Tom Wenseleers

Dave Queller and colleagues have written about this - you can check out royalsocietypublishing.org/doi/abs/10.1... royalsocietypublishing.org/doi/epdf/10.... www.cell.com/current-biol... The PDFs are on Sci-Hub I believe (sci-hub.hkvisa.net).

How often are children genetically unrelated to their presumed fathers? Tackling a touchy subject, genetic detective finds only 1% of European children have unexpected paternity

Honoured to see an article about my research on extra-pair paternity in @science.org. Many thanks to journalist @spoke32.bsky.social for the story and to all colleagues for their quotes!
Hopefully, future visitors to my office will be less surprised by a (reproduction) 17th-century painting...

Big congrats Maarten!

Assessing the role of children in the COVID-19 pandemic in Belgium using perturbation analysis - Nature Communications Estimating the contribution of different age groups to COVID-19 epidemic dynamics is complicated by changing variants, non-pharmaceutical interventions, and vaccination levels. Here, the authors use l...

Cool new article out in Nature Communications by @hensniel.bsky.social and colleagues: "Assessing the role of children in the COVID-19 pandemic in Belgium using perturbation analysis". www.nature.com/articles/s41...

Well Swiss NSF has 15% overhead and Belgian FWO has 6% overhead. But hard to compare countries...

Yes I get that - a greater share of the operating costs of European universities is state funded compared to in US public universities. But I would still find it a little excessive that for a 1 million USD grant 0.5 million would be needed to pay for the heating & cleaning of my lab & admin support.

What do you mean exactly? It just struck me that those high overheads in the US don't seem to translate into making their universities more democratic. Tuition fees are sky high. But OK taxes are a little higher in Europe...

In Belgium the overhead on FWO projects is 6% and the Swiss NSF has 15% overhead. EU horizon 25%. Why are the percentages so much higher in the US?

If you opted-in then in principle - yes, any confidential fine grained facts could leak into the next model. But not if you opt out. Example above is a bit more tricky: the student could opt in, but the lecturer might prefer this material not to be used as training material for the next model...

How your data is used to improve model performance | OpenAI Help Center Learn more about how OpenAI uses content from our services to improve and train our models.

Well from what is written there it seems if you opt in they can use your chats to help train subsequent models & presumably also use it for the RLHF part. But before doing so they would remove any personal information. So in example above the course material could become part of the training set.

That option of allowing your data to be used for the next model can be switched on or off in the settings... But yes, a bit tricky potentially when people are pasting in copyrighted material. I recall that when you paste in PDFs you are not allowed to publicly share those conversations though.

Happy New Year to you all! My Christmas holidays involved helping my son make a DNA molecule model using a 3D pen & getting a new pet for him: leaf insects! Great fun!

In first grade we just demonstrated a Makeblock robot with a mobile phone mounted on its back with ChatGPT Advanced Voice Mode, to pretend the robot was able to talk! The kids & teacher absolutely loved it! It answered all the kids' questions for a full hour! It could even talk with a robot voice!

Well since it needed about twice as much time to do the 35 questions (almost 40 minutes instead of the 25 allowed minutes) it was not particularly fast no. With 40 minutes it got an IQ of 88. For a vision AI that's very good - other vision AIs only get around 70. But I had expected more...

That's definitely not true - in nearly all of the benchmarks ChatGPT o1-preview comes out as the best model & for most of my coding needs & scientific work it works admirably well. An IQ score of 88 is still quite a good score for a vision AI. Previously, the best vision AIs only got around 70.

If I gave it twice as much time to be able to get through all 35 puzzles, ChatGPT o1 gets an IQ score of just 88... Not great...

Massive breakthrough in AI intelligence: OpenAI passes IQ 120 OpenAI's new "o1" model may have an IQ as high as 120

So that earlier post was probably overblown... www.maximumtruth.org/p/massive-br...

Now that ChatGPT o1 actually supports image upload I thought I would give it that Mensa Norway IQ test once more, test.mensa.no/Home/Test/en, but out of the 35 puzzles it only manages to solve 20 within the given 25 min time limit & then ends with an IQ below 100. :-(

A new platform, a new thread. Here's a thread explaining our recent work on #evolution in finite populations, now available ahead of print in @asn-amnat.bsky.social 🧬🧪

Do you want to #model quantitative trait #evolution? Do you like #SPDEs? Happy to say my first solo author paper, now out in Theoretical Population Biology, is about just this! Check it out here:

www.sciencedirect.com/science/arti...

I'm super proud of this y'all :')

#scisky 🧪 🧬 #mathsky #physics

And if course there is different philosophies in terms of either liking simple highly abstracted models vs more complex and realistic ones. Or ones focusing more on particular inheritance mechanisms vs studying the abstract dynamics as such...

Equally, people studying eco-evolutionary dynamics rarely if ever can make use of inclusive fitness techniques. All the stochastic effects in their models usually forces them to use statistical physics approaches or individual based models. Yet this is a fruitful field...

Not sure if they are pointless. Different evolutionary frameworks can be more or less intuitive for studying particular systems. For social insects I found inclusive fitness intuitive and helpful. For microbial systems where mutation typically needs to be added to any model I don't like to use it.

As my son thought some of the graphs were not quite exciting enough, I've also included a fun script that produces various cool visualisation - various fractals etc..., including a couple of 3D ones : www.dropbox.com/scl/fi/qbwo2...

Enjoy!

The lecture material of my latest data visualisation class in R, using both base R & ggplot, which was part of our Advanced Biological Data Analysis course here : www.dropbox.com/scl/fi/0via9... #rstats

The popular summary in the Economist was also cool: www.economist.com/science-and-...

Nice perspectives article about our earlier Current Biology study on the inheritance of alternative nest architectures in stingless bees led by my postdoc Viviana Di Pietro: link.springer.com/article/10.3... Original article here: www.researchgate.net/publication/...

Last Sunday, we had a blast at Day of Science! Truly amazing to see kids (and parents) having fun with Biology. We made bee hotels, gave some info and got them acquainted with some common social insect species in Belgium.

@kuleuven.bsky.social
@twenseleers.bsky.social

Ruben's self balancing robot just about made it to the finish line. Great fun those Makeblock robots! :-)

Was a bit confused though by the fact that on page 12 he suggests to just fit a maximum likelihood lmer mixed model. Or is that part heavily outdated?