Adrian de Wynter (@deWynterruption) header
Adrian de Wynter (@deWynterruption) profile picture

Scientist at Microsoft/University of York. Opinions are my own.

Both Cambridges, sort of
Joined May 2025
Photo by Adrian de Wynter (@deWynterruption)Photo by Adrian de Wynter (@deWynterruption)Photo by Adrian de Wynter (@deWynterruption)Photo by Adrian de Wynter (@deWynterruption)Photo by Adrian de Wynter (@deWynterruption)Photo by Adrian de Wynter (@deWynterruption)Photo by Adrian de Wynter (@deWynterruption)Photo by Adrian de Wynter (@deWynterruption)Photo by Adrian de Wynter (@deWynterruption)
Since the AoE 2 paper seems to have picked up steam (ayyy), and IT IS 13 dense af PAGES, I would like to clarify a few things: 1. I do not deal with consciousness bc it isn't well-defined nor measurable, 2. The AoE II bit is for substrate dependence, and 3. The core argument/proof of the paper is that claiming existence (or lack thereof) human-like attributes needs better experimental setups. Using AoE II (and later Boston) is to emphasise the distinction between our interpretation of what AI does when we observe it, vs what it does. E.g., an LLM outputting an explanation is a sequence of tokens. Calling it an 'explanation' is our observation of it. Assuming it means something (like self-awareness) is an assumption impacting your experimental setup and thus your conclusions. The paper is dense because one needs to be careful when providing these types of arguments (otherwise, trust me, it'd been a lot shorter). So I wrote a less-formal, more-digestible thing here: adriandewynter.substack.com/…
4
3
1
27
103,966
Adrian de Wynter retweeted
Mathematics via infinite monkeys theorem I mean... a proof at 250,000 pages, what are we even doing at this point
246
382
72
13,464
563,907
Not sharing pic/person, but so lame, TU Darmstadt
23
I have nothing against AI-generated papers provided that they do contribute to scientific knowledge, but oh god the writing. E.g., Claude in an abstract 101: everything's passive voice and sounds dorky af with words like 'instrument' and 'mechanism'
1
2
76
Adrian de Wynter retweeted
"three reasons why AI isn't conscious." really? "consciousness is not intelligence": not a reason why AI isn't conscious, just rebuts a bad argument for AI consciousness (AI: intelligent therefore conscious). "consciousness is substrate-dependent"/"noncomputational": would be a reason if true, but truth needs to be established. more accurate title: "three reasons why AI might not be conscious". teapawt.org/Nature/status/21121504…
219
194
82
1,324
72,767
How shameless can some people be that they post the list of papers they got accepted to #EMNLP2026; and some of these are literal paraphrases of each other? The titles are literally the same, and I checked the content: THE SAME. This field has become such a paper mill. It's depressing.
1
79
Late to the party, but our paper 'Evaluating Style-Personalized Text Generation: Challenges and Directions' has been accepted at #EMNLP2026 main--this paper, led by @JangraAnubhav, critically examines measurement of subjective (style) metrics and shows that (1) we really do not agree on how to measure it; and (2) our metrics are unreliable. See you all in Budapest! arxiv.org/abs/2508.06374
1
2
7
236
Bruh why is #OpenAI saying that 'the LLM broke out of its enclosure' like a velociraptor. Blaming the LLM instead of whoever gave it access to HF...? This is anthropomorphism 101
2
60
Adrian de Wynter retweeted
For a variety of reasons I find it genuinely harrowing that a big portion of our jobs have become responding to whatever new thingTM the marketing teams of big labs have decided to come up with on any given day, both for folks inside these labs and outside. I imagine at the core a lot of us are scientists who want to innovate and go do cool things unbothered instead of this being day 24343 of dropping whatever interesting thing I was upto and reading the news/X/marketing copies. Imo, a lot of these complaints would be ameliorated with frontier American open source models (and AI sovereignty built on OSS in general), which is what we have been screeching from the rooftops for as long as I can remember. Used to be the norm pre-2022, and this current mess (unironically trying to further concentrate power via anticompetitive measures) is entirely a choice. AI should be public infrastructure -- if you are learning from all of human knowledge they should also have corresponding agency in the outcome, and open sourcing is one piece of the puzzle to get there. We should have public options for the entire AI stack (see: publicai.network/publication…).
