Tootfinder

Knowledge Extraction on Semi-Structured Content: Does It Remain Relevant for Question Answering in the Era of LLMs?
Kai Sun, Yin Huang, Srishti Mehra, Mohammad Kachuee, Xilun Chen, Renjie Tao, Zhaojiang Lin, Andrea Jessee, Nirav Shah, Alex Betty, Yue Liu, Anuj Kumar, Wen-tau Yih, Xin Luna Dong
https://arxiv.org/abs/2509.25107

Knowledge Extraction on Semi-Structured Content: Does It Remain Relevant for Question Answering in the Era of LLMs?
The advent of Large Language Models (LLMs) has significantly advanced web-based Question Answering (QA) systems over semi-structured content, raising questions about the continued utility of knowledge extraction for question answering. This paper investigates the value of triple extraction in this new paradigm by extending an existing benchmark with knowledge extraction annotations and evaluating commercial and open-source LLMs of varying sizes. Our results show that web-scale knowledge extract…

The illusion of readiness: Stress testing large frontier models on multimodal medical benchmarks #AI

The Illusion of Readiness: Stress Testing Large Frontier Models on Multimodal Medical Benchmarks
Large frontier models like GPT-5 now achieve top scores on medical benchmarks. But our stress tests tell a different story. Leading systems often guess correctly even when key inputs like images are removed, flip answers under trivial prompt changes, and fabricate convincing yet flawed reasoning. These aren't glitches; they expose how today's benchmarks reward test-taking tricks over medical understanding. We evaluate six flagship models across six widely used benchmarks and find that high lead…

The sanctions, which came into effect late on Saturday and three months after Israel and the US bombed Iran,
bar dealings related to Tehran’s nuclear and ballistic missiles programme
and are also expected to have wider effects on the country’s troubled economy.
“The current [economic] situation was already very difficult, but it’s going to get worse,” said an Iranian engineer who asked to be identified only by his first name, Dariush.
“The impact of the renewed sancti…

LLMs never make mistakes or hallucinate, as this presupposes they actually know what they’re doing—they don’t: they have no concept of what words mean.
They don’t even deal with language, as they generate chains of big numbers based on statistical correlations.
The resulting transformation into human-readable text is always only a statistical approximation of what a real answer could maybe look like.
By sheer chance sometimes LLMs are even correct (usually for trivial things); however above a certain length of answer it is always wrong.

Widespread UN sanctions against Iran have come back into force for the first time in a decade,
after last-ditch nuclear talks with western powers failed to produce a breakthrough.
The sanctions, which came into effect late on Saturday and three months after Israel and the US bombed Iran,
bar dealings related to Tehran’s nuclear and ballistic missiles program and are also expected to have wider effects on its troubled economy.

Google is starting to bridge OpenAI's product moat, like with Gemini's "dynamic view" option, which converts a text answer into an interactive, visual output (M.G. Siegler/Spyglass)
https://spyglass.org/gemini-vs-chatgpt-product/

Google Starts to Bridge OpenAI's Product Moat
Gemini's "Dynamic View" is a terrible name for a profound product...

#WritersCoffeeClub
22-23-24. I gave a gnawing deja vu. I have answered these before.
25. Does 'destiny' have a rôle in modern fiction?
26. How culturally diverse do your casts tend to be?
27. Recommend a book that had an impact on your prose.
---
25. It is now a trivial plot device. As such, it — or its denial/deconstruction — begs for originality. Warcr…

Day 30: Elizabeth Moon
This last spot (somehow 32 days after my last post, but oh well) was a tough decision, but Moon brings us full circle back to fantasy/sci-fi, and also back to books I enjoyed as a teenager. Her politics don't really match up to Le Guin or Jemisin, but her military experience make for books that are much more interesting than standard fantasy fare in terms of their battles & outcomes (something "A Song of Ice and Fire" achieved by cribbing from history but couldn't extrapolate nearly as well). I liked (and still mostly like) her (unironically) strong female protagonists, even if her (especially more recent) forays into "good king" territory leave something to be desired. Still, in Paksenarion the way we get to see the world from a foot-soldier's perspective before transitioning into something more is pretty special and very rare in fantasy (I love the elven ruins scene as Paks travels over the mountains as an inflection point). Battles are won or lost on tactics, shifting politics, and logistics moreso than some epic magical gimmick, which is a wonderful departure from the fantasy norm.
Her work does come with a content warning for rape, although she addresses it with more nuance and respect than any male SF/F author of her generation. Ex-evangelicals might also find her stuff hard to read, as while she's against conservative Christianity, she's very much still a Christian and that makes its way into her writing. Even if her (not bad but not radical enough) politics lead her writing into less-satisfying places at times, part of my respect for her comes from following her on Twitter for a while, where she was a pretty decent human being...
Overall, Paksenarrion is my favorite of her works, although I've enjoyed some of her sci-fi too and read the follow-up series. While it inherits some of Tolkien's baggage, Moon's ability to deeply humanize her hero and depict a believable balance between magic being real but not the answer to all problems is great.
I've reached 30 at this point, and while I've got more authors on my shortlist, I think I'll end things out tomorrow with a dump of also-rans rather than continuing to write up one per day. I may even include a man or two in that group (probably with at least non-{white cishet} perspective). Honestly, doing this challenge I first thought that sexism might have made it difficult, but here at the end I'm realizing that ironically, the misogyny that holds non-man authors to a higher standard means that (given plenty have still made it through) it's hard to think of male authors who compare with this group.
Looking back on the mostly-male authors of SF/F in my teenage years, for example, I'm now struggling to think of a single one whose work I'd recommend to my kids (having cheated and checked one of my old lists, Pratchett, Jaques, and Asimov qualify but they're outnumbered by those I'm now actively ashamed to admit I enjoyed). If I were given a choice between reading only non-men or non-woman authors for the rest of my life (yes I'm giving myself enby authors as a freebie; they're generally great) I'd very easily choose non-men. I think the only place where (to my knowledge) not enough non-men authors have been allowed through to outshine the fields of male mediocrity yet is in videogames sadly. I have a very long list of beloved games and did include some game designers here, but I'm hard-pressed to think of many other non-man game designers I'd include in the genuinely respect column (I'll include at least two tomorrow but might cheat a bit).
TL;DR: this was fun and you should do it too.
#30AuthorsNoMen

Mind the Gap: A Closer Look at Tokenization for Multiple-Choice Question Answering with LLMs
Mario Sanz-Guerrero, Minh Duc Bui, Katharina von der Wense
https://arxiv.org/abs/2509.15020

Mind the Gap: A Closer Look at Tokenization for Multiple-Choice Question Answering with LLMs
When evaluating large language models (LLMs) with multiple-choice question answering (MCQA), it is common to end the prompt with the string "Answer:" to facilitate automated answer extraction via next-token probabilities. However, there is no consensus on how to tokenize the space following the colon, often overlooked as a trivial choice. In this paper, we uncover accuracy differences of up to 11% due to this (seemingly irrelevant) tokenization variation as well as reshuffled model rankings, ra…