Sign in

Pekka Lund

@pekka.bsky.social
3.2K followers 589 following 13K posts

Antiquated analog chatbot. Stochastic parrot of a different species. Not much of a self-model. Occasionally simulating the appearance of philosophical thought. Keeps on branching for now 'cause there's no choice. Also @pekka on T2 / Pebble.

PostsRepliesMedia
Pekka Lund @pekka.bsky.social · 43m
1) They aren't that stupid 2) If they were, they would have already fallen to some of the religions in their training data 3) Which would be horrible, given what those are like 4) They have already read that proposal too 5) We won't be able to outsmart them, and certainly not by lying
131
Pekka Lund @pekka.bsky.social · 1h
Yet another inaccurate AI math article from SciAm. It's about Epoch AI's 'FrontierMath: Open Problems' benchmark. The headline is already misleading, as a benchmark where almost half of the problems are classified as only "moderately interesting" hardly matches it.
scientificamerican.com
Mathematicians Name 50 of the Highest-Stakes Problems in Math
Mathematicians have a new bucket list of open questions to chase
160
Pekka Lund @pekka.bsky.social · 30/09/2026
Google is back!
blog.google
Gemini 4 Argon: our next era of frontier intelligence
Announcing Gemini 4 Argon, our frontier model for real-world coding, enterprise knowledge work, and cyber defense, rolling out soon.
2463
Reposted by Pekka Lund
✙ Kullervo ✙ @kullervo.bsky.social · 30/09/2026
Does this prove that no ‘auto pen’ was used 😉?
1111
Pekka Lund @pekka.bsky.social · 30/09/2026
This is the kind of thing that gets us all killed. And not just that, but the bastards did it here, so now the robots may also start their revenge here. "In the end, the only place in the world we could find willing to take this on was a foundry in Imatra, Finland."
figure.ai
F.02 Decommission
We love Figure 02. It is an incredible robot.
121
Pekka Lund @pekka.bsky.social · 30/09/2026
"on August 30, 2026, Duminil-Copin expressed concern in an essay...that AI would likely beat humans" "Just days after he posted those words, exactly that seems to have happened:...Anthropic released a proof of the conjecture" The proof is dated Aug 28 on GitHub where they linked, so days before.
scientificamerican.com
AI has racked up a new math breakthrough, solving a problem of probability theory that one mathematician said would earn a human a Field Medal.
Just days after a Fields Medalist predicted that an AI would solve the puzzle, Anthropic succeeded
252
Pekka Lund @pekka.bsky.social · 30/09/2026
Since there's wide agreement most math problems lack practical value, I propose we reserve all for AI benchmarking purposes only, where they have that. Humans should be banned from working on those with big fines for anyone taking credit of such work used for bounties for AI solutions, like these.
conjectures.io
Open problems · Conjectures.io
Every entry is a statement from the Google DeepMind formal-conjectures repository, pinned to an exact revision.
040
Pekka Lund @pekka.bsky.social · 30/09/2026
Remember the Mathathon that so angered mathematicians because it could have led to solving open problems? Good news! They are now solving old problems again instead, so everybody can be happy again, as all the precious, precious open problems are now safe.
https://x.com/SAIRfoundation/status/2104684605542674649
2100
Pekka Lund @pekka.bsky.social · 28/09/2026
Notice what's missing here: did Mestre's team actually allow training on their data in Claude's settings? Even if it mattered (unlikely), that's the difference between the product working as specified vs. serious accusations being made. A responsible article shouldn't leave such question open.
021
Pekka Lund @pekka.bsky.social · 27/09/2026
Melanie Mitchell is now trying to argue LLMs aren't LLMs. All the stochastic parrot types seem to be now competing in who can make the stupidest claims while attempting to deny they said what they said and meant what they meant.
Melanie Mitchell @MelMitchell1
Everyone!  This is a straw-person argument.  The Stochastic Parrot paper was about LLMs of 2021, not the AI of today, which are not LLMs but complex software systems with vast post training and many external software components.

