces a new multi-task, parameter-efficient language model (LM) tuning method that learns to transfer knowledge across different tasks via a mixture of soft prompts-small prefix embedding vectors pre-trained for different tasks. Our method, called ATTEMPT (ATTEntional Mixtures of Prompt Tuning), obtains source prompts as encodings of large-scale source tasks into a small number of parameters and trains an attention module to interpolate the source prompts and a newly initialized target prompt for every instance in the target task. During training, only the target task prompt and the attention weights, which are shared between tasks in multi-task training, are updated, while the original LM and source prompts are intact. ATTEMPT is highly parameter-efficient (e.g., updates 2,300 times fewer parameters than full fine-tuning) while achieving high task performance using knowledge from high-resource tasks. Moreover, it is modular using pre-trained soft prompts, and can flexibly add or remove source prompts for effective knowledge transfer. Our experimental results across 21 diverse NLP datasets show that ATTEMPT significantly outperforms prompt tuning and outperforms or matches fully fine-tuned or other parameter-efficient tuning approaches that use over ten times more parameters. Finally, ATTEMPT outperforms previous work in few-shot learning settings."]], "0_corpus": "arxiv", "1_conv_id": "6584a2b86688459ca86f0cbfc241df00", "1_model_name": "mixedbread-ai/mxbai-embed-large-v1", "1_prompt": "A new method called ATTEMPT utilizing attentional mixtures of soft prompts for knowledge transfer.", "1_output": [["A new method called ATTEMPT utilizing attentional mixtures of soft prompts for knowledge transfer.", "Title: ATTEMPT: Parameter-Efficient Multi-task Tuning via Attentional Mixtures of Soft Prompts\n\nAbstract: This work introduces a new multi-task, parameter-efficient language model (LM) tuning method that learns to transfer knowledge across different tasks via a mixture of soft prompts-small prefix embedding vectors pre-trained for different tasks. Our method, called ATTEMPT (ATTEntional Mixtures of Prompt Tuning), obtains source prompts as encodings of large-scale source tasks into a small number of parameters and trains an attention module to interpolate the source prompts and a newly initialized target prompt for every instance in the target task. During training, only the target task prompt and the attention weights, which are shared between tasks in multi-task training, are updated, while the original LM and source prompts are intact. ATTEMPT is highly parameter-efficient (e.g., updates 2,300 times fewer parameters than full fine-tuning) while achieving high task performance using knowledge from high-resource tasks. Moreover, it is modular using pre-trained soft prompts, and can flexibly add or remove source prompts for effective knowledge transfer. Our experimental results across 21 diverse NLP datasets show that ATTEMPT significantly outperforms prompt tuning and outperforms or matches fully fine-tuned or other parameter-efficient tuning approaches that use over ten times more parameters. Finally, ATTEMPT outperforms previous work in few-shot learning settings."]], "1_corpus": "arxiv"}
{"tstamp": 1722310105.1175, "task_type": "retrieval", "type": "tievote", "models": ["", ""], "ip": "", "0_conv_id": "daca567e0fff4693a219f1ccdcb2330c", "0_model_name": "BAAI/bge-large-en-v1.5", "0_prompt": "who votes for nba defensive player of year", "0_output": [["who votes for nba defensive player of year", "NBA Defensive Player of the Year Award\n\nThe NBA Defensive Player of the Year Award is an annual National Basketball Association (NBA) award given since the 1982–83 NBA season to the best defensive player of the regular season. The winner is selected by a panel of 124 sportswriters and broadcasters throughout the United States and Canada, each of whom casts a vote for first, second and third place selections. Each first-place vote is worth five points, second-place votes are worth three points, and a third-place vote is worth one. The player with the highest point total, regardless of the number of first-place votes, wins the award. Since the 2022–23 NBA season, winners receive the Hakeem Olajuwon Trophy, named after the two-time defensive player of the year winner.\nDikembe Mutombo, Ben Wallace, and Rudy Gobert have each won the award a record four times. Dwight Howard has won the award three times, and is the only player ever to have won it in three consecutive seasons. Sidney Moncrief, Mark Eaton, Dennis Rodman, Hakeem Olajuwon, Alonzo Mourning and Kawhi Leonard have each won it twice. The most recent award recipient is Rudy Gobert of the Minnesota Timberwolves."]], "0_corpus": "wikipedia", "1_conv_id": "2eb4d1f035c0457d8b0f19fc28c2af8b", "1_model_name": "text-embedding-3-large", "1_prompt": "who votes for nba defensive player of year", "1_output": [["who votes for nba defensive player of year", "NBA Defensive Player of the Year Award\n\nThe NBA Defensive Player of the Year Award is an annual National Basketball Association (NBA) award given since the 1982–83 NBA season to the best defensive player of the regular season. The winner is selected by a panel of 124 sportswriters and broadcasters throughout the United States and Canada, each of whom casts a vote for first, second and third place selections. Each first-place vote is worth five points, second-place votes are worth three points, and a third-place vote is worth one. The player with the highest point total, regardless of the number of first-place votes, wins the award. Since the 2022–23 NBA season, winners receive the Hakeem Olajuwon Trophy, named after the two-time defensive player of the year winner.\nDikembe Mutombo, Ben Wallace, and Rudy Gobert have each won the award a record four times. Dwight Howard has won the award three times, and is the only player ever to have won it in three consecutive seasons. Sidney Moncrief, Mark Eaton, Dennis Rodman, Hakeem Olajuwon, Alonzo Mourning and Kawhi Leonard have each won it twice. The most recent award recipient is Rudy Gobert of the Minnesota Timberwolves."]], "1_corpus": "wikipedia"}
{"tstamp": 1722310164.7577, "task_type": "retrieval", "type": "bothbadvote", "models": ["", ""], "ip": "", "0_conv_id": "fc308a6564e84da2b576fda8f5b0f500", "0_model_name": "jinaai/jina-embeddings-v2-base-en", "0_prompt": "Study on improving computational efficiency and reducing parameters in recommender systems using prompt tuning and pruning techniques", "0_output": [["Study on improving computational efficiency and reducing parameters in recommender systems using prompt tuning and pruning techniques", "Title: Spatio-temporal dynamics and plastic flow of vortices in superconductors with periodic arrays of pinning sites\n\nAbstract: We present simulations of flux-gradient-driven superconducting rigid vortices interacting with square and triangular arrays of columnar pinning sites in an increasing external magnetic field. These simulations allow us to quantitatively relate spatio-temporal microscopic information of the vortex lattice with typically measured macroscopic quantities, such as the magnetization $M(H)$. The flux lattice does not become completely commensurate with the pinning sites throughout the sample at the magnetization matching peaks, but forms a commensurate lattice in a region close to the edge of the sample. Matching fields related to unstable vortex configurations do not produce peaks in $M(H)$. We observe a variety of evolving complex flux profiles, including flat terraces or plateaus separated by winding current-carrying strings and, near the peaks in $M(H)$, plateaus only in certain regions, which move through the sample as the field increases."]], "0_corpus": "arxiv", "1_conv_id": "659b5d7e65414017a44941dc4860d179", "1_model_name": "nomic-ai/nomic-embed-text-v1.5", "1_prompt": "Study on improving computational efficiency and reducing parameters in recommender systems using prompt tuning and pruning techniques", "1_output": [["Study on improving computational efficiency and reducing parameters in recommender systems using prompt tuning and pruning techniques", "Title: Local Hypercomplex Analyticity\n\nAbstract: The notion of analyticity is studied in the context of hypercomplex numbers. A critical review of the problems arising from the conventional approach is given. We describe a local analyticity condition which yields the desired type of hypercomplex solutions. The result is the definition of a generalized complex analyticity to hypercomplex space. 02.10.Tq/Vr, 02.30.-f/Dk, 02.90.+p"]], "1_corpus": "arxiv"}
{"tstamp": 1722310269.5459, "task_type": "retrieval", "type": "leftvote", "models": ["", ""], "ip": "", "0_conv_id": "356862c28de946ae8a434438c62a06e4", "0_model_name": "GritLM/GritLM-7B", "0_prompt": "diablo 2 is not working on windows 7?", "0_output": [["diablo 2 is not working on windows 7?", "Gaming Stackexchange\n\nQ: Playing Diablo 1 on Windows 7 I just found an old copy of Diablo (the first game) in a box and whacked it into my laptop I purchased just yesterday.\nAfter the quick install I hit play and found that the screen remained black (nothing was rendering). There was however sound.\nI downloaded the patch but still nothing.\nWith it being a new laptop, I figure I might be missing some required software like Direct X maybe? I'm not really sure.\nIf anyone has any suggestions or knows how to fix this issue, that would be really helpful.\n\nA: This may sound weird (and I have actually done this to play Diablo on my Win 7 laptop), but right click on your desktop and click on Screen Resolution. Leave that open and then run Diablo.\n"]], "0_corpus": "stackexchange", "1_conv_id": "375b8963e38a4bd191128977a52d9b97", "1_model_name": "BM25", "1_prompt": "diablo 2 is not working on windows 7?", "1_output": [["diablo 2 is not working on windows 7?", "Gaming Stackexchange\n\nQ: How do I configure Diablo 2 in Wine to -skiptobnet and -w? I am using Ubuntu Gnome and Diablo 2 works ok with Wine. How do I add shortcut options to the game? I want the -skiptobnet and -w for windowed mode.\nPath is ~/.wine/drive_c/Program Files (x86)/Diablo II/Game.exe.\n\nA: I was able to get options to work using the windows .lnk file that was created in the wine start menu directory during installation.\nYou can launch Diablo II through terminal or create a new launcher using this command replacing PATH_TO_WINE and PATH_TO_SHORTCUT:\nenv WINEPREFIX=\"PATH_TO_WINE\" wine C:\\\\windows\\\\command\\\\start.exe PATH_TO_SHORTCUT -skiptobnet -w\n\nFor example:\nenv WINEPREFIX=\"/home/nonterrorist/.wine\" wine C:\\\\windows\\\\command\\\\start.exe /Unix /home/nonterrorist/.wine/dosdevices/c:/users/Public/Start\\ Menu/Programs/Diablo\\ II/Diablo\\ II\\ -\\ Lord\\ of\\ Destruction.lnk -skiptobnet -w\n\nNote in the example above, you will need to escape any spaces in the shortcut path using \\\n"]], "1_corpus": "stackexchange"}
