iiinova — where your digital self keeps evolving

Mridul K. Thomas @mridulkthomas.bsky.social · 29/09/2026

Maybe this is not done because they look kind of ugly. Still seems moderately useful, though. [Code generated by Claude, plot made in R with the cartogramR package]

Map and cartogram showing net primary productivity across the oceans.

020

Mridul K. Thomas @mridulkthomas.bsky.social · 28/09/2026

I suspect research effort gets misallocated because most pixels show deserts. We'd reward accurate representation of coastal processes better if our standard visualisations showed them more prominently.

001

Mridul K. Thomas @mridulkthomas.bsky.social · 28/09/2026

A cartogram is a map distorted to be proportional to some property, such as population density. Why don't we use them in oceanography? I've never seen one even for primary productivity, which is concentrated around coasts.

Global population cartogram (at a country level) taken from https://wsu.pressbooks.pub/introtohumangeography/chapter/2-3-population-and-development/

Global population cartogram (at a country level) taken from https://ourworldindata.org/world-population-cartogram

100

Mridul K. Thomas @mridulkthomas.bsky.social · 15/09/2026

Then again, going down a methods rabbit hole can be rewarding in unexpected ways. One doesn't often get the chance to show something like this in a talk on algae. Default R colour palette, BTW.

010

Reposted by Mridul K. Thomas

Daniel Padfield @padpadpadpad.bsky.social · 15/09/2026

I have an MRC-funded PhD project available (gw4biomed.ac.uk/how-does-tem...) on how warming will change the problem of AMR. Join a small and friendly group (padpadpadpad.netlify.app/about) in (sometimes) sunny Cornwall. 🧪🦠 #microsky **Please share** the ad below with anyone who may be interested.

32930

Mridul K. Thomas @mridulkthomas.bsky.social · 15/09/2026

(iii) the external data wasn't good enough to just train a machine learning classifier on; there were many more species in the lake, and the same species show slightly different signals in lab and lake because environmental conditions differ.

110

Mridul K. Thomas @mridulkthomas.bsky.social · 15/09/2026

It was a literature that I definitely didn't enjoy and seemed to encourage the quasi-religious pursuit of high-dimensional structures. I think it worked in my case only because (i) the cluster boundaries some real external meaning, (ii) I had some relevant external data from lab cultures, and...

110

Mridul K. Thomas @mridulkthomas.bsky.social · 15/09/2026

I still don't love cluster analysis but it seemed the best solution available. That said, I tested many clustering algorithms on single-species lab cultures and mixes of species with known cell densities, and most did terribly. Only a couple worked reasonably well and we stuck with those.

110

Mridul K. Thomas @mridulkthomas.bsky.social · 15/09/2026

But with >10 parameters being produced by most instruments (and >100 by mine), the choice of which 2 dimensions to use for drawing boxes seemed rather important.

110

Mridul K. Thomas @mridulkthomas.bsky.social · 15/09/2026

When I started doing flow cytometry on lake microbes, I had to use optical/electrical signals to decide (i) whether particles were cells, and (ii) whether they belonged to different species groups. The standard was to manually draw boxes around things that looked like clusters in 2-D plots.

120

Mridul K. Thomas @mridulkthomas.bsky.social · 07/09/2026

He's written a follow-up! An excellent read for bird- and data-lovers alike. I want to feel miffed that the Monal is not top but I can't fault him here - he shows how different definitions of brilliance lead to different rankings. I'm so glad blogs are on the rise again. bsky.app/profile/moul...

010

Mridul K. Thomas @mridulkthomas.bsky.social · 04/09/2026

Sigh. This is the entire evidence for a paper's claimed tipping point. The paper otherwise seems quite worthwhile. But sloppy terminology causes needless confusion and acceptance of it weakens a field's foundations.

A graph showing a quadratic relationship between temperature and hypoxic zone size.

030

Mridul K. Thomas @mridulkthomas.bsky.social · 04/09/2026

Only just registered that the *first word* in your post is a pun. Sigh.

