The Who's Who of AI

What the AI community is actually reading and debating — the articles its researchers, builders, ethicists and critics are sharing, and the discussions they're having, ranked by standing inside the field.

What the directory is reading

What they're debating

Casey Fiesler on why LLMs don't really cite their sources
4 members · 26 posts ·6d ago
Casey Fiesler · Luke Stark · Eileen Clancy 🧿 · Prof. Catherine Flick · Thomas Dietterich
Casey Fiesler: Particularly in the context of (appropriate) crackdowns on LLM-generated papers (e.g., on arXiv) and more and more tales of fabricated citations, two reminders: (1) LLMs do not …
Casey Fiesler: A common comment on my videos about LLM "hallucinations" is that this isn't as much of a problem because LLMs "cite their sources" now. This is a category error. Citing a source…
Casey Fiesler: So when you get LLM output with links or citations to scholarly work: those citations aren't there because the model consulted them and they actually represent the source of the…
Casey Fiesler: Therefore, the possible failure mode here isn't just fabricated citations - a model can include a real paper in a real journal by a real author, but be *completely* wrong about …
Open the thread →
8 members · 22 posts ·3d ago
SE Gyges · tweety fish · Singularity's Bounty e/cc · Pwnallthethings · Dustin Moskovitz · rev. howard arson · Doll · David Marx
SE Gyges: sorry third time's the charm. since timnit has premium twitter and is just over the character limit text kept getting folded under "show more" in the screenshot
SE Gyges: yes it is explicitly about this
tweety fish: [ sepulchral, directionless, booming sigh echoes through the ancient and musty halls ]
Singularity's Bounty e/cc: Pope TESCREAL
Open the thread →
Researchers weigh arXiv's new policy penalizing LLM-generated paper slop
6 members · 9 posts ·13d ago
Mark Riedl · Michael Saxon · critical slop studies · Dominik Moritz · The Data Therapist in the Blue Sky · jake · Anna Rogers
Mark Riedl: ArXiV has a new LLM policy (Screenshots with alt text so you don’t have to click through to the other place and see all the stupid responses)
Michael Saxon: there will be much gnashing of teeth when the slop penalty for an undergrad-prompted related work section which was delegated by a phd student delegated by an absentee advisor h…
Michael Saxon: still, it's absolutely for the best! We need to pull as many emergency brakes on the paper production rate as possible
critical slop studies: Oh good there was a paper I was planning to report and this makes it more clear cut
Open the thread →
5 members · 19 posts ·7d ago
Anil Dash · Alex Hanna · Michael Ekstrand · Hypervisible · Amy Hoy
Anil Dash: The AI of drugs. The same tech financiers have been saying for years that they want GLP-1s in the water supply.
Anil Dash: Tressie you can’t take my one idea per year. (It’s fine, much better than when the tech financiers take my one idea per year. 😭)
Anil Dash: I see tons of folks finding utility in LLMs for helping them code. That’s a separate thing from the agenda and negative externalities of the biggest proponents and manufacturers.
Anil Dash: Yyyyep
Open the thread →
6 members · 18 posts ·2d ago
rev. howard arson · SE Gyges · tweety fish · critical slop studies · Colin · Chris Paxton
rev. howard arson: it is kind of sad to see her reduced to this.
rev. howard arson: like the argument here can be reduced to "i am timnit gebru"
SE Gyges: she has me blocked for saying her primary credential is having been fired by google
SE Gyges: charitably? she simultaneously dealt with a genocide back home and getting fired from her job, which i am not sure i would handle better
Open the thread →
7 members · 26 posts ·4d ago
Thomas Dietterich · Eugene Vinitsky · Martin Jaggi · The Data Therapist in the Blue Sky · Marco Z · David Picard · Andrew Gordon Wilson · Suresh Venkatasubramanian · Naomi Saphra
Thomas Dietterich: At @arxiv.bsky.social, we are receiving a new type of paper that I call an "I did this experiment" paper. These papers typically report some experiment with an LLM or LLM "agent…
Thomas Dietterich: For example, they might tweak the method for context compression or the routing network in an MoE model. They are typically written by a single author using an LLM system and te…
Thomas Dietterich: What do ML researchers here think? Should these be considered research contributions and released on arXiv? Or do these authors need to formulate higher level research questions…
Thomas Dietterich: The influx of first-time, single author, AI-assisted work suggests that these new entrants to the field would benefit from some mentoring about what constitutes a research contr…