{"tstamp": 1722310291.9285, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "9776843ed00d4cab89c0894cad832500", "0_model_name": "sentence-transformers/all-MiniLM-L6-v2", "0_prompt": "what are chapters in the bible called?", "0_output": [["what are chapters in the bible called?", "Stackoverflow Stackexchange\n\nQ: How to strip executables thoroughly I'd like to strip as much as I can - on Linux: an ELF. I only want in there the stuff I need to run it.\nI tried using strip:\nstrip --strip-all elf\n\nBut it doesn't seem to do a good job: nm still displays lots of stuff, and the binary is still big.\nWhat should I do?\n\nA: I would check out this great article that goes into depth on making an ELF executable as small as possible. Maybe it has info that could help!\nhttp://www.muppetlabs.com/~breadbox/software/tiny/teensy.html\n"]], "0_corpus": "stackexchange", "1_conv_id": "87810785bece4d6dab306293a64980f2", "1_model_name": "BAAI/bge-large-en-v1.5", "1_prompt": "what are chapters in the bible called?", "1_output": [["what are chapters in the bible called?", "English Stackexchange\n\nQ: Why are the divisions of the Bible called \"verses\"? In common parlance, a verse is a writing arranged with a metrical rhythm, typically having a rhyme. Most of the books of Bible contain prose which do not follow metrical rhythm or rhyme. But their sentences are called verses. My question therefore is: Why are the divisions of the Bible called verses?\n\nA: A number of questions  on Stack Exchange - English Language and Usage have commented on the words 'stanza' and 'verse' in relation to poetry and to scripture.\n'Verses' are also used in Shakespeare.\nMerriam Webster defines a 'verse' in three ways : metrical verse, a stanza or a biblical verse.\nIn any work which requires to be referenced and quoted as part of its function, stanzas or verses are numbered accordingly so that they can be cited.\n"]], "1_corpus": "stackexchange"}
{"tstamp": 1722310355.5437, "task_type": "retrieval", "type": "tievote", "models": ["", ""], "ip": "", "0_conv_id": "256e5bb6399a4ba7b599c9ba9a402e95", "0_model_name": "embed-english-v3.0", "0_prompt": "Learning word representations to improve IR tasks like query expansion and classification", "0_output": [["Learning word representations to improve IR tasks like query expansion and classification", "Title: Relevance-based Word Embedding\n\nAbstract: Learning a high-dimensional dense representation for vocabulary terms, also known as a word embedding, has recently attracted much attention in natural language processing and information retrieval tasks. The embedding vectors are typically learned based on term proximity in a large corpus. This means that the objective in well-known word embedding algorithms, e.g., word2vec, is to accurately predict adjacent word(s) for a given word or context. However, this objective is not necessarily equivalent to the goal of many information retrieval (IR) tasks. The primary objective in various IR tasks is to capture relevance instead of term proximity, syntactic, or even semantic similarity. This is the motivation for developing unsupervised relevance-based word embedding models that learn word representations based on query-document relevance information. In this paper, we propose two learning models with different objective functions; one learns a relevance distribution over the vocabulary set for each query, and the other classifies each term as belonging to the relevant or non-relevant class for each query. To train our models, we used over six million unique queries and the top ranked documents retrieved in response to each query, which are assumed to be relevant to the query. We extrinsically evaluate our learned word representation models using two IR tasks: query expansion and query classification. Both query expansion experiments on four TREC collections and query classification experiments on the KDD Cup 2005 dataset suggest that the relevance-based word embedding models significantly outperform state-of-the-art proximity-based embedding models, such as word2vec and GloVe."]], "0_corpus": "arxiv", "1_conv_id": "2f7dc574c8b94ae7a397af80c4cf05f5", "1_model_name": "mixedbread-ai/mxbai-embed-large-v1", "1_prompt": "Learning word representations to improve IR tasks like query expansion and classification", "1_output": [["Learning word representations to improve IR tasks like query expansion and classification", "Title: Relevance-based Word Embedding\n\nAbstract: Learning a high-dimensional dense representation for vocabulary terms, also known as a word embedding, has recently attracted much attention in natural language processing and information retrieval tasks. The embedding vectors are typically learned based on term proximity in a large corpus. This means that the objective in well-known word embedding algorithms, e.g., word2vec, is to accurately predict adjacent word(s) for a given word or context. However, this objective is not necessarily equivalent to the goal of many information retrieval (IR) tasks. The primary objective in various IR tasks is to capture relevance instead of term proximity, syntactic, or even semantic similarity. This is the motivation for developing unsupervised relevance-based word embedding models that learn word representations based on query-document relevance information. In this paper, we propose two learning models with different objective functions; one learns a relevance distribution over the vocabulary set for each query, and the other classifies each term as belonging to the relevant or non-relevant class for each query. To train our models, we used over six million unique queries and the top ranked documents retrieved in response to each query, which are assumed to be relevant to the query. We extrinsically evaluate our learned word representation models using two IR tasks: query expansion and query classification. Both query expansion experiments on four TREC collections and query classification experiments on the KDD Cup 2005 dataset suggest that the relevance-based word embedding models significantly outperform state-of-the-art proximity-based embedding models, such as word2vec and GloVe."]], "1_corpus": "arxiv"}
{"tstamp": 1722310388.2572, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "8c01a7990c7d433c8f9c9a38ccfd45b3", "0_model_name": "voyage-multilingual-2", "0_prompt": "who starred in the film far from the madding crowd", "0_output": [["who starred in the film far from the madding crowd", "Far from the Madding Crowd (2015 film)\n\nPlot The film is set in 1870 in Britain. While working on her aunt's farm in Dorset, Bathsheba Everdene (Carey Mulligan) meets neighbouring farmer, Gabriel Oak (Matthias Schoenaerts). He eventually proposes, but the headstrong Bathsheba declines, saying she is too independent. One night, Gabriel's new sheepdog herds his entire flock off a steep cliff. Penniless, Gabriel leaves to search for work. In contrast, Bathsheba inherits an uncle's farm and leaves to run it.\nWhile at a fair trying to find employment, Gabriel sees recruiting officers. A girl there, Fanny Robbin, points out one soldier, Sergeant Frank Troy, her sweetheart. She suggests Gabriel seek employment at a farm in Weatherbury. Upon Gabriel's arrival, several buildings are on fire. He saves the barn from destruction. At dawn the next day, he meets the farm's new mistress, Bathsheba, who hires him as a shepherd.\nWhile in town trading her seed, Bathsheba sees her neighbour William Boldwood (Michael Sheen), a prosperous and mature bachelor. Bathsheba sends Boldwood a Valentine as a joke. He is both offended and intrigued. In the meantime, Fanny goes to the wrong church for her wedding; Troy, believing he's been jilted, is devastated. Boldwood comes to speak to Bathsheba, sees her handling sheep, and is charmed. He invites her to his home, where he proposes marriage. Bathsheba delays giving him a final answer, and Gabriel admonishes her for toying with Boldwood's affections. Stung by the criticism, she fires him. The next day, a crisis arises with the sheep that only Gabriel can manage. She successfully persuades him to return."]], "0_corpus": "wikipedia", "1_conv_id": "2dafecf528e44c34a578dd2bad99f101", "1_model_name": "GritLM/GritLM-7B", "1_prompt": "who starred in the film far from the madding crowd", "1_output": [["who starred in the film far from the madding crowd", "Far from the Madding Crowd (2015 film)\n\nFar from the Madding Crowd is a 2015 British romantic drama film directed by Thomas Vinterberg and starring Carey Mulligan, Matthias Schoenaerts, Tom Sturridge, Michael Sheen, and Juno Temple. An adaptation by David Nicholls of the 1874 novel Far from the Madding Crowd by Thomas Hardy, it is the fourth film adaptation of the novel."]], "1_corpus": "wikipedia"}
{"tstamp": 1722310737.3362, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "0494368f367f4ce891d1645f49397dce", "0_model_name": "text-embedding-004", "0_prompt": "what happened to sokka after the last airbender", "0_output": [["what happened to sokka after the last airbender", "Sokka\n\nSokka is a fictional character in Nickelodeon's animated television series Avatar: The Last Airbender and its sequel series The Legend of Korra. The character, created by Michael Dante DiMartino and Bryan Konietzko, is voiced by Jack DeSena in the original series and by Chris Hardwick in the sequel series. He is a warrior of the Water Tribe and the son of Chief Hakoda and Kya. In the live-action film adaptation, he is portrayed by Jackson Rathbone, while in the  live-action television series adaptation, he is portrayed by Ian Ousley."]], "0_corpus": "wikipedia", "1_conv_id": "2a889447fde54cc5b596b7983ab9b19e", "1_model_name": "GritLM/GritLM-7B", "1_prompt": "what happened to sokka after the last airbender", "1_output": [["what happened to sokka after the last airbender", "List of Avatar: The Last Airbender characters\n\nSokka Sokka () (voiced by Jack DeSena in the original series, Chris Hardwick in the sequel series, portrayed by Jackson Rathbone in the live-action film, Ian Ousley in the live-action series) is a 15-year-old warrior of the Southern Water Tribe, and Katara's older brother. With no bending power of his own, Sokka relies largely on a metallic boomerang, a blunt metal club, a machete, and later a black jian, or sword, created from the metals of a meteorite. Surprisingly in an inhabitant of a mystical world, Sokka is an engineer and something of a jack-of-all-trades, in which respect he is easily able to understand the Fire Nation's advanced technology, and perfects the design of the hot air balloon. In addition, he is both heterodox and resourceful in his endeavors, and a source of comic relief throughout the series. Sokka was in love with the Northern Water Tribe princess Yue at the end of Book One and later shifted his affections to the Kyoshi Warriors' leader Suki in Books Two and Three. In the sequel series, flashbacks reveal Sokka was the first representative of the Southern Water Tribe to sit on the Republic City Council, and possibly its first chairman. He died a few years after Aang, when the next Avatar, Korra, was still a child."]], "1_corpus": "wikipedia"}
