ChatGPT5.4 (Extended Thinking) feels no better at math problem solving than 5.2 did. This is the first model that does not feel significantly better than the previous ones at this. Are we finally hitting a ceiling? My timeline to math research automation and AGI just got a bit longer.
The main file is called CMakeLists.txt. I mean, already off to a bad start. Good names are important, and this is not that. And it only gets worse from there.
I hate CMake with a burning passion. It is the worst thing ever. An unmitigated disaster. A throwback to the 90s. Which is impressive as it was released in the 00s.
Castles in the sky.
64GB of DDR5 RAM costs... 1000 dollars now? Bullish for succinct data structures!
Donald Knuth is impressed by the creative problem solving capabilities of Claude.
Kuva means "picture" in Finnish. Is that where the name is from?
Something that used to be a weekend project is now a 15-minute project. This unlocks so much that wasn't worth doing before.
The latest Claude seems to be able to synthesize small apps from a short description pretty much autonomously. I'm having hard time adjusting to this reality. I can now make small bespoke productivity apps just for myself. Computers just become a lot more useful for a lot of things.
Once you see it, you cannot unsee it.
It's open for contributions of new constants! github.com/teorth/optim...
Cool repository from Terence Tao: A list of various mathematical constants whose exact value is unknown. The page lists the best known lower and upper bounds for each constant. TCS is represented: e.g. undecidable busy beavers and random 3-SAT satisfiability threshold. github.com/teorth/optim...
Here is a small research project that I did recently with chatbots: github.com/suomela/fin-... — everything is vibe-formalized in Lean 4 (approx. 24500 lines of Lean code).
A quick review of Venice #dsb2026
blog.jnalanko.net/2026/02/20/v...
I just checked your paper and yes, roughly the same idea appears in our paper. We also have an incremental hash function on the color sets. Ours is commutative, atomically updatable and has a clean collision analysis. But it's on sets, not multisets. Looking forward to reading your paper in detail!
It's called Venice. The ceiling is also very nice!
The architecture and art at the DSB 2026 workshop is out of this world. The talks are also great! #dsb2026
The feeling when you try to update some dependencies but get just a compiler error linking to the xkcd comic about all modern digital infrastructure relying on thanklessly maintained open source projects.
An attempt at explaining why I'm finding it hard to get really high productivity boosts from AI-assisted programming. #AI
blog.jnalanko.net/2026/02/02/t...
DSB Program is out !
Seems incredible (as ever)
dsb-meeting.github.io/DSB2026/
I pointed ChatGPT5.2 at the greedy superstring conjecture, open since 1988. It could not prove it for all n, but it found simple arguments for n = 3 and n = 4 strings, which haven't been mentioned in the literature to my (limited) knowledge. n = 5 still open? blog.jnalanko.net/2026/01/20/a... #math
AI fails the Turing test again. It's too smart and knowledgeable to be believable.
I think the time has come to modernize math/CS research. Now that AI is starting to be capable of formalizing proofs, it might soon be inexcusable not to provide formally verified proofs with papers, like trying to ship software without a test suite. This would also make peer review so much easier!
Yeah, it's just numeric answers, like Project Euler. Trivial to verify, but also sometimes guessable without proof.
Thoughts on the FrontierMath benchmark on AI. #AI #math
blog.jnalanko.net/2025/12/26/t...
Why are there no European frontier AI labs?
Korkeakoulujen aloituspaikkoja ei voi mitenkään tolkullisesti lisätä kertaluontoisella rahoituksella, vaan siihen tarvitaan nimenomaan perusrahoituksen kasvattamista. Sisäänotettua opiskelijaa pitää palvella 5+ vuotta, tarvitaan pysyvä lisäys opetushenkilöstöön.
www.hs.fi/politiikka/a...
"Valtiovarainministeriön kansliapäällikkö Juha Majanen sanoi [..] että ylimpien marginaaliveroasteiden keventämisen hyödyistä on vankkaa tutkimustietoa ja toimi rahoittaisi pitkällä aikavälillä itse itsensä."
Mikä ihmeen vankka näyttö? www.uusijuttu.fi/juttu/sZRTks...
#politiikka #puoliväliriihi
More on latest AI models. The AI deep research feature from OpenAI is impressive, but critically, it has no BS detector. I asked for a literature review of a field I know well. It repeats all the hype authors put into introductions of papers and all the cherry-picked highlights in the conclusions.
4/3 on the other hand, it hallucinates trying to describe what the Spectral BWT is. It confidently says that it's the "Suffix-bounded wavelet tree (SBWT)".