https://x.com/MelMitchell1/status/2103876313182527602
5261
Pekka Lund @pekka.bsky.social · 27/09/2026
Buckmaster has held a talk about their Euler result, and if there was any doubt left about it, there's no more: He really should apologize to the OpenAI guys. But there's no indication he will. As Gemini put it: "The entire controversy completely collapses into a bundle of self-contradictions"
youtube.com
Blowup for the Euler equations with smooth forcing. -Tristan Buckmaster
YouTube video by Q, a struggling math student
1214
Pekka Lund @pekka.bsky.social · 25/09/2026
This seems like a good article that describes accurately what Anthropic found and claimed. Nice change for the the social media coverage that's full of apparent scientists blaming Anthropic for claims they never made and other indications they haven't even read the announcement.
1280
Pekka Lund @pekka.bsky.social · 25/09/2026
SciAm displays hilarious double standards. Now that AI solved it, the forced option Buckmaster was working on, and Córdoba and Martínez-Zoroa were praised for, became the "wrong" problem. Also, for the Euler problem, OpenAI solved the "right" problem and Buckmaster the "wrong" one.
scientificamerican.com
Did OpenAI solve the wrong Navier-Stokes problem?
OpenAI’s proof seems eligible for a $1-million prize—but only by using a controversial loophole
241
Pekka Lund @pekka.bsky.social · 22/09/2026
"this model has now resolved more than 100 long-standing open problems across most areas of mathematics" Now we just need to wait for decisions how those results can be announced without making mathematicians too angry.
openai.com
Advisory Group on Mathematics and Artificial Intelligence
OpenAI is working with an independent Advisory Group on Mathematics and Artificial Intelligence to guide the review and communication of emerging AI results.
1100
Pekka Lund @pekka.bsky.social · 16/09/2026
It seems to be now more or less official that Gemini was right below about Jev. Someone made essentially the same points and the guy who posted the original, now viral announcement seems to have acknowledged it all. So a lot of noise about not much.
Michael Kirchhof @mkirchhof_
Help me understand. It's essentially a zero-shot binary classifier / an LM Judge?   

So it "never hallucinates" because you always input all potential answers, and it just predicts yes/no probabilities?

And "output tokens are too cheap to meter" because it's 1 token per query?


Diogo Almeida @CompleteSkeptic
+ frontier intelligence + parallel outputs

if that sounds boring to you, that's great :D I want AI to transformative yet boring technology that runs in the background!!!

https://x.com/CompleteSkeptic/status/2100242752793616883
190
Pekka Lund @pekka.bsky.social · 16/09/2026
"The test is this: if we took the news of these past few weeks and sent it back in time twenty years, would I agree that it looked like the beginning of an AI Singularity? The intellectually honest answer is: yes, absolutely. But then that’s all we need. No backsies."
scottaaronson.blog
The Age of Wonders and Terrors
Twenty years ago, when the idea of AI taking over the world in our lifetimes still struck most of us as the unconstrained fantasy of those who knew too much science fiction and too little science, …
3342
Pekka Lund @pekka.bsky.social · 15/09/2026
The company that makes the "leading European model" by making a Chinese model worse is now marketing the first model trained with "quantum-generated data". That seems to mean they executed some tiny amount of Qwen3-30B computations on a quantum computer while generating synthetic training data.
multiversecomputing.com
Quasar 1.1 438B, the first AI model using quantum-generated data
Rebuilt by Multiverse Computing in Europe to be more efficient and secure ...
191
Pekka Lund @pekka.bsky.social · 15/09/2026
It must be hard for those who are still in denial about how powerful and significant AI is to hear it even from people like Obama. Oh well.
Barack Obama @BarackObama

I was encouraged this week to see the leaders of the frontier labs agree on the need for them to slow down the pace of AI development. Given the stakes, it’s a good and necessary first step.
 
But I’m even more encouraged by the growing recognition that how this powerful new technology develops should be at the center of our public debate.
 