{"tstamp": 1722310798.4596, "task_type": "retrieval", "type": "leftvote", "models": ["", ""], "ip": "", "0_conv_id": "2ff5ab5b630040549b9bacc23cc1ea95", "0_model_name": "GritLM/GritLM-7B", "0_prompt": "do fled pokemon come back?", "0_output": [["do fled pokemon come back?", "Gaming Stackexchange\n\nQ: Do I have to physically return a gym to retrieve a pokemon when the gym is defeated? I'm thinking about leaving my Vaporeon at the gym near where I work. I know that I might be lucky to keep it there one or two days, but darnit, I wanna do it. So if when the gym falls, do I need to physically return to it to get my pokemon back? \n\nA: No you do not! It returns to you automatically. As of Android version 0.35.0/iOS version 1.50, ousted defenders return to you with 0 HP, per the patch notes.\n"]], "0_corpus": "stackexchange", "1_conv_id": "f684351f5e9a4cbd9e9e2e62282e624c", "1_model_name": "jinaai/jina-embeddings-v2-base-en", "1_prompt": "do fled pokemon come back?", "1_output": [["do fled pokemon come back?", "Stackoverflow Stackexchange\n\nQ: How to find out cl.exe's built-in macros Does anyone know how could I find out which are cl.exe's builtin/predefined macros?\nFor example for gcc the following command line will list all the compiler's builtin macros\ngcc -dM -E - </dev/null\n\nEDIT: I'm interested in a way similar to gcc's that is \"ask the actual compiler\".\nThanks\n\nA: /P preprocessor flag will emit the currently active macros based on the project build settings. I am not sure if it is exactly the equivalent of gcc command you have shown. The output is in .I file.\n"]], "1_corpus": "stackexchange"}
{"tstamp": 1722310831.0005, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "3495497b338a42f68c1ed1ff16f8d3ab", "0_model_name": "BM25", "0_prompt": "a great benchmark for text embeddings", "0_output": [["a great benchmark for text embeddings", "Title: Recent advances in text embedding: A Comprehensive Review of Top-Performing Methods on the MTEB Benchmark\n\nAbstract: Text embedding methods have become increasingly popular in both industrial and academic fields due to their critical role in a variety of natural language processing tasks. The significance of universal text embeddings has been further highlighted with the rise of Large Language Models (LLMs) applications such as Retrieval-Augmented Systems (RAGs). While previous models have attempted to be general-purpose, they often struggle to generalize across tasks and domains. However, recent advancements in training data quantity, quality and diversity; synthetic data generation from LLMs as well as using LLMs as backbones encourage great improvements in pursuing universal text embeddings. In this paper, we provide an overview of the recent advances in universal text embedding models with a focus on the top performing text embeddings on Massive Text Embedding Benchmark (MTEB). Through detailed comparison and analysis, we highlight the key contributions and limitations in this area, and propose potentially inspiring future research directions."]], "0_corpus": "arxiv", "1_conv_id": "732203ab60e34d408f5ea6db08909f9a", "1_model_name": "text-embedding-004", "1_prompt": "a great benchmark for text embeddings", "1_output": [["a great benchmark for text embeddings", "Title: MTEB: Massive Text Embedding Benchmark\n\nAbstract: Text embeddings are commonly evaluated on a small set of datasets from a single task not covering their possible applications to other tasks. It is unclear whether state-of-the-art embeddings on semantic textual similarity (STS) can be equally well applied to other tasks like clustering or reranking. This makes progress in the field difficult to track, as various models are constantly being proposed without proper evaluation. To solve this problem, we introduce the Massive Text Embedding Benchmark (MTEB). MTEB spans 8 embedding tasks covering a total of 58 datasets and 112 languages. Through the benchmarking of 33 models on MTEB, we establish the most comprehensive benchmark of text embeddings to date. We find that no particular text embedding method dominates across all tasks. This suggests that the field has yet to converge on a universal text embedding method and scale it up sufficiently to provide state-of-the-art results on all embedding tasks. MTEB comes with open-source code and a public leaderboard at https://github.com/embeddings-benchmark/mteb."]], "1_corpus": "arxiv"}
{"tstamp": 1722310835.8058, "task_type": "retrieval", "type": "bothbadvote", "models": ["", ""], "ip": "", "0_conv_id": "51662d2391894809a37610cb3c419e8a", "0_model_name": "intfloat/multilingual-e5-large-instruct", "0_prompt": "what is the difference between field and sigma field?", "0_output": [["what is the difference between field and sigma field?", "Math Stackexchange\n\nQ: $\\sigma$-algebra vs. $\\sigma$-field: is there any difference? The subject says it all: is there any difference between the two concepts of $\\sigma$-algebra and $\\sigma$-field? In probability theory, they seem to be used more or less interchangeably. If there is no difference, is there any historical reason why some people/schools use the term $\\sigma$-algebra, while others use the term $\\sigma-$field?\n\nA: If you want to be completely sure you can see Taylor´s Introduction to Measure and Integration where he states that both concepts refer to the same thing.\n"]], "0_corpus": "stackexchange", "1_conv_id": "c60e866a11304ec09c51a718d9b9bbb8", "1_model_name": "mixedbread-ai/mxbai-embed-large-v1", "1_prompt": "what is the difference between field and sigma field?", "1_output": [["what is the difference between field and sigma field?", "Math Stackexchange\n\nQ: $\\sigma$-algebra vs. $\\sigma$-field: is there any difference? The subject says it all: is there any difference between the two concepts of $\\sigma$-algebra and $\\sigma$-field? In probability theory, they seem to be used more or less interchangeably. If there is no difference, is there any historical reason why some people/schools use the term $\\sigma$-algebra, while others use the term $\\sigma-$field?\n\nA: If you want to be completely sure you can see Taylor´s Introduction to Measure and Integration where he states that both concepts refer to the same thing.\n"]], "1_corpus": "stackexchange"}
{"tstamp": 1722310863.6196, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "4f330aeb5b2b442d8190b75b12ed1f29", "0_model_name": "nomic-ai/nomic-embed-text-v1.5", "0_prompt": "New search algorithm inspired by AlphaZero for better theorem proving results", "0_output": [["New search algorithm inspired by AlphaZero for better theorem proving results", "Title: On the propagation of semiclassical Wigner functions\n\nAbstract: We establish the difference between the propagation of semiclassical Wigner functions and classical Liouville propagation. First we re-discuss the semiclassical limit for the propagator of Wigner functions, which on its own leads to their classical propagation. Then, via stationary phase evaluation of the full integral evolution equation, using the semiclassical expressions of Wigner functions, we provide the correct geometrical prescription for their semiclassical propagation. This is determined by the classical trajectories of the tips of the chords defined by the initial semiclassical Wigner function and centered on their arguments, in contrast to the Liouville propagation which is determined by the classical trajectories of the arguments themselves."]], "0_corpus": "arxiv", "1_conv_id": "86fc44429afe4954a77719d25bf4ef90", "1_model_name": "GritLM/GritLM-7B", "1_prompt": "New search algorithm inspired by AlphaZero for better theorem proving results", "1_output": [["New search algorithm inspired by AlphaZero for better theorem proving results", "Title: HyperTree Proof Search for Neural Theorem Proving\n\nAbstract: We propose an online training procedure for a transformer-based automated theorem prover. Our approach leverages a new search algorithm, HyperTree Proof Search (HTPS), inspired by the recent success of AlphaZero. Our model learns from previous proof searches through online training, allowing it to generalize to domains far from the training distribution. We report detailed ablations of our pipeline's main components by studying performance on three environments of increasing complexity. In particular, we show that with HTPS alone, a model trained on annotated proofs manages to prove 65.4% of a held-out set of Metamath theorems, significantly outperforming the previous state of the art of 56.5% by GPT-f. Online training on these unproved theorems increases accuracy to 82.6%. With a similar computational budget, we improve the state of the art on the Lean-based miniF2F-curriculum dataset from 31% to 42% proving accuracy."]], "1_corpus": "arxiv"}
