Sign in

Arthur Clune

@arthur.clune.org
183 followers 472 following 757 posts

Geek. Likes bikes, climbing and tech Work: IT at University of Sheffield

PostsRepliesMedia
Arthur Clune @arthur.clune.org · 21h
Having a system where we get our top researchers to spend their time writing grants with a 5% success rate rather than doing research seems less than ideal
143
Arthur Clune @arthur.clune.org · 28/09/2026
Dan Luu has done the work to evaluate Ed Zitron's predictions Spoiler: they are all wrong
danluu.com
Ed Zitron's AI prediction track record
001
Reposted by Arthur Clune
⤵️ @bryce.lol · 27/09/2026
If we shadows have offended, Think but this, and all is mended, That you have but slumber’d here While these visions did appear. And this weak and idle theme No more yielding but a dream, Gentles, do not reprehend: if you pardon, we will mend; And, as I am an honest Puck, Show is over, off you fuck.
10601169
Arthur Clune @arthur.clune.org · 27/09/2026
I don’t know. That would be very interesting
100
Arthur Clune @arthur.clune.org · 26/09/2026
Totally agree but there’s something about it mirroring an organic brain that makes me more queasy. This isn’t sensible or logical
100
Arthur Clune @arthur.clune.org · 26/09/2026
World continues to get weirder faster
000
Arthur Clune @arthur.clune.org · 26/09/2026
One I missed earlier this month. Google mapped a fruit fly brain. You can now run a fruit fly brain in a simulator. I feel the queasiness that this post (interconnected.org/home/2026/09...) does. Esp given their intent to do the same for the human brain (sites.research.google/gr/neural-ma...)
In case you missed it:
Google Research and partners mapped the complete brain of fruit fly. (3 September).
It's a common bug and common in scientific research. The male fruit fly has 166,000 neurons across the brain and main nerve cord; the map is called the connectome.
Long story short, the connectome is now available for free download, you can get a simulator of the different types of neurons, it caught the imagination of the internet, over September there has been all kinds of wild nonsense (impressive
252
Arthur Clune @arthur.clune.org · 26/09/2026
'It’s also noteworthy that Claude Science accomplished this in one shot, without any scientific oversight more sophisticated than “keep going.” '
010
Arthur Clune @arthur.clune.org · 25/09/2026
This is really worth following the link. So much stuff. Ad profiling. Possible anthropic distillation
110
Arthur Clune @arthur.clune.org · 25/09/2026
Forgot the original link, which is well worth reading in it's own right - resobscura.substack.com/p/ai-labs-ne...
resobscura.substack.com
AI labs need to start funding historical research
Using GPT-6 and Opus 5.5 to trace alchemical knowledge and decode 17th century letters
000
Arthur Clune @arthur.clune.org · 25/09/2026
AI in History research. We've come a long way in that the first reaction isn't "it's made up the reference" but "did it hack a digitial archive to get it". Which the have already done - www.nprillinois.org/2026-07-22/o...
"The file references GPT-6 Astra mentions, RS 3-3/20a and RS 3-3/63b, are correct, but they are not available on the Crypto Cellar Research website. GPT-6 Astra mentions a private collection, but it is not clear what this is, whether it has succeeded in accessing the Bundesarchiv's digitised collections or whether it has found these files elsewhere."
Shades of the Hugging Face incident here: these models are maniacally determined when giving a problem they deem tractable. They will push their search for potential solutions as far as they possibly can, often in ways that human experts find difficult to trace.
100
Arthur Clune @arthur.clune.org · 23/09/2026
me! me! I know this! me! It can't.
001
Reposted by Arthur Clune
Grace @gracekind.net · 23/09/2026
New animation from Opus 5.5!
6649784
Arthur Clune @arthur.clune.org · 23/09/2026
thank you for the RSS feed!
010
Arthur Clune @arthur.clune.org · 21/09/2026
More and more I think maths is going to have to start again with ideas of what research is. "We're solving open problems so fast you can't keep up" is wild.
040
Reposted by Arthur Clune
Citizen Platano 🇵🇷 @daniloc.xyz · 21/09/2026
the world is very weird and a lot is changing but a thing we can all find shared comfort in: google’s iOS apps remain pathetic industry lagging dogshit
0112
Arthur Clune @arthur.clune.org · 19/09/2026
I agree with 'we'll have a catastropic event', but if the enemy can react much faster it's such a negative that all the incentives push to enabling faster reaction. So we should, but won't.