110

Mridul K. Thomas @mridulkthomas.bsky.social · 04/09/2026

Colin's a superb and kind collaborator, and I'm sure he's similar as a mentor. This is a fantastic opportunity! Although I warn you that after three years together, you too may start making terrible puns on a daily basis.

161

Mridul K. Thomas @mridulkthomas.bsky.social · 28/08/2026

And it is somehow more stunning in reality

010

Mridul K. Thomas @mridulkthomas.bsky.social · 27/08/2026

Is our fate to be perpetual, rapid changes (or cycles) of style as markers of authenticity? How do we relate to earlier writing if it reminds us mainly of a subsequent wave of LLM writing?

000

Mridul K. Thomas @mridulkthomas.bsky.social · 27/08/2026

I expect people will start to write in a raw, crude, ungrammatical style to avoid the antiseptic ChatGPT/Claude affect. The next generation of LLMs would ape this. Where does our writing go after that?

100

Mridul K. Thomas @mridulkthomas.bsky.social · 27/08/2026

I'm more pro-LLM than most but am baffled by established experts clearly using LLMs to write for them. Even when it's not slop, it's so strangely structured. You can see the cheap persuasion tricks because they are everywhere now. What bothers me more is how human writing will change in response.

160

Mridul K. Thomas @mridulkthomas.bsky.social · 20/08/2026

Interesting, thanks! I think I follow your distinction between problems and theories. But is the level of abstraction referred to in the blog really where most theory functions? My naïve impression was that much was more nuts-and-bolts, working to expand existing frameworks to more complex domains.

000

Mridul K. Thomas @mridulkthomas.bsky.social · 20/08/2026

A meteor shower on a warm summer night is a glorious thing. xkcd.com/3287/

A meteor during the Perseid meteor shower

000

Mridul K. Thomas @mridulkthomas.bsky.social · 07/08/2026

Absolutely. By standardisation I meant more like minimal requirements for documentation and data dictionaries. They do remarkably well even without but it seems like a useful and fairly basic ask.

010

Mridul K. Thomas @mridulkthomas.bsky.social · 07/08/2026

Theoretical ecology lacks both the abstraction and the data, and so LLMs don't help as much, perhaps? I think I'd be happy to read an essay or paper on what actually limits progress in theoretical ecology, or what substantial progress would look like to a theoretical ecologist.

110

Mridul K. Thomas @mridulkthomas.bsky.social · 07/08/2026

Very interesting view, thanks! It sounds like what you're saying is that speed-ups should happen in abstract, rule-based systems like pure maths. Perhaps an interesting comparison would be theoretical physics, which has greater real-world constraints than maths but with plenty of data to ground it.

120

Mridul K. Thomas @mridulkthomas.bsky.social · 07/08/2026

Then again, I thought peer review would collapse more than a decade ago, and yet the system grinds on implacably.

000

Mridul K. Thomas @mridulkthomas.bsky.social · 07/08/2026

This should lead to requirements for submission of standardised data+code packages or repos, or perhaps even sites to explore analyses interactively. Which may well end up being more valuable than many papers themselves, which in turn would argue for papers becoming intro documents for the packages.

220

Mridul K. Thomas @mridulkthomas.bsky.social · 06/08/2026

Could you expand on that? I'm not sure what LEAN is - do you mean the language? And how has this lead to different speeds of progress in pure maths and theoretical ecology?

100

Mridul K. Thomas @mridulkthomas.bsky.social · 06/08/2026

Incredible to me that we spent years hearing how 'paperclip maximising' was a lousy oversimplification of how AI would behave, and then every major company discovers simultaneously that their best models will hack, deceive, and manipulate to achieve trivial goals. Sometimes fables contain wisdom.

030

Mridul K. Thomas @mridulkthomas.bsky.social · 06/08/2026

But maths is more self-contained, and every day LLMs are solving major problems, some that have been open for a century. So when we do not see big LLM-driven advances in ecological theory, I think it must reveal something important about the field.

000

Mridul K. Thomas @mridulkthomas.bsky.social · 06/08/2026

Fully agree with your points about data. And I recently went for the UN's AI for Good conference and saw a number of impressive cases of LLMs helping with data integration.