{"tstamp": 1722310892.1304, "task_type": "retrieval", "type": "tievote", "models": ["", ""], "ip": "", "0_conv_id": "c998ef114db443b0a8ea43dcaa79997c", "0_model_name": "GritLM/GritLM-7B", "0_prompt": "can I put metal plate in a microwave", "0_output": [["can I put metal plate in a microwave", "Microwave oven\n\nMetal objects Contrary to popular assumptions, metal objects can be safely used in a microwave oven, but with some restrictions. Any metal or conductive object placed into the microwave oven acts as an antenna to some degree, resulting in an electric current. This causes the object to act as a heating element. This effect varies with the object's shape and composition, and is sometimes utilized for cooking.\nAny object containing pointed metal can create an electric arc (sparks) when microwaved. This includes cutlery, crumpled aluminium foil (though some foil used in microwave ovens is safe, see below), twist-ties containing metal wire, the metal wire carry-handles in oyster pails, or almost any metal formed into a poorly conductive foil or thin wire, or into a pointed shape. Forks are a good example: the tines of the fork respond to the electric field by producing high concentrations of electric charge at the tips. This has the effect of exceeding the dielectric breakdown of air, about 3 megavolts per meter (3×106 V/m). The air forms a conductive plasma, which is visible as a spark. The plasma and the tines may then form a conductive loop, which may be a more effective antenna, resulting in a longer lived spark. When dielectric breakdown occurs in air, some ozone and nitrogen oxides are formed, both of which are unhealthy in large quantities."]], "0_corpus": "wikipedia", "1_conv_id": "088ae8addacd4fa3a40cb59053e8abe6", "1_model_name": "BAAI/bge-large-en-v1.5", "1_prompt": "can I put metal plate in a microwave", "1_output": [["can I put metal plate in a microwave", "Microwave oven\n\nMetal objects Contrary to popular assumptions, metal objects can be safely used in a microwave oven, but with some restrictions. Any metal or conductive object placed into the microwave oven acts as an antenna to some degree, resulting in an electric current. This causes the object to act as a heating element. This effect varies with the object's shape and composition, and is sometimes utilized for cooking.\nAny object containing pointed metal can create an electric arc (sparks) when microwaved. This includes cutlery, crumpled aluminium foil (though some foil used in microwave ovens is safe, see below), twist-ties containing metal wire, the metal wire carry-handles in oyster pails, or almost any metal formed into a poorly conductive foil or thin wire, or into a pointed shape. Forks are a good example: the tines of the fork respond to the electric field by producing high concentrations of electric charge at the tips. This has the effect of exceeding the dielectric breakdown of air, about 3 megavolts per meter (3×106 V/m). The air forms a conductive plasma, which is visible as a spark. The plasma and the tines may then form a conductive loop, which may be a more effective antenna, resulting in a longer lived spark. When dielectric breakdown occurs in air, some ozone and nitrogen oxides are formed, both of which are unhealthy in large quantities."]], "1_corpus": "wikipedia"}
{"tstamp": 1722310932.6557, "task_type": "retrieval", "type": "leftvote", "models": ["", ""], "ip": "", "0_conv_id": "cbdc637378324499be0368c4cf0b44de", "0_model_name": "embed-english-v3.0", "0_prompt": "when was the abacus invented in ancient china", "0_output": [["when was the abacus invented in ancient china", "Abacus\n\nEgypt Greek historian Herodotus mentioned the abacus in Ancient Egypt. He wrote that the Egyptians manipulated the pebbles from right to left, opposite in direction to the Greek left-to-right method. Archaeologists have found ancient disks of various sizes that are thought to have been used as counters. However, wall depictions of this instrument are yet to be discovered.\nPersia\nAt around 600 BC, Persians first began to use the abacus, during the Achaemenid Empire. Under the Parthian, Sassanian, and Iranian empires, scholars concentrated on exchanging knowledge and inventions with the countries around them – India, China, and the Roman Empire- which is how the abacus may have been exported to other countries.\nGreece\nThe earliest archaeological evidence for the use of the Greek abacus dates to the 5th century BC. Demosthenes (384 BC–322 BC) complained that the need to use pebbles for calculations was too difficult. A play by Alexis from the 4th century BC mentions an abacus and pebbles for accounting, and both Diogenes and Polybius use the abacus as a metaphor for human behavior, stating \"that men that sometimes stood for more and sometimes for less\" like the pebbles on an abacus. The Greek abacus was a table of wood or marble, pre-set with small counters in wood or metal for mathematical calculations. This Greek abacus was used in Achaemenid Persia, the Etruscan civilization, Ancient Rome, and the Western Christian world until the French Revolution."]], "0_corpus": "wikipedia", "1_conv_id": "f78b452ffa04499e97fec3be70a373c3", "1_model_name": "jinaai/jina-embeddings-v2-base-en", "1_prompt": "when was the abacus invented in ancient china", "1_output": [["when was the abacus invented in ancient china", "Abacus\n\nEach rod typically represents one digit of a multi-digit number laid out using a positional numeral system such as base ten (though some cultures used different numerical bases). Roman and East Asian abacuses use a system resembling bi-quinary coded decimal, with a top deck (containing one or two beads) representing fives and a bottom deck (containing four or five beads) representing ones. Natural numbers are normally used, but some allow simple fractional components (e.g. , , and  in Roman abacus), and a decimal point can be imagined for fixed-point arithmetic.\nAny particular abacus design supports multiple methods to perform calculations, including addition, subtraction, multiplication, division, and square and cube roots. The beads are first arranged to represent a number, then are manipulated to perform a mathematical operation with another number, and their final position can be read as the result (or can be used as the starting number for subsequent operations).\nIn the ancient world, abacuses were a practical calculating tool. Although calculators and computers are commonly used today instead of abacuses, abacuses remain in everyday use in some countries. The abacus has an advantage of not requiring a writing implement and paper (needed for algorism) or an electric power source. Merchants, traders, and clerks in some parts of Eastern Europe, Russia, China, and Africa use abacuses. The abacus remains in common use as a scoring system in non-electronic table games. Others may use an abacus due to visual impairment that prevents the use of a calculator. The abacus is still used to teach the fundamentals of mathematics to children in most countries."]], "1_corpus": "wikipedia"}
{"tstamp": 1722310979.3311, "task_type": "retrieval", "type": "tievote", "models": ["", ""], "ip": "", "0_conv_id": "b6d7ab0bc1bb42cb8e93438eef835621", "0_model_name": "GritLM/GritLM-7B", "0_prompt": "where does the spinal cord become cauda equina", "0_output": [["where does the spinal cord become cauda equina", "Cauda equina\n\nDevelopment In humans, the spinal cord stops growing in infancy.  At birth the end of the spinal cord is about the level of the third lumbar vertebra, or L3. Because the bones of the vertebral column continue to grow, by about 12 months of age the end of the cord reaches its permanent position. Typically this is at the level of L1 or L2 (closer to the head), ranging due to normal anatomical variations anywhere from the twelfth thoracic vertebra (T12) to L3. Individual spinal nerve roots arise from the cord as they get closer to the head, but as the differential growth occurs, the top end of the nerve stays attached to the spinal cord while the lower end of the nerve exits the spinal column at its proper level. This results in a \"bundle\"-like structure of nerve fibers that extends caudally, within the spinal column, from the end of the spinal cord, gradually declining in number further down as individual pairs leave the spinal column.\nClinical significance\nNeedle insertion\nThe cauda equina exists within the lumbar cistern, a gap between the arachnoid membrane and the pia mater of the spinal cord, called the subarachnoid space. Cerebrospinal fluid also exists within this space. Because the spinal cord terminates at level L1/L2, lumbar puncture (or colloquially, \"spinal tap\") is performed from the lumbar cistern between two vertebrae at level L3/L4, or L4/L5, where there is no risk of accidental injury to the spinal cord, when a sample of CSF is needed for clinical purposes."]], "0_corpus": "wikipedia", "1_conv_id": "8dd7868f665946e3bc2d5ff340ff7f95", "1_model_name": "text-embedding-004", "1_prompt": "where does the spinal cord become cauda equina", "1_output": [["where does the spinal cord become cauda equina", "Cauda equina\n\nDevelopment In humans, the spinal cord stops growing in infancy.  At birth the end of the spinal cord is about the level of the third lumbar vertebra, or L3. Because the bones of the vertebral column continue to grow, by about 12 months of age the end of the cord reaches its permanent position. Typically this is at the level of L1 or L2 (closer to the head), ranging due to normal anatomical variations anywhere from the twelfth thoracic vertebra (T12) to L3. Individual spinal nerve roots arise from the cord as they get closer to the head, but as the differential growth occurs, the top end of the nerve stays attached to the spinal cord while the lower end of the nerve exits the spinal column at its proper level. This results in a \"bundle\"-like structure of nerve fibers that extends caudally, within the spinal column, from the end of the spinal cord, gradually declining in number further down as individual pairs leave the spinal column.\nClinical significance\nNeedle insertion\nThe cauda equina exists within the lumbar cistern, a gap between the arachnoid membrane and the pia mater of the spinal cord, called the subarachnoid space. Cerebrospinal fluid also exists within this space. Because the spinal cord terminates at level L1/L2, lumbar puncture (or colloquially, \"spinal tap\") is performed from the lumbar cistern between two vertebrae at level L3/L4, or L4/L5, where there is no risk of accidental injury to the spinal cord, when a sample of CSF is needed for clinical purposes."]], "1_corpus": "wikipedia"}