1
3
29
917
Adrian de Wynter retweeted
hello there the jacobian conjecture is false thanx to my close friend akhil for asking about it and my other close friend fable for working during the world cup final ((1+xy)^3 z + y^2 (1+xy) (4+3xy), y + 3 x (1+xy)^2 z + 3 x y^2 (4+3xy), 2 x - 3 x^2 y - x^3 z): \C^3\to \C^3, has jacobian determinant -2, and sends (0, 0, -1/4), (1, -3/2, 13/2), and (-1, 3/2, 13/2) to (-1/4, 0, 0)
1,698
5,362
3,157
43,828
40,132,414
What drives me up the wall on using AI for maths is that they are SO missing the point--maths is not about marketing. Maths is about the pursuit of truth, of beauty, and the journey that takes you there. Doing these proofs feels a lot like turning mathematical inquiry into these tacky stable diffusion images that plague the internet nowadays. Thanks for coming to my TED talk.
3
6
10,568
I will be at #ICML2026 for the next two days! Come say hi! (Yes, I am severely jet lagged post ACL. Yes, I am late. Yes, I am still willing to talk your ear off about machine cognition)
92
The burnt bits on my toast this morning looked a lot like Jesus. I saw a strikingly similar pattern in Mel Gibson's face on that one film teapawt.org/AnthropicAI/status/207…
New Anthropic research: A global workspace in language models. Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with. We found a strikingly similar divide inside Claude.
2
119
Adrian de Wynter retweeted
One of the only times I remind people I have a PhD in computational neuroscience is when people without a neuroscience background say their model works "like the brain." In these cases, I put on my neuroscience hat, put on my PhD cloak, and say in my important voice: "No, your model is not behaving like the brain."
New Anthropic research: A global workspace in language models. Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with. We found a strikingly similar divide inside Claude.
230
126
22
1,880
266,158
Come to Jiseon Kim and Xiangjue Dong's BoF session: social agents in the wild: reasoning, simulation and evaluation. I will be talking a bit about my work on machine cognition so you can leave before that so you don't fall asleep! 📅 𝘛𝘰𝘥𝘢𝘺, 𝘑𝘶𝘭𝘺 5 | 10:30 𝘈𝘔 𝘵𝘰 12:30 𝘗𝘔 | 𝘎𝘢𝘴𝘭𝘢𝘮𝘱 𝘊-𝘋 #ACL2026 #NLP #LLM #SocialAgents #MultiAgentSystems #AIResearch
69
Gotta love #ACL2026 's food options. Everything has nuts. Me: is there anything to eat that won't kill me? Catering: Me: Catering: ... the dried fruit toppings don't have nuts :D m.youtube.com/watch?v=6fv_Zf…
130
Do LLMs understand what you're talking about? We ran a Wittgenstein- and a Turing-inspired test to answer this, so come to my presentation at #ACL2026 to find out. Pictured: me happy, explaining why juicy salif is iconic, 2hr before I remember why penalties suck😭😭😭 Not pictured: me packing
5
130
Just to reiterate: I do not have any affiliation with the memecoin used for the goat paper. It's funny, but it could also be a scam. Keep an eye out. I can't believe I had to write these words.
Replying to @deWynterruption
Y'all this is very funny but Google says I should put a disclaimer: I don't have any relation with this project--legit don't even know how crypto works 🥲
1
148
My dream: people discussing papers who actually read the paper 🥲
1
4
368
Adrian de Wynter retweeted
Excited to be back at ACM @FAccTConference after 2 years! I will be there in person in Montreal this week, presenting two things: - Paper: "What if AI Systems Weren't Chatbots?" (arxiv.org/abs/2605.07896), in which we discuss problems with the current Conversational AI paradigm, and call for alternative futures for AI that serve the unmet needs of people (physical labor, non-textual scientific research, global south languages) who are currently not reaping benefits from the current direction of AI development. - Tutorial: Every Eval Ever, But For Everyone? Open Challenges in Building Community-Governed AI Evaluation Infrastructure (evalevalai.com/events/2026-f…), in which we walk participants through all the progress we have made at Evaluating Evaluations, introduce various stakeholders in policy, research, governance, procurement, and others to our community tools, Every Eval Ever and Eval Cards. Also, see you at the EvalEval social after! 🥂 (programs.sigchi.org/facct/20…) If you want to chat about Evals, AI for Science, or AI for the Global South, please say hi!
2
6
25
1,688