iiinova — where your digital self keeps evolving

Tom Wenseleers @twenseleers.bsky.social · 07/10/2026

All the AI haters on bluesky is definitely one of the single most annoying aspects of this platform...

Tom Wenseleers @twenseleers.bsky.social · 07/10/2026

Open AI ran their GPT Pro model on 4000 open maths problems for 3h each & ended up with 722 manuscripts & 377 families of solutions. The rest it didn't manage (in 3h). Pretty amazing! Manuscripts were all posted at github.com/openai/math. openai.com/index/sharin... Science renaissance coming.

010

Tom Wenseleers @twenseleers.bsky.social · 10/09/2026

I still have these 3D printed Antikythera parts lying around - still have to put it together with my youngest son! printables.com/model/284372... An even better working version here etsy.com/be/listing/1... Not to be underestimated, those Old Greeks! Just a shame they put the earth in the middle!

printables.com

Antikythera Mechanism by FreddyZ | Download free STL model | Printables.com

First Mechanical Computer. Replica of the ancient mechanism found in a shipwreck near Antikythera. | Download free 3D printable STL models

011

Tom Wenseleers @twenseleers.bsky.social · 10/09/2026

Amazing! An interactive 3D replica of the Antikythera Mechanism, straight out of GPT-6 Astra! With 407 downloadable parts to 3D print. antikythera-cosmos-in-bronze.alexandrosm.chatgpt.site

150

Reposted by Tom Wenseleers

John Burn-Murdoch @jburnmurdoch.ft.com · 03/09/2026

I hereby declare the "learn to code" era officially dead: Big declines in the number of people studying computer science in the last year or two 📉 Chart from this week’s edition of our newsletter on AI and the labour market www.ft.com/content/9183...

67945236

Tom Wenseleers @twenseleers.bsky.social · 29/08/2026

Just send me a message at tom.wenseleers@kuleuven.be :-)

010

Tom Wenseleers @twenseleers.bsky.social · 28/08/2026

This was a short talk "Working Smarter with AI: Coding, Analysis and Scientific Writing" that I gave here in our department in January 2026. As things are moving so fast it needs a little updating, as now I work a lot more with Codex, which is way better still... www.dropbox.com/scl/fi/ta9k3...

210

Reposted by Tom Wenseleers

Tom Wenseleers @twenseleers.bsky.social · 22/08/2026

Tip of the day: ask Codex to help you spot mistakes or issues in the print proofs you receive... In this Scientific Reports article of mine it picked up way more issues than I could ever have spotted myself by hand... Works like a charm, and took just half an hour...

131

Tom Wenseleers @twenseleers.bsky.social · 23/08/2026

If you would give it a Scientific Reports Latex template and my paper it would have made far fewer mistakes... And then they would not have to charge us 2500 USD in author processing charges...

010

Tom Wenseleers @twenseleers.bsky.social · 23/08/2026

It also costs next to nothing: about 0.25 USD. Makes me wonder why those publishers don't use Codex themselves to typeset the papers in the first place...

110

Tom Wenseleers @twenseleers.bsky.social · 22/08/2026

131

Tom Wenseleers @twenseleers.bsky.social · 07/08/2026

Even if not standardised I noticed that the current LLMs can already redo your analyses. When I let Codex work on a manuscript of mine and gave it the manuscript, data (550 Excels) & R script it would tell me it found a data entry error in one of the Excels, which none of reviewers had picked up on

110

Tom Wenseleers @twenseleers.bsky.social · 07/08/2026

I bet that AI reviewing manuscripts to check for technical issues will soon become routine in many fields... It will make it possible for the stats & data analysis to actually be redone to check for errors, which just isn't feasible by hand...

141

Tom Wenseleers @twenseleers.bsky.social · 04/08/2026

Fair enough, I would of course also like to see it verified by others & formally refereed & published!

000

Tom Wenseleers @twenseleers.bsky.social · 02/08/2026

OK but 9/10 is still pretty good, no? :-) The initial human proof of Fermat's last theorem initially also still contained an error, right?