{"tstamp": 1722311003.1834, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "14396d4a28ae4f839eb1d20af5252e72", "0_model_name": "text-embedding-3-large", "0_prompt": "Transformer-based model designed specifically for Arabic language understanding", "0_output": [["Transformer-based model designed specifically for Arabic language understanding", "Title: Interpreting Arabic Transformer Models\n\nAbstract: Arabic is a Semitic language which is widely spoken with many dialects. Given the success of pre-trained language models, many transformer models trained on Arabic and its dialects have surfaced. While these models have been compared with respect to downstream NLP tasks, no evaluation has been carried out to directly compare the internal representations. We probe how linguistic information is encoded in Arabic pretrained models, trained on different varieties of Arabic language. We perform a layer and neuron analysis on the models using three intrinsic tasks: two morphological tagging tasks based on MSA (modern standard Arabic) and dialectal POS-tagging and a dialectal identification task. Our analysis enlightens interesting findings such as: i) word morphology is learned at the lower and middle layers ii) dialectal identification necessitate more knowledge and hence preserved even in the final layers, iii) despite a large overlap in their vocabulary, the MSA-based models fail to capture the nuances of Arabic dialects, iv) we found that neurons in embedding layers are polysemous in nature, while the neurons in middle layers are exclusive to specific properties."]], "0_corpus": "arxiv", "1_conv_id": "f9f953034cbb4a10a1b425e550d64406", "1_model_name": "voyage-multilingual-2", "1_prompt": "Transformer-based model designed specifically for Arabic language understanding", "1_output": [["Transformer-based model designed specifically for Arabic language understanding", "Title: AraBERT: Transformer-based Model for Arabic Language Understanding\n\nAbstract: The Arabic language is a morphologically rich language with relatively few resources and a less explored syntax compared to English. Given these limitations, Arabic Natural Language Processing (NLP) tasks like Sentiment Analysis (SA), Named Entity Recognition (NER), and Question Answering (QA), have proven to be very challenging to tackle. Recently, with the surge of transformers based models, language-specific BERT based models have proven to be very efficient at language understanding, provided they are pre-trained on a very large corpus. Such models were able to set new standards and achieve state-of-the-art results for most NLP tasks. In this paper, we pre-trained BERT specifically for the Arabic language in the pursuit of achieving the same success that BERT did for the English language. The performance of AraBERT is compared to multilingual BERT from Google and other state-of-the-art approaches. The results showed that the newly developed AraBERT achieved state-of-the-art performance on most tested Arabic NLP tasks. The pretrained araBERT models are publicly available on https://github.com/aub-mind/arabert hoping to encourage research and applications for Arabic NLP."]], "1_corpus": "arxiv"}
{"tstamp": 1722311101.6953, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "42b61a56dbda4f8fb93f621e039f5ce4", "0_model_name": "intfloat/multilingual-e5-large-instruct", "0_prompt": "do switches use ip or mac?", "0_output": [["do switches use ip or mac?", "Superuser Stackexchange\n\nQ: IP Address vs MAC addresses why do we need ip addresses if mac address is enough to identify the machine???\n\nA: There are many reasons. For example: MAC address is bound to hardware. It means you can't or rather are not supposed to have the same MAC on different hardware. Imagine that google has to change DNS binding of \"google.com\" to \"MAC\" address each time they change some network card on front-end server.\nAnother reason is that MAC addresses are assigned by producers while IP addresses are assigned by network admins. This allows building hierarchy of addresses that is root of packets routing. This is because way from IP A1.A2.A3.A4 to B1.B2.B3.B4 is easily predictable and way from MAC#1 to MAC#2 isn't.\nNext reason is that MAC is Ethernet-related stuff and there are other network technologies that might not use MAC at all.\n"]], "0_corpus": "stackexchange", "1_conv_id": "ae5bb24c1351491fa7a7bb739099f035", "1_model_name": "BM25", "1_prompt": "do switches use ip or mac?", "1_output": [["do switches use ip or mac?", "Serverfault Stackexchange\n\nQ: Deny Static IP Assignments I am currently using a SonicWall Pro 2040 with DHCP enabled, but only with static IPs mapped to specific MAC addresses. There is no dynamic IP scope being used. Currently, no DHCP request will resolve an IP, as desired.\nHowever, if a server or workstation network interface is still configured with a user-defined, static IP address, it works fine. I would like potential users who assign their machines with static IPs to not work.\nIs there a way to deny these types of connections whose MAC address is not approved? \n\nA: There is no way to configure a DHCP server to deny static IP addresses. If you think about it, there is a direct path between hosts on the network that simply does not go through the SonicWall:\n            SonicWall\n            LAN Port\n                ^\n                |\n                v\nHostA <----> Switch <----> HostB\n\nSo if you want to filter on MAC address to stop HostA from advertising an IP address on the network, you need to do so at the switch. As an example, if your switch is a Cisco, the command to use is switchport port-security.\n"]], "1_corpus": "stackexchange"}
{"tstamp": 1722311103.6017, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "dd1cd051afc647eabf636fc6055dea05", "0_model_name": "intfloat/multilingual-e5-large-instruct", "0_prompt": "How can retrieval-based methods be optimized to balance the trade-off between retrieval precision and computational efficiency in large-scale neural information retrieval systems?", "0_output": [["How can retrieval-based methods be optimized to balance the trade-off between retrieval precision and computational efficiency in large-scale neural information retrieval systems?", "Evaluation measures (information retrieval)\n\nSome authors choose to interpolate the  function to reduce the impact of \"wiggles\" in the curve. For example, the PASCAL Visual Object Classes challenge (a benchmark for computer vision object detection) until 2010 computed the average precision by averaging the precision over a set of evenly spaced recall levels {0, 0.1, 0.2, ... 1.0}:\nwhere  is an interpolated precision that takes the maximum precision over all recalls greater than :\n.\nAn alternative is to derive an analytical  function by assuming a particular parametric distribution for the underlying decision values. For example, a binormal precision-recall curve can be obtained by assuming decision values in both classes to follow a Gaussian distribution.\nThe minimum achievable AveP for a given classification task is given by:\nPrecision at k\nFor modern (web-scale) information retrieval, recall is no longer a meaningful metric, as many queries have thousands of relevant documents, and few users will be interested in reading all of them. Precision at k documents (P@k) is still a useful metric (e.g., P@10 or \"Precision at 10\" corresponds to the number of relevant results among the top 10 retrieved documents), but fails to take into account the positions of the relevant documents among the top k.  Another shortcoming is that on a query with fewer relevant results than k, even a perfect system will have a score less than 1.  It is easier to score manually since only the top k results need to be examined to determine if they are relevant or not."]], "0_corpus": "wikipedia", "1_conv_id": "181c322773ac4bf5a3d0c6321237b30a", "1_model_name": "Salesforce/SFR-Embedding-2_R", "1_prompt": "How can retrieval-based methods be optimized to balance the trade-off between retrieval precision and computational efficiency in large-scale neural information retrieval systems?", "1_output": [["How can retrieval-based methods be optimized to balance the trade-off between retrieval precision and computational efficiency in large-scale neural information retrieval systems?", "Information retrieval\n\nInformation retrieval (IR) in computing and information science is the task of identifying and retrieving information system resources that are relevant to an information need.  The information need can be specified in the form of a search query. In the case of document retrieval, queries can be based on full-text or other content-based indexing. Information retrieval is the science of searching for information in a document, searching for documents themselves, and also searching for the metadata that describes data, and for databases of texts, images or sounds.\nAutomated information retrieval systems are used to reduce what has been called information overload. An IR system is a software system that provides access to books, journals and other documents; it also stores and manages those documents. Web search engines are the most visible IR applications."]], "1_corpus": "wikipedia"}
{"tstamp": 1722311155.3291, "task_type": "retrieval", "type": "tievote", "models": ["", ""], "ip": "", "0_conv_id": "af5d587baf874dcaa68ca13be655b9ec", "0_model_name": "mixedbread-ai/mxbai-embed-large-v1", "0_prompt": "Impact of AI chatbots on academic integrity and plagiarism", "0_output": [["Impact of AI chatbots on academic integrity and plagiarism", "Title: Will ChatGPT get you caught? Rethinking of Plagiarism Detection\n\nAbstract: The rise of Artificial Intelligence (AI) technology and its impact on education has been a topic of growing concern in recent years. The new generation AI systems such as chatbots have become more accessible on the Internet and stronger in terms of capabilities. The use of chatbots, particularly ChatGPT, for generating academic essays at schools and colleges has sparked fears among scholars. This study aims to explore the originality of contents produced by one of the most popular AI chatbots, ChatGPT. To this end, two popular plagiarism detection tools were used to evaluate the originality of 50 essays generated by ChatGPT on various topics. Our results manifest that ChatGPT has a great potential to generate sophisticated text outputs without being well caught by the plagiarism check software. In other words, ChatGPT can create content on many topics with high originality as if they were written by someone. These findings align with the recent concerns about students using chatbots for an easy shortcut to success with minimal or no effort. Moreover, ChatGPT was asked to verify if the essays were generated by itself, as an additional measure of plagiarism check, and it showed superior performance compared to the traditional plagiarism-detection tools. The paper discusses the need for institutions to consider appropriate measures to mitigate potential plagiarism issues and advise on the ongoing debate surrounding the impact of AI technology on education. Further implications are discussed in the paper."]], "0_corpus": "arxiv", "1_conv_id": "6912b24d5d0444babee7fd61715355bc", "1_model_name": "text-embedding-004", "1_prompt": "Impact of AI chatbots on academic integrity and plagiarism", "1_output": [["Impact of AI chatbots on academic integrity and plagiarism", "Title: Will ChatGPT get you caught? Rethinking of Plagiarism Detection\n\nAbstract: The rise of Artificial Intelligence (AI) technology and its impact on education has been a topic of growing concern in recent years. The new generation AI systems such as chatbots have become more accessible on the Internet and stronger in terms of capabilities. The use of chatbots, particularly ChatGPT, for generating academic essays at schools and colleges has sparked fears among scholars. This study aims to explore the originality of contents produced by one of the most popular AI chatbots, ChatGPT. To this end, two popular plagiarism detection tools were used to evaluate the originality of 50 essays generated by ChatGPT on various topics. Our results manifest that ChatGPT has a great potential to generate sophisticated text outputs without being well caught by the plagiarism check software. In other words, ChatGPT can create content on many topics with high originality as if they were written by someone. These findings align with the recent concerns about students using chatbots for an easy shortcut to success with minimal or no effort. Moreover, ChatGPT was asked to verify if the essays were generated by itself, as an additional measure of plagiarism check, and it showed superior performance compared to the traditional plagiarism-detection tools. The paper discusses the need for institutions to consider appropriate measures to mitigate potential plagiarism issues and advise on the ongoing debate surrounding the impact of AI technology on education. Further implications are discussed in the paper."]], "1_corpus": "arxiv"}