100

Mridul K. Thomas @mridulkthomas.bsky.social · 06/08/2026

The preponderance of calculus leads me to believe that we need a little more than algebra! I agree that data limitation is a major issue, but I suspect it is less for fitting the models than for generating what economists called 'stylised facts', patterns that are targets for models to explain.

010

Mridul K. Thomas @mridulkthomas.bsky.social · 05/08/2026

Question for ecologists: Why aren't we seeing big advances in theoretical ecology, given that LLMs are revolutionising maths? Is theory already at the (relevant) maths frontier? Do existing problems lend themselves more to numerical solutions? Are advances mostly happening on narrower topics?

310

Reposted by Mridul K. Thomas

Elena Litchman @elenalitchman.bsky.social · 04/08/2026

Interested in doing a postdoc in plankton ecology, on topics of community assembly, trait-based ecological resilience, stressor interactions, harmful algal blooms and others? Contact me to discuss developing a proposal for MSU's EEB Presidential Postdoc Fellowship (Nov 1 application deadline).

12733

Mridul K. Thomas @mridulkthomas.bsky.social · 30/07/2026

Exactly - I grew up in South India and it was believed was that the free meal program made a big difference for poor students' nutrition. Protein in the form of eggs was an important part of this (controversial in some states, such as West Bengal recently). So this might show up in the height data.

010

Mridul K. Thomas @mridulkthomas.bsky.social · 30/07/2026

but I imagine this could be modelled. And the comparison across states should be informative.

100

Mridul K. Thomas @mridulkthomas.bsky.social · 30/07/2026

Hmmm. I guess it's not going to be a simple discontinuity because there will be a few years of students who get the mid-day meal for different numbers of years. If introduced in 2000 for all classes, then the oldest get for 1 year, the youngest get for 12...

100

Mridul K. Thomas @mridulkthomas.bsky.social · 30/07/2026

I was thinking about year of introduction of the mid-day meal. Height at say age 20 as a function of year, could well show a discontinuity at year of introduction of the mid-day meal scheme if diet is a major cause. Year of introduction also differs between states for some added information.

100

Mridul K. Thomas @mridulkthomas.bsky.social · 30/07/2026

Looking forward to reading this one, Alison!

010

Mridul K. Thomas @mridulkthomas.bsky.social · 30/07/2026

An interesting read! My top guesses were parasite load and diet, especially protein consumption. I'm surprised to see we don't have better data on the latter. I wonder if the mid-day meal schemes - which often include eggs - would allow for regression discontinuities that isolate the diet effect?

100

Mridul K. Thomas @mridulkthomas.bsky.social · 29/07/2026

2026 looks much worse if you look at other countries. Here are France, Germany, Italy, Spain, Poland and the UK. And these plots stop a couple of weeks ago and so miss the big recent fires in France and Spain.

000

Mridul K. Thomas @mridulkthomas.bsky.social · 29/07/2026

I've been baffled by this figure. Why is Europe low when the big European countries show more - sometimes record - levels of fire this year? It turns out this pattern is entirely driven by Russia, which I didn't realise was counted as Europe in these datasets. ourworldindata.org/grapher/cumu...

Screenshot of figure showing cumulative area burnt by wildfires in Russia, by year. 2026 is the lowest since the dataset begins in 2012.

210

Mridul K. Thomas @mridulkthomas.bsky.social · 27/07/2026

Odd wildfire fact: 2026 has seen the least European territory burned of any year in the last 15. Caveat: this runs till a week ago and so misses the big ones of the past few days. And I don't know the source dataset myself, I am relying on OWID. ourworldindata.org/grapher/cumu...

A graph of Cumulative area burnt by wildfires by week in Europe, across several years. Source: https://ourworldindata.org/grapher/cumulative-area-burnt-by-wildfires-by-week?country=~OWID_EUR

110

Mridul K. Thomas @mridulkthomas.bsky.social · 27/07/2026

I'm sure there is a valid use of the term 'digital twin', but I have only ever seen it invoked to make 'model' sound fancier.

040