Open the thread →
Emily Bender on ChatGPT as a product that harvests what you type
2 members · 2 posts ·14d ago
Emily M. Bender · Baldur Bjarnason · The Data Therapist in the Blue Sky
Emily M. Bender: Always worth remembering: ChatGPT isn't a tool, it isn't a companion. It's a product -- and everything you type in that box is data you are sending to OpenAI.
Baldur Bjarnason: Always worth remembering: ChatGPT isn't a tool, it isn't a companion. It's a product -- and everything you type in that box is data you are sending to OpenAI.
The Data Therapist in the Blue Sky : Not quite! That was the good ol’ days! According to the law suit, now everything you type in that box is data you are sending to OpenAI, and much of it (or derivatives) is also …
Open the thread →
4 members · 9 posts ·1d ago
Melanie Mitchell · Suresh Venkatasubramanian · Marc Lanctot · James MacGlashan · Jasmijn Bastings
Melanie Mitchell: “If a machine can defeat the best of our species at chess, ‘a domain that humans have viewed as sacrosanct, something that is quintessentially possessed by human intelligence,’ …
Melanie Mitchell: It turned out that beating the best humans at chess had nothing to do with more general intelligence. Any analogy with what we're seeing now with AI and math?
Suresh Venkatasubramanian: It's hard to say. But the unit distance conjecture has the feel of deep blue. In that the machine was able to go deeper and explore a space of possibilities more than a person c…
Marc Lanctot: I wouldn't say absolutely nothing but I mostly agree. Of course.. this is the Washington Post's take. Pretty sure most AI researchers didn't think a highly specialized minimax w…
Open the thread →
5 members · 16 posts ·4d ago
Cat Hicks · Maria Antoniak · Adam L · Carlos Scheidegger · Eryk Salvaggio · Stephen Turner
Cat Hicks: Same religion is the reason neither of our dads would come to our wedding. So bizarre every time Bsky fawns over the pope. My dad sent me an encyclical letter to explain WHY he …
Maria Antoniak: exactly, it sends me into a spiral every time this happens
Cat Hicks: He does not want people to come to harm, so in a way, repaired a lot. But on this, no. He once said with great pain that he would've loved my wife if the church would let him an…
Adam L: Cynical theory: the pope puts out progressive-sounding statements strategically, trying to win back people who left over child abuse/anti-LGBT/anti-abortion stuff, while explici…
Open the thread →
How arXiv will detect and penalize LLM-slop papers
5 members · 10 posts ·12d ago
David Picard · Thomas Dietterich · Mary Branscombe · Joshua Grochow · Iris van Rooij
David Picard: The classifier that says if there is incontrovertible evidence of LLM slop in a paper, is it open source? I'd like to see how it's made and also I'd like to use it.
Thomas Dietterich: If you haven't read the thing you are citing, you are already in trouble
Mary Branscombe: yeah, like the mimimum issue with including someone else's citation you haven't bothered to go read is plagiarism?
Joshua Grochow: Is there a way to report slop papers? I've definitely seen a few on arXiv (cs.CC) already.
Open the thread →
AI harms: Skynet hypotheticals vs. what powerful people do with AI
2 members · 3 posts ·15d ago
Julian Sanchez · Luis Villa
Julian Sanchez: In my more cynical moods I suspect the Skynet nonsense is to equate discussion of “AI harms” with fanciful hypotheticals rather than more mundane stuff that’s actually happening.
Julian Sanchez: I am not worried software is suddenly going to decide it wants power over humanity. I am worried about what people who already have power over humanity can do with that software.
Luis Villa: That’s sometimes the purpose, and something just that people who should know better are lazy or bored or pressed for time. In the meantime, we have literal (potential) killer ro…
Open the thread →
4 members · 4 posts ·3d ago
Abeba Birhane · Olivia Guest · Andrew D Wilson · Rua M. Williams · Marielza
Abeba Birhane: misogyny, racism, & fascist ideologies ai systems encode and exacerbate that we've been telling you about for yrs never went away. it's just that governments & those in position…
Olivia Guest: misogyny, racism, & fascist ideologies ai systems encode and exacerbate that we've been telling you about for yrs never went away. it's just that governments & those in position…
Andrew D Wilson: I’ve seen this one arrive in discussions at work and it’s said with such confidence and it shuts the discussion down. People are buying the line that it’s assistive tech and tha…
Rua M. Williams: I love/hate that this term is useful.
Open the thread →