100

Reposted by Tom Wenseleers

Tom Wenseleers @twenseleers.bsky.social · 24/07/2026

Really enjoyed the Mathematical Models in Ecology and Evolution (MMEE) meeting in Cork! If anyone missed my talk "Early-season helping yields increasing returns to scale at the onset of eusociality" (Tom Wenseleers, Viviana Di Pietro & Ricardo Caliari Oliviera), the PPT is here: lnkd.in/eGXBJRuq

121

Tom Wenseleers @twenseleers.bsky.social · 02/08/2026

How many PIs would universities have to hire to try to solve any of these problems, without getting anywhere?

100

Tom Wenseleers @twenseleers.bsky.social · 02/08/2026

An internal version of Astra, OpenAIs next major model, solved ten major open math problems, for a total cost of $2,000 at Sol API rates. Truly impressive! openai.com/index/ten-ad...

310

Tom Wenseleers @twenseleers.bsky.social · 24/07/2026

PS I am still looking for someone to show me how to derive the population genetic model results using an inclusive fitness logic [equation 6 in the appendix, see below]. If anyone can I'm giving away a crate of good Belgian beer!

010

Tom Wenseleers @twenseleers.bsky.social · 24/07/2026

121

Tom Wenseleers @twenseleers.bsky.social · 07/07/2026

PS Funny that in Shenzhen they put the stuff that's banned at KU Leuven (OpenClaw & other agentic AI) because "too new and dangerous" already in kids toys!

020

Tom Wenseleers @twenseleers.bsky.social · 07/07/2026

Great way also to learn a bit of Linux & Python for both young & old! The LLM & AI integration makes all the programming very engaging! And code is GPL2 open source so you can modify as you wish! docs.sunfounder.com/projects/pid... docs.sunfounder.com/projects/pid... www.youtube.com/watch?v=U18H...

youtube.com

I Built A Robot Dog To Guard My Studio - SunFounder PiDog Review

YouTube video by ctrl-alt-rees

120

Tom Wenseleers @twenseleers.bsky.social · 07/07/2026

Assembling it was great fun & the video with the instructions was very clear! The Raspberry Pi 4B has to be purchased separately, but that's also quite cheap! docs.sunfounder.com/projects/pid...

100

Tom Wenseleers @twenseleers.bsky.social · 07/07/2026

Just got my son a new toy - a Sunfounder robot dog kit! With voice & face recognition, inbuilt text to speech & LLMs (ChatGPT, Gemini, Deepseek) so it can talk back to you & it even has Openclaw to be able to perform entirely new behaviours! Very impressive for just 150 euros!

141

Tom Wenseleers @twenseleers.bsky.social · 15/06/2026

Even more elegantly, Codex could also readily write OpenSCAD code to produce this bee comb design, including debossed letters for the different rows, in just 230 lines of code. STLs and code here: www.dropbox.com/scl/fi/58ags...

100

Tom Wenseleers @twenseleers.bsky.social · 09/06/2026

Python scripts and STLs here for anyone that would like to use them: www.dropbox.com/scl/fi/kplv5...

120

Tom Wenseleers @twenseleers.bsky.social · 09/06/2026

With minimal instruction & in just 13 mins, Codex wrote me a Python script to procedurally generate the STL of a piece of honeybee worker comb & I then had it 3D printed for just 8 euros by the great people at Fablab! Not allowed at KU Leuven, as it's agentic AI. But works great!

150

Tom Wenseleers @twenseleers.bsky.social · 05/06/2026

Link to a presentation I gave earlier this year at the Biology Research Day about "Working Smarter with AI: Coding, Analysis and Scientific Writing" on how to productively use LLMs as part of your research workflow : www.dropbox.com/scl/fi/ta9k3...

020

Reposted by Tom Wenseleers

Jerry Coyne @evolutionistrue.bsky.social · 25/03/2026

There are two new obituaries of Robert Trivers, though the mainstream media, to its shame, still hasn't published one. Steve Pinker has an excellent and long obituary in Quillette, and there's a shorter one in the Times of London. whyevolutionistrue.com/2026/03/25/t...

131

Reposted by Tom Wenseleers

Jerry Coyne @evolutionistrue.bsky.social · 07/04/2026

Some new obituaries of Robert Trivers. whyevolutionistrue.com/2026/04/07/t...

011

Reposted by Tom Wenseleers

Tom Wenseleers @twenseleers.bsky.social · 23/04/2026

Cool - first time I had a LLM produce correct code to calculate a Romanesco broccoli natural fractal mesh & display it in realtime in 3D in R using rgl... ChatGPT 5.4 Extended Thinking R code here: www.dropbox.com/scl/fi/zrmqo...