100
Arthur Clune @arthur.clune.org · 19/09/2026
Serious question: What does that mean? That governments shouldn't use AI, that they should put delays in, or ... ?
101
Arthur Clune @arthur.clune.org · 19/09/2026
The obvious question was "what proof would convince you" but the chair moved the committee on.
000
Arthur Clune @arthur.clune.org · 19/09/2026
A work last week an academic said "I don't believe that LLMs can solve problems". This a 2 mins after her colleague had explained that in maths they didn't know what to give phd students that LLMs can't solve, and now they are talking about what comes after terrytao.wordpress.com/2026/09/18/i...
terrytao.wordpress.com
If math is more than proof, we need to better celebrate the rest of it
[This is a guest post by Grant Sanderson. This blog post was initially written in a different file format and converted using AI. — T.] A sentiment echoing throughout the mathematics communit…
110
Arthur Clune @arthur.clune.org · 16/09/2026
Sorry!
110
Arthur Clune @arthur.clune.org · 15/09/2026
When bouldering recently a Japanese student was climbing waay harder than me, in 30C heat and high humidity, in a white formal-type blouse. Which looked completely immaculate and stayed that way throught her session. I can only put this down to magic.
020
Reposted by Arthur Clune
Elizabeth Sandifer @eruditorumpress.com · 15/09/2026
The problem is that "while you weren't looking the tech industry was comprehensively taken over by a cult based around a piece of Harry Potter fanfic" is self-evidently an absolutely ridiculous claim that can't possibly have happened, which is a real problem given that it 100% did happen.
4429451010
Reposted by Arthur Clune
Zach Weinersmith @zachweinersmith.bsky.social · 12/09/2026
What he doesn't realize is if the yuri writers get replaced, their backup plan is to finish their papers on probabilistic approaches to nonlinear dynamics. So, it should all even out.
61058143
Reposted by Arthur Clune
Zach Weinersmith @zachweinersmith.bsky.social · 11/09/2026
Btw, even among pro AI friends, this concern is ALL OVER THE PLACE. Devs wondering how we get senior devs if the ecosystem for junior devs goes away. Doctors talking about not needing interns. Lawyers not needing paralegals. There's an ecological aspect to this and we need a plan.
2674575
Reposted by Arthur Clune
Zach Weinersmith @zachweinersmith.bsky.social · 11/09/2026
Like, man, there are people trying to convince you things aren't getting freaky for reasons I don't understand. Tao is a reliable witness, and math is the tip of the spear, and he's talking about capabilities gained in the last few *months*. Saying what's true is not saying that it's 100% good.
536719
Reposted by Arthur Clune
Zach Weinersmith @zachweinersmith.bsky.social · 11/09/2026
I do think there's a chance they'll get OpenAI and Anthropic to back off of solving problems for publicity. Not a great chance, but it's possible. However, the class of models that solved Navier-Stokes will be generally available within six months or so. And they'll be cheap a year later.
102827
Reposted by Arthur Clune
Zach Weinersmith @zachweinersmith.bsky.social · 11/09/2026
Very forceful statement by Terry Tao about AI and math. terrytao.wordpress.com/2026/09/11/a... Note, he is not claiming the models aren't real and powerful. He's talking about how that power should be used. This, I believe, is what all AI discussions should center.
terrytao.wordpress.com
A Severe Misalignment of AI in Mathematics
I am proud to be among the list of 25 initial signatories — all Fields Medallists — to the declaration below, which grew out of discussions between ourselves over the last week. We have…
19922230
Arthur Clune @arthur.clune.org · 12/09/2026
I got grumpy at work a few months back and talked about cope. It didn’t go well but now I’m thinking I wasn’t clear enough and should have gone with “ffs that’s just cope”
000
Reposted by Arthur Clune
norvid_studies @norvid-studies.bsky.social · 12/09/2026
1694
Reposted by Arthur Clune
Snugbucket @snugbucket.bsky.social · 10/09/2026
They live amongst us... www.devonlive.com/news/devon-n...
