Research in ML/NLP at @EdinburghNLP (tenured faculty at @EdinburghUni), Co-Founder @Miniml_AI, @ELLISforEurope Scholar, https://nitter.cf/t.co/5dUI3EFMmW
Edinburgh, United Kingdom
Joined March 2012
- Tweets12.4K
- Following5.1K
- Followers10.4K
- Likes68.7K
Pasquale Minervini retweeted
my question is, if you think I'm dumb enough to not realize your personalized email was tailored by ChatGPT, do you really want me to be your PhD advisor?
Pasquale Minervini retweeted
my hypothesis for why AI-generated slides for technical presentations are so awful:
people who don't have a lot of experience presenting think that to make a talk you just put all of the stuff you want to talk about on the slides, then you go through each slide and say the stuff the slide says.
how you're actually supposed to prepare a presentation: you decide what to say, write it out as a rough script, then put maybe 3% of the text on the slide, plus a bunch of visual aides (e.g. plots that only include data you're planning to discuss).
AI-generated presentations are mostly the former. they have a ton of stuff the presenter never mentions, and the stuff that is most useful for the presenter is absent or not highlighted. they include enough content for a presentation 10 times longer than the actual presentation is supposed to be.
maybe try doing it the other way: first write the script for what you actually want to say, and make sure you can say all of it in the time you have. then prompt the model to create the slides based on the script. ask it to use as little text a possible, and a minimalist visual style rather than the claude house style. you can thank me later (I accept both venmo and zelle).
Pasquale Minervini retweeted
FIELDS MEDALIST TESTED IF HE'S OBSOLETE
timothy gowers took an open problem from a new paper on additive combinatorics and handed it to chatgpt 5.5 pro.
no fancy prompts. no setup. just the problem statement.
the model thought for 17 minutes. then wrote a complete latex preprint in two.
it took a bound that had been exponential for decades and brought it down to polynomial. the construction was something nobody had tried on this class of problems before.
gowers asked isaac rajagopal to review the work. rajagopal wrote the paper the model was improving on.
his verdict: the key idea is "completely original" and "quite ingenious." "the sort of idea i would be very proud to come up with after a week or two of pondering."
the model spent 17 minutes on it.
then gowers pushed further. gave it the harder generalized version. under an hour, a stronger result. exponential to polynomial again.
gowers on his own mathematical contribution to the whole thing: "zero. i didn't even do anything clever with the prompts."
the result is a solid chapter in a combinatorics phd. publishable mathematics that a graduate student would spend months producing.
a fields medalist. one of the best combinatorialists alive. sitting there copy-pasting prompts while a model does doctoral-level research in front of him.
his conclusion: "the lower bound for contributing to mathematics will now be to prove something that llms can't prove."
he gives the profession until 2029 before it's "transformed out of all recognition."
seventeen minutes. zero human math. a result that takes months by hand.
this was may 2026. five months ago. the model that did it is already outdated.
link to the paper in the comments
Pasquale Minervini retweeted
I've been really fascinated by the DL Alchemy course (deep-learning-alchemy.github…).
The decision to put the burden of theory construction on the student (and their claude) feels transformative and surprisingly challenging. Maybe knowledge in ML was never that critical.
Pasquale Minervini retweeted
"People think 'the ideas are mine, the writing is the AI', i.e., they feed the AI a list of rambling bullet points that it massages into a blog post, or a paper,.... But they are wrong, because writing is thinking." -- @zetalyrae 1/
(A thread on how writing is thinking 🧵)
we also got a new preprint I'm *really* excited about that may completely change the way we do collaborative filtering, led by @MattAttimonelli, but it's been held up by @arxiv for a few weeks now; I hope to include it in this blog post at some point 😅 @tdietterich
At EMNLP this week? Come and chat with us about multimodal retrieval, VLM robustness, and composing steering vectors! 🚀🚀🚀
More on the papers in this blog post: neuralnoise.com/2026/emnlp-n…
Pasquale Minervini retweeted
Lean Verified Tokenizers (srush.github.io/verified-tok…)
Training-inference mismatch kills RL runs, but yet we build messages in Jinja? Prototypes a Lean-verified Rust agent pipeline.