{"tstamp": 1722311155.3294, "task_type": "retrieval", "type": "leftvote", "models": ["", ""], "ip": "", "0_conv_id": "c96bbb99cf3546ba97838b73048bd5fa", "0_model_name": "BM25", "0_prompt": "What are the main causes and effects of the Industrial Revolution?", "0_output": [["What are the main causes and effects of the Industrial Revolution?", "Industrial Revolution\n\nPrimitivism Primitivism argues that the Industrial Revolution have created an un-natural frame of society and the world in which humans need to adapt to an un-natural urban landscape in which humans are perpetual cogs without personal autonomy.\nCertain primitivists argue for a return to pre-industrial society, while others argue that technology such as modern medicine, and agriculture are all positive for humanity assuming they are controlled by and serve humanity and have no effect on the natural environment.\nPollution and ecological collapse\nThe Industrial Revolution has been criticised for leading to immense ecological and habitat destruction. It has led to immense decrease in the biodiversity of life on Earth. The Industrial revolution has been said to be inherently unsustainable and will lead to eventual collapse of society, mass hunger, starvation, and resource scarcity.\nThe Anthropocene\nThe Anthropocene is a proposed epoch or mass extinction coming from humanity (anthropo- is the Greek root for humanity). Since the start of the Industrial revolution humanity has permanently changed the Earth, such as immense decrease in biodiversity, and mass extinction caused by the Industrial revolution. The effects include permanent changes to the Earth's atmosphere and soil, forests, the mass destruction of the Industrial revolution has led to catastrophic impacts on the Earth. Most organisms are unable to adapt leading to mass extinction with the remaining undergoing evolutionary rescue, as a result of the Industrial revolution."]], "0_corpus": "wikipedia", "1_conv_id": "babef639f1a24f2a9ac89cb3279d6177", "1_model_name": "Alibaba-NLP/gte-Qwen2-7B-instruct", "1_prompt": "What are the main causes and effects of the Industrial Revolution?", "1_output": [["What are the main causes and effects of the Industrial Revolution?", "De-industrialisation of India\n\nThe fall in the hegemony of Mughals reduced the overall productivity of agriculture and reduced the supply of grains. The grain was the primary consumption good for the Indian workers and was non-tradeable. The reduction in the supply of grain resulted in the rise of its prices. This rise in prices and negative supply shock led to a rise in the nominal wages in the cotton and weaving industry. The increased competition from British cotton and rising nominal wages reduced the profitability of the cotton industry of India. Thus, the negative supply shock in agricultural production is also an important reason behind the de-industrialisation of cotton–industries.\nThe short run as well as long run impact on living standards and growth rate of GDP providing agriculture sector competitive advantage with strengthening of the productivity advance on the land at home or increasing openness to world in turn increases GDP in the short run. The causes of de-industrialisation are region or country specific as in the case of India in the 19th and 20th century. The colonial rule under the British led to the decline of textile and handicrafts industries through their policies and introduction of machine made goods in to the Indian market. Some of the causes of de-industrialisation in India during that period were:\nIntroduction of machine made goods in the Indian subcontinent at a cheaper rate, which led to the decline of the traditional textile industry of India."]], "1_corpus": "wikipedia"}
{"tstamp": 1722311183.6493, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "344ba32a2d4e44d4857f9fcc603f8ef8", "0_model_name": "sentence-transformers/all-MiniLM-L6-v2", "0_prompt": "What are the differences between machine learning and deep learning?", "0_output": [["What are the differences between machine learning and deep learning?", "Comparison of deep learning software\n\nThe following table compares notable software frameworks, libraries and computer programs for deep learning.\nDeep-learning software by name\nComparison of compatibility of machine learning models"]], "0_corpus": "wikipedia", "1_conv_id": "b3eb01f72a77480caae9a4025986769f", "1_model_name": "voyage-multilingual-2", "1_prompt": "What are the differences between machine learning and deep learning?", "1_output": [["What are the differences between machine learning and deep learning?", "Deep learning\n\nOverview Most modern deep learning models are based on multi-layered neural networks such as convolutional neural networks and transformers, although they can also include propositional formulas or latent variables organized layer-wise in deep generative models such as the nodes in deep belief networks and deep Boltzmann machines.\nFundamentally, deep learning refers to a class of machine learning algorithms in which a hierarchy of layers is used to transform input data into a slightly more abstract and composite representation. For example, in an image recognition model, the raw input may be an image (represented as a tensor of pixels). The first representational layer may attempt to identify basic shapes such as lines and circles, the second layer may compose and encode arrangements of edges, the third layer may encode a nose and eyes, and the fourth layer may recognize that the image contains a face.\nImportantly, a deep learning process can learn which features to optimally place in which level on its own. Prior to deep learning, machine learning techniques often involved hand-crafted feature engineering to transform the data into a more suitable representation for a classification algorithm to operate upon. In the deep learning approach, features are not hand-crafted and the model discovers useful feature representations from the data automatically. This does not eliminate the need for hand-tuning; for example, varying numbers of layers and layer sizes can provide different degrees of abstraction."]], "1_corpus": "wikipedia"}
{"tstamp": 1722311199.5255, "task_type": "retrieval", "type": "leftvote", "models": ["", ""], "ip": "", "0_conv_id": "92e783c936084f088d8757c497e00510", "0_model_name": "intfloat/multilingual-e5-large-instruct", "0_prompt": "are electrical outlets on studs?", "0_output": [["are electrical outlets on studs?", "Electronics Stackexchange\n\nQ: Purpose of breadboard \"studs\" on base off micro breadboards? I recently bought some micro breadboards on Ebay.  Extremely useful little things.  However each have two awkward studs on their bases.  See the attached photo.   What I'm curious about is their purpose?  What might these \"plugin\" to or be used for?  I think I'm going to file them off.\n\n\nA: They connect to a base that fixes them with regards to each other.\n\n"]], "0_corpus": "stackexchange", "1_conv_id": "5a38498ae0f7449186feb5beb2f13faf", "1_model_name": "sentence-transformers/all-MiniLM-L6-v2", "1_prompt": "are electrical outlets on studs?", "1_output": [["are electrical outlets on studs?", "Math Stackexchange\n\nQ: Write $\\cos^2(x)$ as linear combination of $x \\mapsto \\sin(x)$ and $x \\mapsto \\cos(x)$ \nCan we write $\\cos^2(x)$ as linear combination of $x \\mapsto \\sin(x)$ and  $x \\mapsto \\cos(x)$?\n\nI know\n$$\n\\cos^2(x)\n= \\frac{\\cos(2x) + 1}{2}\n= 1 - \\sin^2(x)\n= \\cos(2x) + \\sin^2(x)\n$$\nbut none of these helped.\nThen, I tried to solve\n$$\n\\cos^2(x) = \\alpha \\sin(x) + \\beta \\cos(x)\n$$\nfor the coefficients $\\alpha, \\beta \\in \\mathbb{R}$.\nBut when plugging in $x = 0$ I get $\\beta = 1$ and for $x = \\frac{\\pi}{2}$ I get $\\alpha = 0$. Plugging those values back in I obtain a false statement, and WolframAlpha can't do better!\nThis is from a numerical analysis exam and the second function is $x \\mapsto \\sqrt{2}\\cos\\left(\\frac{\\pi}{4} - x \\right)$, which can easily be expressed in terms of $x \\mapsto \\sin(x)$ and $x \\mapsto \\cos(x)$ by the corresponding addition formula.\n\nA: The function $f(x):=\\cos^2 x$ has $f(x+\\pi)\\equiv f(x)$, but any linear combination $g$ of $\\cos$ and $\\sin$ has $g(x+\\pi)\\equiv -g(x)$.\n"]], "1_corpus": "stackexchange"}
{"tstamp": 1722311222.231, "task_type": "retrieval", "type": "tievote", "models": ["", ""], "ip": "", "0_conv_id": "de94bd25010e43bda926db85c7971b20", "0_model_name": "text-embedding-3-large", "0_prompt": "when to use an air gap for dishwasher", "0_output": [["when to use an air gap for dishwasher", "Air gap (networking)\n\nAn air gap, air wall, air gapping or disconnected network is a network security measure employed on one or more computers to ensure that a secure computer network is physically isolated from unsecured networks, such as the public Internet or an unsecured local area network. It means a computer or network has no network interface controllers connected to other networks, with a physical or conceptual air gap, analogous to the air gap used in plumbing to maintain water quality."]], "0_corpus": "wikipedia", "1_conv_id": "436c2a44d65c4bb68f57a571cada2260", "1_model_name": "nomic-ai/nomic-embed-text-v1.5", "1_prompt": "when to use an air gap for dishwasher", "1_output": [["when to use an air gap for dishwasher", "Air gap (networking)\n\nAn air gap, air wall, air gapping or disconnected network is a network security measure employed on one or more computers to ensure that a secure computer network is physically isolated from unsecured networks, such as the public Internet or an unsecured local area network. It means a computer or network has no network interface controllers connected to other networks, with a physical or conceptual air gap, analogous to the air gap used in plumbing to maintain water quality."]], "1_corpus": "wikipedia"}