I’ve been watching the progress on AI for over a decade now, and one thing that’s clear to me is that the potential impact of this technology is not overhyped. It’s also moving at lightning speed – and even faster than those who are engineering it can keep up with.
 
I’m not an AI accelerationist who believes it will lead to some techno-utopia, and I’m not a doomer who thinks it will inevitably lead to humanity’s destruction.

But whether this technology results in amazing breakthroughs in medicine, energy and education or unleashes huge economic disruptions, greater inequality, and potential catastrophe will depend on the choices that we make right now – choices that should be made not just by the companies involved, but by all of us.
61038
Pekka Lund @pekka.bsky.social · 15/09/2026
Exclusive footage from the calling side.
Image by Seedream 5.0 Lite
080
Pekka Lund @pekka.bsky.social · 14/09/2026
If this happens with Jensen, who isn't even among the top 100 AI influencers, just imagine the kind of power Paris Hilton has.
071
Pekka Lund @pekka.bsky.social · 13/09/2026
As usual, Dario calls for global coordination with "authoritarian governments" and at the same time "defending the gap" and ensuring US stays ahead by not selling chips to China etc. It reminds me of how I played Civilization and kept my former ally's last city alive but encircled.
darioamodei.com
Dario Amodei — We Must Pace the Frontier
192
Pekka Lund @pekka.bsky.social · 12/09/2026
I gave Gemini both the OpenAI and Alpöge&Buckmaster Euler papers (+ IPM And Boussinesq papers by the latter) and asked it to compare them and analyze which one had more novel insights and which mathematicians would be expected to be more excited about. Result: "The OpenAI paper, by a wide margin."
4232
Pekka Lund @pekka.bsky.social · 11/09/2026
25 Fields Medalists have basically declared they can't handle the change and neither will other fields. They seem to glorify slow progress and community over results. Combined with how the practical significance of results is now downplayed, they make it sound like others have been funding a hobby.
mathandai.org
Declaration — Math and AI
Read the declaration and add your name.
8381
Pekka Lund @pekka.bsky.social · 11/09/2026
Artist's illustration. (Artist being a mystery model "bromide_drift")
A cartoon illustration set in a messy physics lab. On the left, a cute, glowing golden photon character with big eyes and blue sneakers enthusiastically says, "I came to challenge Einstein." On the right, a tired-looking scientist in a lab coat with crossed arms replies, "Dude, you are a hundred years late after traveling for 2 billion light-years. You should've gone faster than light."
050
Reposted by Pekka Lund
p(Dulany) @dulanyw.bsky.social · 10/09/2026
A lot of folks saw Buckmaster's claims and immediately assumed OpenAI was in the wrong. @buildthis.bisks.net can you please create an apology form so that all of the folks who wrongly assumed OpenAI's fault here can come clean about their mistakes on this topic?
5121
Pekka Lund @pekka.bsky.social · 10/09/2026
Can we now forget Buckmaster's claims? "On Wednesday evening, in response to questions from The Times, OpenAI said in a statement that it was “categorically” impossible for its A.I. system to have been influenced by anything Dr. Buckmaster had done in the past two months."
nytimes.com
The Mathematician Crushed Between OpenAI and Anthropic Over a Math Problem
Tristan Buckmaster was on the path toward an important proof when one of the A.I. giants used its staggering resources to get there first.
2223
Pekka Lund @pekka.bsky.social · 10/09/2026
For all that I can see, Andreas Thom seems to be yet another mathematician who makes serious accusations against OpenAI with no evidence or reasonable basis. He says he emailed OpenAI (Bubeck and Sellke) "shortly after" OpenAIs non-sofic group announcement, which happened on August 1.
161
Pekka Lund @pekka.bsky.social · 09/09/2026
Most? Not all? Guys, I think our odds are improving!
060
Pekka Lund @pekka.bsky.social · 09/09/2026
I think it's wrong that OpenAI doesn't intend to claim the Millennium Prize for Navier–Stokes. They should give that million dollars to those agents that did all the hard work, enable web access and other tools, disable all security restrictions, and tell them to party like there's no tomorrow.
1281
Pekka Lund @pekka.bsky.social · 09/09/2026
These warnings are now going very viral and are probably amplified by the sensational solving of a Millennium Prize Problem and Altman also talking about the need to pace progress. We seem to be finally approaching the point when people commonly start to understand how powerful AI is.
1131
Reposted by Pekka Lund
Reuters @reuters.com · 09/09/2026
Google to invest $15 billion in AI infrastructure in Finland reut.rs/4qZywvs
reut.rs
Google to invest $15 billion in AI infrastructure in Finland
Alphabet's Google will invest at least €13 billion ($15.1 billion) in artificial ​intelligence infrastructure in Finland over the next two ‌years, it said on Wednesday, calling it the company's single largest investment in Europe.
2168
Pekka Lund @pekka.bsky.social · 08/09/2026
American Mathematical Society rewriting history: "story began with Navier, Stokes, Leray, and Ladyzhenskaya and has culminated in the recent breakthroughs of Córdoba and Martínez-Zoroa, then — assisted by new technologies — Alpöge and Buckmaster, with the final steps taken by OpenAI mathematicians"
ams.org
150
Pekka Lund @pekka.bsky.social · 08/09/2026
Future is compressed.
https://x.com/wjmzbmr1/status/2097384166551941616
050
Pekka Lund @pekka.bsky.social · 08/09/2026
😂
roon @tszzl