012

Tom Wenseleers @twenseleers.bsky.social · 23/04/2026

012

Tom Wenseleers @twenseleers.bsky.social · 14/03/2026

First time that a vision model, GPT 5.4 Pro (Vision), now gets the best score on the Mensa Norway IQ Test - IQ 145, 33 / 35 questions correct, just by visually looking at the IQ test! Or IQ 130 (15/16 correct) in the offline test. Amazing how far we have come!

100

Tom Wenseleers @twenseleers.bsky.social · 12/02/2026

FYI @andygardner.bsky.social @stuwest.bsky.social - feel free to forward to anyone you think who might like to take up the challenge :-)

000

Tom Wenseleers @twenseleers.bsky.social · 12/02/2026

For those not so familiar with the whole controversy two nice overview papers by @birchlse.bsky.social giving a nice and balanced overview of the current state of affairs here: www.journals.uchicago.edu/doi/full/10.... academic.oup.com/bioscience/a...

141

Tom Wenseleers @twenseleers.bsky.social · 12/02/2026

Mathematica notebook with our analytical popgen solution can be found here: www.dropbox.com/scl/fi/h7ery... Or if you don't have Mathematica, PDF here: www.dropbox.com/scl/fi/sbkm4...

dropbox.com

100

Tom Wenseleers @twenseleers.bsky.social · 12/02/2026

Now that the Epstein files are apparently also bringing up the old Nowak inclusive fitness controversy - here a fun challenge: show me how to derive our recent eusociality model www.researchgate.net/publication/... using inclusive fitness methodology! A crate of Belgian beer awaits those who can!

researchgate.net

(PDF) Early-season helping yields increasing returns to scale at the onset of eusociality

PDF | The evolution of eusociality, characterized by cooperative brood care, reproductive division of labor, and overlapping generations, represents a... | Find, read and cite all the research you nee...

150

Tom Wenseleers @twenseleers.bsky.social · 08/02/2026

From the Epstein Files, the stated goals of Nowak's Program for Evolutionary Dynamics: "curing the world of cancer, infectious disease, selfishness, and inclusive fitness theory." www.justice.gov/epstein/file...

031

Tom Wenseleers @twenseleers.bsky.social · 07/02/2026

What struck me in this is that Robert Trivers decided not to sign the reply to the Nowak et al 2010 paper based at least in part on that email he received from Karl Sigmund (who also appeared a bit critical of the statistical version of inclusive fitness theory) : www.justice.gov/epstein/file...

justice.gov

020

Tom Wenseleers @twenseleers.bsky.social · 31/01/2026

When was the last time you tried LLMs btw and what model was that? Insane to truly believe they are not useful for coding... The theoretical modelling projects of my students ar overall much better and ambitious than before...

100

Tom Wenseleers @twenseleers.bsky.social · 31/01/2026

If bad quality science ends up published then it's human slop & bad judgement on the parts of authors, reviewers and editors, not AI slop. Like sending in an unedited paper of a student without a thorough check.

100

Tom Wenseleers @twenseleers.bsky.social · 31/01/2026

This was about learning new skills not about efficiency gains in getting the job done: that's at least 39% more efficient - see paper below. And for me: writing roxygen2 R package documentation: 10x faster with LLMs... papers.ssrn.com/sol3/papers....

papers.ssrn.com

AI Agents, Productivity, and Higher-Order Thinking: Early Evidence From Software Development

How are agentic AI systems affecting productivity and the cognitive experience of work? This paper studies the effects of agents on software development by anal

100

Tom Wenseleers @twenseleers.bsky.social · 30/01/2026

Yes exactly it has ChatGPT integrated. That's why I could paste in a ChatGPT written manuscript and ask to have it typeset in Latex...

010

Tom Wenseleers @twenseleers.bsky.social · 30/01/2026

I have much more positive experience with LLMs: allowing me to crack problems that even domain experts couldn't... There will be pros and cons, sure, but overall I can only see it massively lift science to new levels... Code is also relatively easy to check...

100

Tom Wenseleers @twenseleers.bsky.social · 29/01/2026

Why so negative? It's just a tool to make your work easier. The responsibility for the final text lies with the authors, as always. I look as critically at AI output as I look at say a first draft of a student... What's the difference?

120