{"tstamp": 1722311256.1494, "task_type": "retrieval", "type": "leftvote", "models": ["", ""], "ip": "", "0_conv_id": "8685880d067944afa0b8593df3edca7c", "0_model_name": "GritLM/GritLM-7B", "0_prompt": "A paper on scalable and accurate graph transformers using sparse attention and expander graphs", "0_output": [["A paper on scalable and accurate graph transformers using sparse attention and expander graphs", "Title: Exphormer: Sparse Transformers for Graphs\n\nAbstract: Graph transformers have emerged as a promising architecture for a variety of graph learning and representation tasks. Despite their successes, though, it remains challenging to scale graph transformers to large graphs while maintaining accuracy competitive with message-passing networks. In this paper, we introduce Exphormer, a framework for building powerful and scalable graph transformers. Exphormer consists of a sparse attention mechanism based on two mechanisms: virtual global nodes and expander graphs, whose mathematical characteristics, such as spectral expansion, pseduorandomness, and sparsity, yield graph transformers with complexity only linear in the size of the graph, while allowing us to prove desirable theoretical properties of the resulting transformer models. We show that incorporating Exphormer into the recently-proposed GraphGPS framework produces models with competitive empirical results on a wide variety of graph datasets, including state-of-the-art results on three datasets. We also show that Exphormer can scale to datasets on larger graphs than shown in previous graph transformer architectures. Code can be found at \\url{https://github.com/hamed1375/Exphormer}."]], "0_corpus": "arxiv", "1_conv_id": "e515fe1b4bad484eba5701f6c8446cc0", "1_model_name": "intfloat/multilingual-e5-large-instruct", "1_prompt": "A paper on scalable and accurate graph transformers using sparse attention and expander graphs", "1_output": [["A paper on scalable and accurate graph transformers using sparse attention and expander graphs", "Title: Deformable Graph Transformer\n\nAbstract: Transformer-based models have recently shown success in representation learning on graph-structured data beyond natural language processing and computer vision. However, the success is limited to small-scale graphs due to the drawbacks of full dot-product attention on graphs such as the quadratic complexity with respect to the number of nodes and message aggregation from enormous irrelevant nodes. To address these issues, we propose Deformable Graph Transformer (DGT) that performs sparse attention via dynamically sampled relevant nodes for efficiently handling large-scale graphs with a linear complexity in the number of nodes. Specifically, our framework first constructs multiple node sequences with various criteria to consider both structural and semantic proximity. Then, combining with our learnable Katz Positional Encodings, the sparse attention is applied to the node sequences for learning node representations with a significantly reduced computational cost. Extensive experiments demonstrate that our DGT achieves state-of-the-art performance on 7 graph benchmark datasets with 2.5 - 449 times less computational cost compared to transformer-based graph models with full attention."]], "1_corpus": "arxiv"}
{"tstamp": 1722311284.3104, "task_type": "retrieval", "type": "tievote", "models": ["", ""], "ip": "", "0_conv_id": "f4c9ac94173d4247865eb70981697a10", "0_model_name": "mixedbread-ai/mxbai-embed-large-v1", "0_prompt": "What are the key principles of quantum mechanics?", "0_output": [["What are the key principles of quantum mechanics?", "Quantum mechanics\n\nQuantum mechanics is a fundamental theory in physics that describes the behavior of nature at and below the scale of atoms. It is the foundation of all quantum physics, which includes quantum chemistry, quantum field theory, quantum technology, and quantum information science.\nQuantum mechanics can describe many systems that classical physics cannot. Classical physics can describe many aspects of nature at an ordinary (macroscopic and (optical) microscopic) scale, but is not sufficient for describing them at very small submicroscopic (atomic and subatomic) scales. Most theories in classical physics can be derived from quantum mechanics as an approximation valid at large (macroscopic/microscopic) scale.\nQuantum systems have bound states that are quantized to discrete values of energy, momentum, angular momentum, and other quantities, in contrast to classical systems where these quantities can be measured continuously. Measurements of quantum systems show characteristics of both particles and waves (wave–particle duality), and there are limits to how accurately the value of a physical quantity can be predicted prior to its measurement, given a complete set of initial conditions (the uncertainty principle)."]], "0_corpus": "wikipedia", "1_conv_id": "8ed959dba71e4b10b407a68b0b7a8ca9", "1_model_name": "nomic-ai/nomic-embed-text-v1.5", "1_prompt": "What are the key principles of quantum mechanics?", "1_output": [["What are the key principles of quantum mechanics?", "Introduction to quantum mechanics\n\nMany aspects of quantum mechanics are counterintuitive and can seem paradoxical because they describe behavior quite different from that seen at larger scales. In the words of quantum physicist Richard Feynman, quantum mechanics deals with \"nature as She is—absurd\". Features of quantum mechanics often defy simple explanations in everyday language. One example of this is the uncertainty principle: precise measurements of position cannot be combined with precise measurements of velocity. Another example is entanglement: a measurement made on one particle (such as an electron that is measured to have spin 'up') will correlate with a measurement on a second particle (an electron will be found to have spin 'down') if the two particles have a shared history. This will apply even if it is impossible for the result of the first measurement to have been transmitted to the second particle before the second measurement takes place.\nQuantum mechanics helps us understand chemistry, because it explains how atoms interact with each other and form molecules. Many remarkable phenomena can be explained using quantum mechanics, like superfluidity. For example, if liquid helium cooled to a temperature near absolute zero is placed in a container, it spontaneously flows up and over the rim of its container; this is an effect which cannot be explained by classical physics.\nHistory\nJames C. Maxwell's unification of the equations governing electricity, magnetism, and light in the late 19th century led to experiments on the interaction of light and matter. Some of these experiments had aspects which could not be explained until quantum mechanics emerged in the early part of the 20th century.\nEvidence of quanta from the photoelectric effect"]], "1_corpus": "wikipedia"}
{"tstamp": 1722311301.4896, "task_type": "retrieval", "type": "tievote", "models": ["", ""], "ip": "", "0_conv_id": "e25ddbeab37a4e65805ee17fe8dbf844", "0_model_name": "intfloat/multilingual-e5-large-instruct", "0_prompt": "What is the history and significance of the Silk Road?", "0_output": [["What is the history and significance of the Silk Road?", "Silk Road\n\nThe Silk Road was a network of Eurasian trade routes active from the second century BCE until the mid-15th century. Spanning over , it played a central role in facilitating economic, cultural, political, and religious interactions between the Eastern and Western worlds. The name \"Silk Road\" was first coined in the late 19th century, but some 20th- and 21st-century historians instead prefer the term Silk Routes, on the grounds that it more accurately describes the intricate web of land and sea routes connecting Central, East, South, Southeast, and West Asia as well as East Africa and Southern Europe."]], "0_corpus": "wikipedia", "1_conv_id": "f903812140fb434d886b47c056c458bf", "1_model_name": "mixedbread-ai/mxbai-embed-large-v1", "1_prompt": "What is the history and significance of the Silk Road?", "1_output": [["What is the history and significance of the Silk Road?", "Silk Road\n\nThe Silk Road was a network of Eurasian trade routes active from the second century BCE until the mid-15th century. Spanning over , it played a central role in facilitating economic, cultural, political, and religious interactions between the Eastern and Western worlds. The name \"Silk Road\" was first coined in the late 19th century, but some 20th- and 21st-century historians instead prefer the term Silk Routes, on the grounds that it more accurately describes the intricate web of land and sea routes connecting Central, East, South, Southeast, and West Asia as well as East Africa and Southern Europe."]], "1_corpus": "wikipedia"}
{"tstamp": 1722311318.7266, "task_type": "retrieval", "type": "tievote", "models": ["", ""], "ip": "", "0_conv_id": "c1f62d7555804ab3a153de69ca106e31", "0_model_name": "GritLM/GritLM-7B", "0_prompt": "How do black holes form and what are their main properties?", "0_output": [["How do black holes form and what are their main properties?", "Black hole\n\nFormation and evolution Given the bizarre character of black holes, it was long questioned whether such objects could actually exist in nature or whether they were merely pathological solutions to Einstein's equations. Einstein himself wrongly thought black holes would not form, because he held that the angular momentum of collapsing particles would stabilize their motion at some radius. This led the general relativity community to dismiss all results to the contrary for many years. However, a minority of relativists continued to contend that black holes were physical objects, and by the end of the 1960s, they had persuaded the majority of researchers in the field that there is no obstacle to the formation of an event horizon.