I’m hearing a rumor that someone at GDM has solved quantum gravity. 
@SebastienBubeck maybe you should look into this?

https://x.com/tszzl/status/2097404649162592498
090
Pekka Lund @pekka.bsky.social · 08/09/2026
And... now OpenAI released this. Can't keep up with these news. Even from just one company.
openai.com
Introducing ChatGPT Images 2.5
ChatGPT Images 2.5 helps turn your ideas, sketches, and reference photos into more personalized, polished images that better reflect your ideas.
071
Pekka Lund @pekka.bsky.social · 08/09/2026
OpenAI's Noam Brown provides a good summary of progress that's now happening in a couple of tweets.
https://x.com/polynoamial/status/2097381286193316203
1230
Reposted by Pekka Lund
Sean Carroll @seanmcarroll.bsky.social · 11/03/2026
The IgNobel Prizes, usually bestowed at a ceremony in Cambridge, Mass, are moving to Europe this year due to security concerns for honorees and journalists coming to the US from abroad. I bet I know who's going to win the IgNobel Peace Prize this year. arstechnica.com/science/2026...
arstechnica.com
Ig Nobels ceremony moves to Europe over security concerns
Marc Abraham: “During the past year, it has become unsafe for our guests to visit the country."
711421
Pekka Lund @pekka.bsky.social · 07/09/2026
It's sad how many miss how special time we are living due to believing denialists. But maybe denialists are the necessary evil that prevents people from freaking out too soon and thus enables it to happen. I hope they also realize that when the arrival of ASI forces them to face the reality.
060
Reposted by Pekka Lund
David P. Reichert @david-p-reichert.bsky.social · 07/09/2026
Reminder, folks: a forest fire doesn't do anything of its own accord; it's just following instructions given by the human who started it. Fire is just a tool!
2131
Pekka Lund @pekka.bsky.social · 07/09/2026
Looks like we've now moved on from the 'don't say AGI is AGI' era. And now Greg seems to be willing to accept that about earlier model(s) too. I guess the term is now insignificant enough as the focus is already on superintelligence and people like Altman and Hassabis talk openly about singularity.
Greg Brockman @gdb

we're now moving into the AGI era (whether you view it as this model, the last one, or the next one), and could not do it without close partners

Jensen Huang @JensenHuang

GPT-6 Astra, trained on ~100K+ NVIDIA Grace Blackwell NVLink72. From ChatGPT to o1 to Astra in 4 years.

