Geoffrey Hinton's distillation paper might be the most successful workshop (!!) paper of all time.
NIPS Workshop on Deep Learning and Representation Learning.
The workshop topic alone would not make sense anymore now, it is basically "ICLR".
Me agreeing to a new project knowing full well it’s gonna add stress 🇲🇽
New paper: Back into Plato’s Cave
Are vision and language models converging to the same representation of reality? The Platonic Representation Hypothesis says yes. BUT we find the evidence for this is more fragile than it looks.
Project page: akoepke.github.io/cave_umwelten/
1/9
"Thank you for that lecture, it was really interesting and it was clear that you put a lot of effort into preparing it."
:)
This thread is required reading...
I feel like lots of people have been yelling "we need to change the incentive structures of academia!" for many years but nothing ever happens. :-\
But like *what* actually needs to happen to disincentivize whatever has resulted in a 3,000+ paper submission numbers jump in 4 years???
I'm not at CHI but I hear that there's expected "oh god what do we do about reviewing, we have had an exponential explosion in paper submissions" discussion.
I think a good start would be to stop making everyone feel like they have to publish 10 first author CHI papers to get a job. :-\
Man Who Threw Molotov Cocktail At Sam Altman’s Home Claims He Was Following ChatGPT Recipe For Risotto
Man Who Threw Molotov Cocktail At Sam Altman’s Home Claims He Was Following ChatGPT Recipe For Risotto theonion.com/man-who-threw-molotov-co...
1500+ mathematicians are demanding that their field’s most prestigious meeting, the International Congress of Mathematicians, where the prestigious Fields Medal is awarded, be moved from the US because of geopolitical concerns and immigration policy www.scientificamerican.com/article/why-...
There’s some cool recent research on this phenomenon! It turns out vision language models excel at image benchmarks *even when the actual images aren’t provided,* because the answers are implicit in the questions!
arxiv.org/abs/2603.21687
The Position Paper Track is back at NeurIPS 2026 for the second year, with an expanded scope, and better alignment with the main and Evaluation and Dataset tracks!
Isambard Summit 2026 was made a huge success by the speakers, sponsors, and showcased projects. Thank you to everyone who contributed and attended. See you next year 🎩
"NeurIPS' new policy, which was enacted to comply with US laws, means that previous Chinese sponsors of the conference, such as Huawei, are excluded from submitting research papers" www.reuters.com/world/china/...
It's been said plenty of times in plenty of different ways, but it is striking to emphasise thar if you've fit a linear model to data, your eventual plot of the fitted model need not be linear (since you might be linear in features of x rather than in x, etc.).
At some point this was a hill I was prepare to die on, but I gave up: running a Linear layer over all the pixels of your image is a 1x1 convolution.
A ViT is not "convolution-less", since the initial "linear mapping" (sigh) is a 1x1 conv on the patch matrix (channels being all pixels).
Reaching out to potential PhD supervisors purely on the basis of the vibes they give when posting on here
I'm totally biased on this but I think it's wonderful that we now have official NeurIPS parallel satellite events. Parallelization was the trick that allowed us to scale to large data and models. It makes sense to try it for conferences as the community grows.
This is so relevant for huge part of ML. We should all start integrating memory and continual learning in our works yesterday.
To ensure compliance w peer-review policies, ICML has removed 795 reviews (1% of total) by reviewers who used LLMs when they explicitly agreed to not. Consequently, 497 papers (2% of all submissions) of these (reciprocal) reviewers have been desk rejected
Details in blog post 👇
"we are removing the affected reviews and desk rejecting the 497 papers where a violating reviewer was serving as a reciprocal reviewer---approximately 2% of submissions in total." I'm very happy the ICML PC take a hard stance against reviewers not respecting LLM reviewing policy.
📱 New Enterprise Sessions episode!
Dr Myles‑Jay Linton’s commitment to understanding young people’s mental health is driving powerful new research and digital innovation at Bristol 🧠
Listen here: researchpod.org/university-o...
@mylesjaylinton.bsky.social
YouTube thumbnail for the sixth episode of Bristol Data Stories. On the left is a headshot of Jeff Clark. On the right is a headshot of Madlen Williams and Leo Gorman.
Bristol Data Stories Episode 6: Technology for social good: the use of AI in medical imagery with @savvyscientist.bsky.social
In this episode, Jeff discusses how AI can be used in medical imagery and talks about his website, 'The Savvy Scientist'.
Listen now: tinyurl.com/a3xn3y55
I'm organising a seminar at Bristol Digital Futures Institute hosting our wonderful placement PhD researcher, Zia Saylor!
If you're interested in sociotechnical approaches to digital twin research, come join us on the 31st March at 3pm UK time, whether in person or online.
One of the many, many things genAI destroys is human collaboration.
It actively creates loners, guys spinning off into the deep end with literally no one around them to go "hey what the fuck are you talking about" or "hey this sounds like the thing I've been working on" or "check out this book"
Fascinating Bloomberg piece about how AI is used as an excuse for downsizing, but isn't usually the real reason:
> “We’re restructuring around AI” is a growth signal. “We over-hired during the pandemic and revenue softened” is an accountability signal.
/1
Happy Pi Day! In a certain sense, π is not a constant, but a variable. Using our usual Euclidean distance, it is 3.14159… but applying other L^p norms on ℝ², half the unit circle's perimeter will give other values. For instance, at p=1 (taxicab geometry), “π” = 2√2. At p ≈ 2.2, it's 3.20.
"Palantir CEO Alex Karp says his AI technology will lessen the power of “highly educated, often female voters, who vote mostly Democrat” while increasing the power of working-class men."
Just saying it right out loud now.
newrepublic.com/post/207693/...
Photo of a web form saying "Depth of understanding attained: The project blah blah blah. However..."
Making great progress with dissertation marking
When I still worked for a tech company I remember my colleagues responding to things like this with "well they're not proposing solutions". Until I was out of that world, it was hard to see what a failure of imagination it is to see critique and abolition as impractical