\nPenrose demonstrated that once an event horizon forms, general relativity without quantum mechanics requires that a singularity will form within. Shortly afterwards, Hawking showed that many cosmological solutions that describe the Big Bang have singularities without scalar fields or other exotic matter. The Kerr solution, the no-hair theorem, and the laws of black hole thermodynamics showed that the physical properties of black holes were simple and comprehensible, making them respectable subjects for research. Conventional black holes are formed by gravitational collapse of heavy objects such as stars, but they can also in theory be formed by other processes."]], "0_corpus": "wikipedia", "1_conv_id": "1c02620544ee43d78b4c741a10b183e6", "1_model_name": "text-embedding-3-large", "1_prompt": "How do black holes form and what are their main properties?", "1_output": [["How do black holes form and what are their main properties?", "Black hole\n\nFormation and evolution Given the bizarre character of black holes, it was long questioned whether such objects could actually exist in nature or whether they were merely pathological solutions to Einstein's equations. Einstein himself wrongly thought black holes would not form, because he held that the angular momentum of collapsing particles would stabilize their motion at some radius. This led the general relativity community to dismiss all results to the contrary for many years. However, a minority of relativists continued to contend that black holes were physical objects, and by the end of the 1960s, they had persuaded the majority of researchers in the field that there is no obstacle to the formation of an event horizon.\nPenrose demonstrated that once an event horizon forms, general relativity without quantum mechanics requires that a singularity will form within. Shortly afterwards, Hawking showed that many cosmological solutions that describe the Big Bang have singularities without scalar fields or other exotic matter. The Kerr solution, the no-hair theorem, and the laws of black hole thermodynamics showed that the physical properties of black holes were simple and comprehensible, making them respectable subjects for research. Conventional black holes are formed by gravitational collapse of heavy objects such as stars, but they can also in theory be formed by other processes."]], "1_corpus": "wikipedia"}
{"tstamp": 1722313460.3193, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "4e00c146250846939c8ba7c8a40eca43", "0_model_name": "BAAI/bge-large-en-v1.5", "0_prompt": "给我一个吻", "0_output": [["给我一个吻", "Taiwanese Hokkien\n\nSubject–kā–object–verb: Another sentence of roughly equivalent meaning is Goá kā lí phō, with the slight connotation of 'I take you and hold' or 'I get to you and hold'.\nObject hō͘ subject–verb (the passive voice): Then, Lí hō͘ goá phō means the same thing but in the passive voice, with the connotation of 'You allow yourself to be held by me' or 'You make yourself available for my holding'.\nWith this, more complicated sentences can be constructed: Goá hō͘ lí chúi lim ('I give water for you to drink': chúi means 'water'; lim is 'to drink').\nThis article can only give a few very simple examples on the syntax, for flavour. Linguistic work on the syntax of Taiwanese is still a (quite nascent) scholarly topic being explored.\nScripts and orthographies\nUntil the late 19th century, Taiwanese speakers wrote mostly in Classical Chinese, although songbooks using Han characters are attested from the 1820s. Among many systems of writing Taiwanese using Latin characters, the most used is called Pe̍h-ōe-jī (POJ) and was developed in the 19th century, while the Taiwanese Romanization System (Tâi-lô) has been officially promoted since 2006 by Taiwan's Ministry of Education. (For additional romanized systems, see references in \"Orthography in Latin characters\", below.) Nonetheless, Taiwanese speakers nowadays most commonly write in Mandarin, though many of the same characters are also used to write Taiwanese."]], "0_corpus": "wikipedia", "1_conv_id": "3072a5571b6f43429edbd73072654e79", "1_model_name": "Salesforce/SFR-Embedding-2_R", "1_prompt": "给我一个吻", "1_output": [["给我一个吻", "Kiss\n\nIn early Japanese films, kissing and sexual expression were controversial. In 1931, a director slipped a kissing scene past the censor (who was a friend), but when the film opened in a downtown Tokyo theater, the screening was stopped and the film confiscated. During the American occupation of Japan, in 1946, an American censor required a film to include a kissing scene. One scholar says that the censor suggested \"we believe that even Japanese do something like kissing when they love each other. Why don't you include that in your films?\" Americans encouraged such scenes to force the Japanese to express publicly actions and feelings that had been considered strictly private. Since Pearl Harbor, Americans had felt that the Japanese were \"sneaky\", claiming that \"if Japanese kissed in private, they should do it in public too.\"\nNon-sexual kisses\nIn some Western cultures, it is considered good luck to kiss someone on Christmas or on New Year's Eve, especially beneath a sprig of mistletoe. Newlyweds usually kiss at the end of a wedding ceremony.\nFemale friends and relations and close acquaintances commonly offer reciprocal kisses on the cheek as a greeting or farewell.\nWhere cheek kissing is used, in some countries a single kiss is the custom, while in others a kiss on each cheek is the norm, or even three or four kisses on alternating cheeks. In the United States, an air kiss is becoming more common. This involves kissing in the air near the cheek, with the cheeks touching or not. After a first date, it is common for the couple to give each other a quick kiss on the cheek (or lips where that is the norm) on parting, to indicate that a good time was had and perhaps to indicate an interest in another meeting."]], "1_corpus": "wikipedia"}
{"tstamp": 1722342604.8788, "task_type": "retrieval", "type": "leftvote", "models": ["", ""], "ip": "", "0_conv_id": "d863e942d2434ae3b85a725bc295dfa8", "0_model_name": "intfloat/multilingual-e5-large-instruct", "0_prompt": "capital or france", "0_output": [["capital or france", "List of capitals of France\n\nThis is a chronological list of capitals of France. The capital of France has been Paris since its liberation in 1944."]], "0_corpus": "wikipedia", "1_conv_id": "04e69520659e4878ab293a12b90a426c", "1_model_name": "text-embedding-3-large", "1_prompt": "capital or france", "1_output": [["capital or france", "List of capitals of France\n\nVersailles (1871), the French Third Republic established Versailles as its provisional seat of government in March 1871 after the Paris Commune took control of Paris.\nBordeaux (September 1914), the government was relocated from Paris to Bordeaux very briefly during World War I, when it was feared that Paris would soon fall into German hands. These fears were alleviated after the German Army was pushed back at the First Battle of the Marne.\nTours (10–13 June 1940), the city served as the temporary capital of France during World War II after the government fled Paris due to the German advance.\nBordeaux (June 1940), the government was relocated from Paris to Tours then Bordeaux very briefly during World War II, when it became apparent that Paris would soon fall into German hands.\nClermont-Ferrand (29 June 1940), the government was relocated from Bordeaux to Clermont-Ferrand, during a single day, before going to Vichy, which had a larger hotel capacity.\nVichy (1940–1944), the Third Republic was abolished in Vichy and replaced it with the French State.\nBrazzaville (1940–1943), with metropolitan France under Axis powers rule, Brazzaville was announced as the seat of the Free France government.\nAlgiers (1943–1944), the city was made the seat of Free France, to be closer to the war in Europe."]], "1_corpus": "wikipedia"}
{"tstamp": 1722342645.1941, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "6600474578594c0d9192933dc077d8f2", "0_model_name": "BM25", "0_prompt": "who created tensorflow", "0_output": [["who created tensorflow", "TensorFlow\n\nMedical GE Healthcare used TensorFlow to increase the speed and accuracy of MRIs in identifying specific body parts. Google used TensorFlow to create DermAssist, a free mobile application that allows users to take pictures of their skin and identify potential health complications. Sinovation Ventures used TensorFlow to identify and classify eye diseases from optical coherence tomography (OCT) scans.\nSocial media\nTwitter implemented TensorFlow to rank tweets by importance for a given user, and changed their platform to show tweets in order of this ranking. Previously, tweets were simply shown in reverse chronological order. The photo sharing app VSCO used TensorFlow to help suggest custom filters for photos.\nSearch Engine\nGoogle officially released RankBrain on October 26, 2015, backed by TensorFlow.\nEducation\nInSpace, a virtual learning platform, used TensorFlow to filter out toxic chat messages in classrooms. Liulishuo, an online English learning platform, utilized TensorFlow to create an adaptive curriculum for each student. TensorFlow was used to accurately assess a student's current abilities, and also helped decide the best future content to show based on those capabilities.\nRetail\nThe e-commerce platform Carousell used TensorFlow to provide personalized recommendations for customers. The cosmetics company ModiFace used TensorFlow to create an augmented reality experience for customers to test various shades of make-up on their face.\nResearch\nTensorFlow is the foundation for the automated image-captioning software DeepDream."]], "0_corpus": "wikipedia", "1_conv_id": "565cdcfdfa6846e2a40cb0dd2b64a586", "1_model_name": "sentence-transformers/all-MiniLM-L6-v2", "1_prompt": "who created tensorflow", "1_output": [["who created tensorflow", "TensorFlow\n\nIn May 2019, Google announced TensorFlow Graphics for deep learning in computer graphics.\nTensor processing unit (TPU)\nIn May 2016, Google announced its Tensor processing unit (TPU), an application-specific integrated circuit (ASIC, a hardware chip) built specifically for machine learning and tailored for TensorFlow. A TPU is a programmable AI accelerator designed to provide high throughput of low-precision arithmetic (e.g., 8-bit), and oriented toward using or running models rather than training them. Google announced they had been running TPUs inside their data centers for more than a year, and had found them to deliver an order of magnitude better-optimized performance per watt for machine learning.\nIn May 2017, Google announced the second-generation, as well as the availability of the TPUs in Google Compute Engine. The second-generation TPUs deliver up to 180 teraflops of performance, and when organized into clusters of 64 TPUs, provide up to 11.5 petaflops.\nIn May 2018, Google announced the third-generation TPUs delivering up to 420 teraflops of performance and 128 GB high bandwidth memory (HBM). Cloud TPU v3 Pods offer 100+ petaflops of performance and 32 TB HBM.\nIn February 2018, Google announced that they were making TPUs available in beta on the Google Cloud Platform."]], "1_corpus": "wikipedia"}