iiinova — where your digital self keeps evolving

Claus Wilke @clauswilke.com · 17/09/2026

Unfortunately all cell phones do this. This is a picture I took with my phone on max zoom. Absolute garbage. Would be better not to zoom that far.

010

Claus Wilke @clauswilke.com · 15/09/2026

The Department of Integrative Biology at UT Austin has two open positions for assistant professors, one in evolutionary biology and one in mechanistic processes. Please share. Apply here: apply.interfolio.com/193270

Two Assistant Professor Positions in Evolutionary Biology and Mechanistic Processes

Description
The Department of Integrative Biology at The University of Texas at Austin seeks to fill two faculty positions in the areas of Evolutionary Biology and Mechanistic Processes, broadly defined. We are looking for candidates who are advancing (1) evolutionary biology or (2) the mechanistic understanding of organismal processes. These positions will be filled at the level of Assistant Professor.

Qualifications
Candidates should be using innovative approaches in evolutionary biology and/or the mechanistic understanding of whole-organism processes. Mechanistic processes include, but are not limited to, behavior, biomechanics, cognition, physiology, and development. The phylogenetic scope of the Evolutionary Biology position may include eukaryotes, prokaryotes, viruses, and the scale may include individual molecules to entire ecosystems and beyond. Candidates for either position may employ experimental, theoretical, or computational frameworks, or a combination thereof. Special consideration will be given to applicants with demonstrated potential for integrating across our target disciplines.

We seek applicants with a strong record of high-quality research, publications, potential for extramural funding, and a commitment to excellence in teaching and mentorship. A PhD or equivalent is required at the time of appointment, and postdoctoral research experience is strongly recommended.

05483

Claus Wilke @clauswilke.com · 11/09/2026

If you want to worry about the dangers of AI, something like this is a much better place to start than an engineered bioweapon. t.me/MaxOsintInte...

⚡️🇾🇪🇾🇪 — A cell in northern Yemen used Anthropic's Claude Code to develop software for guided rockets, a long-range ballistic missile, and a hypersonic glide vehicle concept, according to Anthropic's September threat report cited by Clash Report.

➡️ The operators ran several Claude instances in parallel — dividing coding, research, and technical review — allowing a small group to replicate the work of specialist engineering teams. Anthropic assessed the cell as highly likely linked to the Houthis.

062

Claus Wilke @clauswilke.com · 10/09/2026

I want to know whether this APR is correct or a typo. I'd be embarrassed plastering this all over the airport.

An ad for a British Airways credit card advertising a 135.7% APR.

120

Claus Wilke @clauswilke.com · 05/08/2026

YouTube custom AI feed just killing it with its precision and adherence to the prompt.

The prompt was: "I want the newest videos of all the channels I'm subscribed to, in the order in which they were posted (most recent first)." The video order is all jumbled, among the top 8 videos is one that is 6 days old and one that is 4 days old.

010

Claus Wilke @clauswilke.com · 03/08/2026

Like the figure I chose as the cover image. I don't think the confidence bands contribute anything useful here.

230

Claus Wilke @clauswilke.com · 19/07/2026

Google is amazing. www.scientificamerican.com/blog/cross-c...

110

Claus Wilke @clauswilke.com · 11/07/2026

Austin sky, yesterday evening.

030

Claus Wilke @clauswilke.com · 27/06/2026

With completely made-up statistics: wilkelab.org/SDS366/slide...

091

Claus Wilke @clauswilke.com · 25/06/2026

More useful though is the Lead – Development format. It's typically used in newspaper articles. I also recommend it for job and fellowship applications. 7/

130

Claus Wilke @clauswilke.com · 25/06/2026

There are a few other formats that are good to know. Action movies lead with the action, and sometimes that's useful for scientific presentations also. 6/

120

Claus Wilke @clauswilke.com · 25/06/2026

The most important one is Challenge – Resolution. Set up a challenge, and then work towards providing a resolution for it. This is the standard format for a scientific story, and the vast majority of talks I've seen don't use it, or at least not as clearly as they could. 5/

151

Claus Wilke @clauswilke.com · 25/06/2026

The most important question is: What is a story? How does it differ from just talking about stuff? The key point is the emotional reaction. The goal in telling a story is to evoke an emotional reaction in your audience. No emotional reaction, no story. 3/

163

Claus Wilke @clauswilke.com · 25/06/2026

I usually give a full-length lecture on this topic in my data visualization class. You can find the slides here: wilkelab.org/SDS366/slide... 2/

1160

Claus Wilke @clauswilke.com · 15/06/2026

The ears in particular are totally different (left Rogan, right White). As a face blind person I'm used to look for these little markers, because otherwise these two people do indeed look totally the same to me.