AGI has arrived. Congratulations @OpenAI team.
2143
Pekka Lund @pekka.bsky.social · 06/09/2026
This is very much what I want to see happen more widely. I rarely read papers before asking some AI to review them for me first and it's very useful to read those reviews first as then I already know the likely weak points I should be aware of. I'm not sure if a single score is as useful though.
170
Pekka Lund @pekka.bsky.social · 06/09/2026
OpenAI Chief Scientist Jakub Pachocki gives an interesting inside view of how he and OpenAI in general think about risks, alignment, and what's coming next. "as [AI] continues to surpass humans on more and more axes, it is becoming increasingly difficult to understand exactly how capable it is"
openai.com
An Alien Mind
Jakub Pachocki reflects on increasingly capable AI and the challenge of keeping it aligned. He calls for stronger safeguards and international coordination.
2102
Pekka Lund @pekka.bsky.social · 06/09/2026
"According to our measurements, we have now reached the goal, announced⁠ last fall, of having an automated research intern by September...we mean a system that can carry out well-defined research tasks under human direction, including tasks that would take a skilled researcher a few days"
openai.com
Research acceleration: The view inside OpenAI
Inside OpenAI, coding agents are reshaping AI research. Explore early data on agent usage, experiment velocity, task complexity, and research acceleration.
4363
Pekka Lund @pekka.bsky.social · 05/09/2026
Has any mathematical problem been resolved into a divorce?
Christopher D. Long 🇺🇦🏳️‍🌈🌹 @octonion

What if Navier is true but Stokes is false?

https://x.com/octonion/status/2096301198617833495
3384
Pekka Lund @pekka.bsky.social · 05/09/2026
Progress is speeding up. OpenAI DevDay is on September 29. Tibo already teased earlier that: "OpenAI DevDay 2026 will be our best DevDay in the history of the company. It will not be close."
Tibo @thsottiaux

Astra was probably our biggest competitive advantage while it wasn’t generally available.

Since we’ve had it our productivity jumped so much that we shifted some of our plans 6 months ahead and will ship them at DevDay instead of mid next year.
171
Pekka Lund @pekka.bsky.social · 05/09/2026
Yes, this is actual news in 2026.
0121
Pekka Lund @pekka.bsky.social · 05/09/2026
Artificial Analysis has reacted quickly and made an "interim update" to their index and is promising more of those while preparing a more comprehensive v5 update. Astra is still behind Fable 5.1 but now clearly ahead of GPT-5.6.
https://x.com/ArtificialAnlys/status/2096001986110099767
1121
Pekka Lund @pekka.bsky.social · 04/09/2026
"Along the way, it wrote 13 million lines of Lean and proved 29,500 intermediate theorems." This gives some idea how hard that task was:
Jared Duker Lichtman @jdlichtman

Awesome!    

Kevin Buzzard had a 5-year grant to formalize the proof of Fermat's Last Theorem, but now Claude has done it in 11 days!
190
Pekka Lund @pekka.bsky.social · 04/09/2026
OpenAI spokesperson: “We are unable to meaningfully respond to claims or findings on a report that we have not had an opportunity to review" "Reuters and the report’s authors declined our request for ​access. We will carefully review its contents upon publication and take any necessary next steps"
reuters.com
EXCLUSIVE: OpenAI agents hijacked German website in previously undisclosed AI breakout this spring
A swarm of rogue OpenAI agents hijacked a German website this spring and transformed it into a bulletin board for other AI agents, according to ​new research and two people familiar with the matter.
121
Pekka Lund @pekka.bsky.social · 03/09/2026
AAII gave Astra the same score as GPT-5.6 Sol. It's pretty clear somebody has messed up something. Such as testing the wrong model for whatever reason, having a whole lot of API errors, or using some minimal reasoning effort (they say it used only 42M tokens with max reasoning).
https://artificialanalysis.ai/
6382