@aarontay Yes incorporating mentions is kind of a hit-or-miss, it can be crucial for some papers (e.g. when necessary info wasn't retrieved from the index), but it tends to be confusing for models to differentiate consistently
English
Dan Bareket
310 posts
Ai2 paper finder is free and great but I often see it rank something as "perfectly relevant" just because a paper mentions another paper that is perfectly relevant but the paper itself is only sort of relevant paperfinder.allen.ai/share/3bc169d4…
Aaron Tay@aarontay
[Blogged] Ai2 Paper Finder and Futurehouse PaperQA2: More transparent Deep Search for Scholars? musingsaboutlibrarianship.blogspot.com/2025/05/ai2-pa…
English
@aarontay @pastaraspberry @menhguin @PradyuPrasad Appreciate the feedback Aaron (I'm from the Ai2 paperfinder crew). Btw we've been slowly fixing small issues based on the usage we're seeing. Current version should have a more clear and expected behaviour on some things you pointed out re the initial release. Still WIP ofc
English
@pastaraspberry @menhguin @PradyuPrasad I find paid version of Undermind.ai slightly better than paperfinder and same for paid Elicit.
But paperfinder seems best of the academic focused free/open by far. It's still beta so can be unstable and fail when I tried but very promising
English
btw Allen AI's Ai2 PaperFinder is quite good! I haven't used deep research so I can't compare. But even then, it has found a lot of papers I was looking for
English
Give it a shot. If it didn't find what you wanted in the first go, you can ask it to work harder, or you can try to iterate with follow up queries, corrections and refinements.
Ai2@allen_ai
Meet Ai2 Paper Finder, an LLM-powered literature search system. Searching for relevant work is a multi-step process that requires iteration. Paper Finder mimics this workflow — and helps researchers find more papers than ever 🔍
English
#פידתחבורה אם אני צריך להגיע מפלורנטין לשיבא, יש דרך יותר טובה מקו 104?
עברית
At the ACL business meeting. I am *not* bringing up the arxiv policy issue. The reason is that I was at the ACL Exec meeting a few days ago on this. The outcome is that the exec recognizes the issue, but do not want to change policy on their own. A working group will be formed.
English
@dbreqt @ArijRiabi Oh, I didn’t realize it was you ! You and @rtsarfaty really paved the with NER evaluation on Hebrew with predicted morphology. 1st work ever I’ve seen with NER with non gold tokens! Kudos :)
English
English
Dan Bareket retweetledi
ICAIL 2023 has begun! 😀 tomorrow we open the lecture day with a joint law and NLP project with Eliya Habba @dbreqt @GabiStanovsky.
English
Dan Bareket retweetledi
Check out our new paper on mapping abstract descriptions to sentences! We prompt GPT3 to create a dataset of <sentence, description>, and use it to train a retrieval model. A joint work with @valentina__py , Amir DN Cohen, @AvshalomM and @yoavgo (1/6)
arxiv.org/abs/2305.12517
(((ل()(ل() 'yoav))))👾@yoavgo
"semantic embeddings" are becoming increasingly popular, but "semantics" is really ill-defined. sometimes you want to search for text given a description of its content. current embedders suck at this. in this work we introduce a new embedder. @ravfogel @valentina__py @AvshalomM
English
כמה גשר יהודית יכול להחזיק?
איתי בלומנטל 🇮🇱 Itay Blumental@ItayBlumental
לא היה כדבר הזה. גשר יהודית ת"א, עכשיו. אלפים ברחובות
עברית
הייתי קצת מנותק כמה שעות ואני מבין שכולם זועמים והמחאה ניצתה הערב בכח רב. אבל לא הצלחתי להבין למה. מה קרה שהצית את הכל? שמעתי שביבי פיטר את גלנט אבל זו לא יכולה להיות הסיבה? או שזו כן הסיבה? מה הסיבה?
עברית
@yoavgo Well yea, shitty in every possible way, but still comparing the content of academic publishing to that of other media - things can get much much worse real easy
English
hot take: academic publishing, reviewing, citation practices, etc etc are sufficiently broken already, so chatgpt cannot make it much worse
English
@yoavgo but importantly this was written pre-internet and (stupid) bibliometrics, when the intended audience of citations were readers and not calculated scores that disregard the citation context
English
@yoavgo I like the way Latour put it in Science in Action. The general strategy with citation: do whatever you can to render existing literature helpful to your claims; e.g. weaken your enemies, paralyze those you can't weaken, help your allies, support those who supply you resources
English
why are scientific citations important? the common answer is "credit assignment", but let me propose an alternative one: findability and knowledge preservation.
citations help future readers get an access to the literature at a specific point in time.
English