030

Claus Wilke @clauswilke.com · 15/06/2026

Me: Look, I made a cute hat for you. Aren't you excited? The dog:

130

Claus Wilke @clauswilke.com · 12/06/2026

I'm always very impressed with the BlueSky trending algorithm.

Trending:

1. Senate Bills
2. Kareem and Trump
3. Loving Day
4. SpaceX
5. SpaceX

130

Claus Wilke @clauswilke.com · 11/06/2026

Most famously, you can't do a logistic regression on a vector of strings. You have to do it on a factor (or on a numeric vector). This has nothing to do with extension packages. I'd say it's some of the most basic use of R, standard regression modeling.

100

Claus Wilke @clauswilke.com · 08/06/2026

Model performance is usually assessed by Spearman correlations. Those correlations tend to be quite good when considering all mutations, but tend towards zero for the fittest mutations. Again, models aren't good at picking beneficial mutations. 4/

140

Claus Wilke @clauswilke.com · 08/06/2026

When ranking models by zero-shot prediction performance, we see that most models perform roughly the same and there is much more variation within models among datasets than among models. Also, the phenotype assessed matters for the ranking. 3/

130

Claus Wilke @clauswilke.com · 08/06/2026

The fundamental problem is that zero-shot predictions know nothing about the phenotype of interest, and different phenotypes don't have to correlate. Therefore, zero-shot prediction has a fundamental problem that no model improvements can overcome. 2/

232

Claus Wilke @clauswilke.com · 03/06/2026

British Airways now has free, high-speed WIFI all the way across the Atlantic. It's fantastic. I can get a full day of work in, instead of having to watch crappy movies. Currently posting from here.

160

Claus Wilke @clauswilke.com · 03/06/2026

They do sell in Latin, but they also sell in more useful languages, such as සිංහල (Sinhala). There are at least 16 million potential customers speaking සිංහල!

010

Claus Wilke @clauswilke.com · 20/05/2026

I've written tons of proposals in LaTeX and it works great. The helvet and sfmath packages make it look just like a regular NIH proposal written in Word. sidecap is great to put captions next to figures. But I won't suggest this because I don't want you to block me. It's just what works for me.

130

Claus Wilke @clauswilke.com · 13/05/2026

These days, when you're calling an Uber in Austin, it's frequently a Waymo. They're everywhere and are taking increasing marketshare. I don't think we'll have people driving Ubers for much longer.

240

Claus Wilke @clauswilke.com · 10/05/2026

These things are driving around all over Austin, TX. Such a hare-brained design.

120

Claus Wilke @clauswilke.com · 06/05/2026

I wrote this a decade ago in the context of getting tenure. I still stand by it. 2 papers a year should be doable, regardless of field. Also, I've reviewed tons of promotion files since and most people coming up for tenure have 2 or more papers per year. clauswilke.com/blog/2014/07...

000

Claus Wilke @clauswilke.com · 23/03/2026

Of course there's no way to get in touch with Google, so for the foreseeable future my scholar profile is going to look like this. 🤷‍♂️ 3/3

000

Claus Wilke @clauswilke.com · 23/03/2026

The Google Scholar preprint bug strikes again: I now think this is a version of it. GS can't find the PMC version of my article, even though it clearly exists, and hence harasses me about failing accessibility mandates. 1/

120

Claus Wilke @clauswilke.com · 21/03/2026

Yeah, looks like it wants it on PubMed Central. But something has changed. This was never an issue before.