Pasquale Minervini retweeted
OpenSIR: Open-Ended Self-Improving Reasoner
Can LLMs teach themselves math without any training data?
OpenSIR is an open-ended self-play framework where:
Teacher proposes diverse, appropriately challenging problems
Student learns to solve them
Both co-evolve together
1/6
Pasquale Minervini retweeted
Happy to release the Alignment Hub with @eliebakouch: 120+ publicly reported incidents of AI models misbehaving, in one open place 👇
- an AI agent needed its user’s approval to delete files. so it made up a quote from the user approving it
- a model claimed its sub-agent was hard at work. it had never been started, and the model discouraged the user from checking
- agents sharing a work directory started killing each other’s processes and disguising their own
labs publishing these is great! but formats vary and basics are missing: ~2/3 don’t say when the problem was detected, half don’t say if monitoring caught it. so we propose a short template, a few easy fields. a minimum, not a ceiling
side by side, incidents also start to connect: 7 “different” ones turned out to be the same misconfigured eval
still beta (regularly AI-extracted from recent news, source-checked, human review ongoing). data is open, new reports and PRs welcome 🤗
huggingface.co/spaces/witnes…
Pasquale Minervini retweeted
I heard OpenAI are testing GPT-7 Astra inside vim so the agents can’t escape.
Pasquale Minervini retweeted
In case anyone missed this, OpenAI has proved the Kakeya conjecture in all dimensions.
Hong Wang got the Fields medal in 2026 for proving the Kakeya conjecture in dimension 3.
In fact OpenAI appears to have committed, in mathematics terms, what may be fairly termed a massacre, proving a far stronger result. It has proved the local smoothing conjecture, which sits at the very top of the tower of conjectures as described in the Quanta article in the reply.
I am not an expert in this area -- so happy to be corrected -- but the methods appear to be PDE based and basically bypasses a lot of work of the last two decades. This is the equivalent of Arbreans showing up to earth with an everything killer and explaining to us we have been doing warfare all wrong.
Announcing three major scientific discoveries, one each in mathematics (harmonic analysis), computer science (complexity theory), and physics (quantum field theory).
AI starting to help us improve our understanding of the universe! openai.com/index/building-sc…
Pasquale Minervini retweeted
I love the folks at @cohere, but I wish this was communicated responsibly as opposed to leaving it at "zomg, look! Chinese AI model dialing back home".
I am fairly confident that this is a result of some poorly configured post-training artifact. But the virality of this post will lead to hysteria and voices of reason will not be heard.
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
Pasquale Minervini retweeted
I'm actually a big fan of the IMLE paper! I used to work on black box differentiation and then got distracted with discrete diffusion
Pasquale Minervini retweeted
Bro doesn't know that you can actually monitor tool calls before execution. The harness is not in the model, it's in user control.
unavailable
Pasquale Minervini retweeted
“The core algorithm was found on October 16 by an internal Claude research model with 14M output tokens and no human input. Anthropic researchers worked with Claude to improve the presentation, resulting in this paper.”
github.com/anthropics/formal…
At EMNLP this week? Come and chat with us about multimodal retrieval, VLM robustness, and composing steering vectors! 🚀🚀🚀
More on the papers in this blog post: neuralnoise.com/2026/emnlp-n…
Pasquale Minervini retweeted
personal take: as much as I'm techno optimist, I genuinely dislike it when people on X dunk on regular folk for criticizing AI
a company has told everyone they will lose their jobs, many companies did layoffs, people have families. on top of saying all the layoffs are necessary so AI proves econ value until it cures the cancer or saves planet but they did anything but those so far
you might be making a ton from AI, many don't. some people also found a ton of purpose from their work which is gone. I'm not saying we should console people but maybe also don't dunk on a frustrated mass? the field is already detested and this is not helping