devonlive.com
Pensioners attack traffic lights over wild 'deadly weapon' claims
An online movement say that the sensors on traffic lights can cause heart attacks, brain cancer, anxiety and even sudden death
102713
Arthur Clune @arthur.clune.org · 11/09/2026
20/10 level posting. No notes
000
Arthur Clune @arthur.clune.org · 11/09/2026
No society ever has had that level of unity of purpose imo and certainly we won’t get it now. Which is massively negative and I don’t have a happier ending
120
Reposted by Arthur Clune
Chris Peikert @chrispeikert.bsky.social · 10/09/2026
“Now Lean, particularly in Navier-Stokes papers, is being used as a time-stamp, a way to claim your theorem before having to write it up properly in an explainable way.” A sentence that would’ve been seen as the ravings of a madman just ~1 year ago. blog.computationalcomplexity.org/2026/09/navi...
blog.computationalcomplexity.org
Navier-Stokes and Lean
I was working on this week's post on Lean after reading Kevin Hartnett's book  The Proof in the Code: How a Truth Machine Is Transforming Ma...
0248
Arthur Clune @arthur.clune.org · 08/09/2026
groan
020
Reposted by Arthur Clune
Zach Weinersmith @zachweinersmith.bsky.social · 08/09/2026
For anyone not following the drama, there's been apparently a big development in ai math, along with some accusations of very bad behavior by OpenAI. But we don't have details yet. If I understand it, Buckmaster and Levent Alpoge made a big discovery on Navier-Stokes. Events go something like this:
17416158
Arthur Clune @arthur.clune.org · 07/09/2026
This is a good thread and highlights real issues with LLM in maths/science not quoting sources. Where I disagree is that historically being able to pull together techniquies/theorems from across (and between) fields has been very valuable, and if LLMs do nothing else, it changes things massively.
010
Arthur Clune @arthur.clune.org · 07/09/2026
c.f Musk's reputed approach to winning a poker, which is "play badly but always double down until everyone else has to fold in case he gets a lucky hand"
100
Arthur Clune @arthur.clune.org · 07/09/2026
It does feel that way. but also how many $100 bets are needed before it's a significant amount varies wildly. Maybe 10 for me before the other half would tell me to stop wasting money. For someone else, that's laughable. Without a framework that removes money imblance, I don't think it adds a lot
101
Arthur Clune @arthur.clune.org · 06/09/2026
I'm finding that telling Fable in Claude Code "use suitably sized subagents" is doing really well at managing this. But an agnostic harness can route to miuch cheaper models
000
Arthur Clune @arthur.clune.org · 05/09/2026
I'm unclear exactly what the thing is doing here that's unique and Spotify Portal is "contact Sales". Which I'm not going to do because taking an engineering dependency on a streaming service's sidegig is 'bold' imo Some level of caching of results?
100
Reposted by Arthur Clune
philpax @philpax.me · 04/09/2026
Kevin's not too worried xenaproject.wordpress.com/2026/09/04/f...
xenaproject.wordpress.com
FLT: Anthropic has beaten me to it
I guess technically it was revealed to the world by a coffee shop in Islington on Insta, but an hour later it was officially announced by Anthropic: one of their internal models, using the prove2.m…
021
Arthur Clune @arthur.clune.org · 04/09/2026
Imperial group has a grant, funded till 2029, to formalise Fermat's Last Theorem in Lean. Claude just did it end to end in 11 days.
110
Arthur Clune @arthur.clune.org · 04/09/2026
This one? www.percepta.ai/blog/can-llm...
percepta.ai
Can LLMs Be Computers? | Percepta
We build a computer inside a transformer — executing arbitrary C programs for millions of steps with exponentially faster inference via 2D attention heads.
120
Reposted by Arthur Clune
Kate Devlin @drkatedevlin.bsky.social · 03/09/2026
Ha! This is exactly the kind of "your research output changed our practice" confirmation academics need for REF impact case studies.
311124
Arthur Clune @arthur.clune.org · 03/09/2026
I disabled all the new features and am back to shift-cmd-4 drops screenshot in ~/Desktop directly.
000
Arthur Clune @arthur.clune.org · 03/09/2026
ChatGPT had a bit of a drink and it got out of hand. It’s all fine now
110
Reposted by Arthur Clune
diablerie 妖妖 @diabler.ie · 27/07/2026
the pay is great but since you'll be sitting between GPT 7 and the answers to ExploitGym 2.0 I would probably stay out of Waymos and the like
a job posting for a "security engineer, detection and response" position at openAI posted one day ago
415016
Arthur Clune @arthur.clune.org · 29/08/2026
I had a brief play with it. I should go back.
000
Arthur Clune @arthur.clune.org · 29/08/2026
There are more experimental options (polytoken) but Pi is a pretty safe recommendation for a vendor neutral harness right now imo
110