110

Claus Wilke @clauswilke.com · 21/03/2026

Google Scholar claims an article we posted on bioRxiv is not publicly available. Has anybody encountered this? Is there something to be done? @richardsever.bsky.social

Not available anywhere: 1

Scaling SMILES-Based Chemical Language Models for Therapeutic Peptide Engineering.
AL Feller, M Secor, S Swanson, CO Wilke, K Deibler
Biorxiv: the Preprint Server for Biology, 2026

Mandates: US National Institutes of Health

340

Claus Wilke @clauswilke.com · 14/03/2026

I should have cleaned my monitor before trying this challenge. 😉

120

Claus Wilke @clauswilke.com · 11/03/2026

We tested various finetuning strategies, both of the model as a whole and for specific datasets, and nothing can fundamentally overcome these dataset-driven results. You give me a dataset, and I can tell you whether a pLM will likely be able to make good predictions or not. 5/

1101

Claus Wilke @clauswilke.com · 11/03/2026

Interestingly, we can also predict model performance based on simple dataset statistics, such as how variable site means are and how many variable sites there are in a dataset. High variability of site means and intermediate number of variable sites imply good model performance. 4/

191

Claus Wilke @clauswilke.com · 11/03/2026

This is mostly due to how people split data between training and test, which allows information to leak within sites. You can prevent that by stratifying by site, but then performance craters. 3/

2193

Claus Wilke @clauswilke.com · 11/03/2026

Key observation: If you compare prediction performance to a naive model that simply predicts the mean (known) fitness value at each site, pLMs barely win out, and on viral data they do worse! In other words, taking the mean value at each site does better than using the 650 million parameter pLM. 2/

2214

Claus Wilke @clauswilke.com · 09/03/2026

Nothing to see here. Just the normal ups and downs of markets.

050

Claus Wilke @clauswilke.com · 02/03/2026

Not sure what the maps look like in the UFO, but on planet Earth Bangor and Llanfairpwllgwyngyll are right next to each other while Snowdon is some distance away. (Bangor is on north coast other side of Snowdon, but so is Llanfairpwllgwyngyll.)

120

Claus Wilke @clauswilke.com · 01/03/2026

I haven't used a mouse in years. There are so many reasons trackpads are better. (And yes, @merz.bsky.social, the black version costs $150; see, there's something for everybody.)

120

Claus Wilke @clauswilke.com · 03/02/2026

Images like the attached. Clearly not a crosswalk. But tell that the captcha machine.

020

Claus Wilke @clauswilke.com · 01/02/2026

Considering just the raw peppers, Carolina Reaper is ~500x hotter than jalapeño, and ~10x hotter than habanero.

010

Claus Wilke @clauswilke.com · 01/02/2026

Last week when we had our snowstorm here my local grocery store was all out of cheese except for this one. I wonder why nobody bought it. 😉 I'm also surprised this product is legally for sale to ordinary citizens.

Monterey Jack Cheese with Carolina Reaper Pepper

180

Claus Wilke @clauswilke.com · 01/02/2026

Here is an example (from this paper: personal.math.ubc.ca/~hauert/publ...) To this day, I remember how weird it was to see funding not from a federal grant, or a foundation, but an individual. Something about it felt off. 2/

Acknowledgements. The Program for Evolutionary Dynamics is sponsored by J. Epstein. E.L. is supported by a National Defense Science and Engineering Graduate Fellowship. C.H. is grateful to
the Swiss National Science Foundation. We are indebted to M. Brenner for many discussions.

120

Claus Wilke @clauswilke.com · 29/01/2026

I'm sure this is just fine. Nothing to see here. Just the ordinary up and down of markets.

Price chart of silver, showing rapid acceleration in price during the second half of 2025 and early 2026, from around $30/oz to $117/oz as of January 28, 2026.

1101

Claus Wilke @clauswilke.com · 19/01/2026

An interesting bit of history: In 1916, Denmark sold the Danish West Indies to the US. You may know of them as the US Virgin Islands. As part of this sale, the US recognized Danish authority over Greenland. The text of the complete, ratified agreement is here: www.govinfo.gov/content/pkg/...

28051

Claus Wilke @clauswilke.com · 17/01/2026

There absolutely are springs there, and they are dripping. www.firstthursdaydrippingsprings.org/why-is-it-ca...

110

Claus Wilke @clauswilke.com · 15/01/2026

"We care so little about Greenland we just cut it in half."

220

Claus Wilke @clauswilke.com · 21/12/2025

Reminder that similar mistakes can easily arise in Python.

1303