iiinova — where your digital self keeps evolving

*Adding to the manuscript-writing skill*: omit articles gratuitously. More seriously, I get where you're coming from, but there's a leap from valuing thought to valuing errors.

010

Anton Akhmerov @antonakhmerov.org · 02/10/2026

To make lean more accessible to physicists I have formalized ≈.

010

Anton Akhmerov @antonakhmerov.org · 01/10/2026

It's a stopgap, but it's still a good call.

000

Anton Akhmerov @antonakhmerov.org · 27/09/2026

I gathered an atlas of literature on Andreev states and the corresponding collaboration network. For now it lives here antonakhmerov.org/misc/andreev... Check it out for the citation hedgehog and to find your friends.

Interactive citation-network visualization of 8,067 papers. Papers are positioned by publication year and annual citation count on a logarithmic vertical scale; with publication year on the horizontal axis. Point color indicates physics subfield and point size reflects citation count. Faint lines connect citing and cited papers, producing a dense fan-like “hedgehog” pattern.

Collaboration network centered on Anton Akhmerov. Nodes represent researchers, with larger nodes indicating stronger collaboration links; lines connect coauthors. Prominent collaborators include Michael Wimmer, C. W. J. Beenakker, Attila Geresdi, Bernard van Heck, Valla Fatemi, Bas Nijholt, David J. van Woerkom, Chun-Xiao Liu, Ion Cosma Fulga, and others.

121

Reposted by Anton Akhmerov

Aaron Roth @aaroth.bsky.social · 24/09/2026

AI tools are very useful for mathematical research. But at present, producing good work requires a labor intensive step that I have been calling "deslopping" - making the thing readable. Please do not skip this step; even if the theorem is great an unreadable paper is worthless.

4273

Anton Akhmerov @antonakhmerov.org · 23/09/2026

Of course the best humans will know what is the right direction.

000

Anton Akhmerov @antonakhmerov.org · 20/09/2026

The best strategy as this is coming is, of course, to spam everyone in and out of your your network. Thankfully with AI it's really easy to do in a way that will be progressively harder to detect. /s

010

Anton Akhmerov @antonakhmerov.org · 18/09/2026

I updated the Fermi surface viewer of elemental metals at solidstate.quantumtinkerer.tudelft.nl/fermi_surfac... by using the Materials Project database and a custom 3D viewer.

Interactive “Fermi surfaces” visualization for technetium (Tc, atomic number 43), showing five color-coded electronic bands as a sixfold-symmetric 3D Fermi-surface plot. The interface includes element selection, band toggles, rotation/zoom controls, and an “About the data” section.

020

Anton Akhmerov @antonakhmerov.org · 14/09/2026

Field-defining paper: Blonder, Tinkham, Klapwijk.

A citation network showing a paper cited by a majority of papers in the field.

020

Anton Akhmerov @antonakhmerov.org · 13/09/2026

The statement on AI in mathematics terrytao.wordpress.com/2026/09/11/a... is fair. One addition. In my community (condensed matter, not math) the misalignment of community itself is already comparably severe: there is little incentive to focus on clarity and coherence rather than paper production.

terrytao.wordpress.com

A Severe Misalignment of AI in Mathematics

I am proud to be among the list of 25 initial signatories — all Fields Medallists — to the declaration below, which grew out of discussions between ourselves over the last week. We have…

030

Anton Akhmerov @antonakhmerov.org · 03/09/2026

Here is monthly publication rate of a colleague. Recent papers (after the dashed line) acknowledge Claude usage, and the publication rate goes 3x. The paper quality stays roughly constant.

Line chart titled “Submission acceleration.” Cumulative arXiv submissions from January 2024 to September 2026 rise roughly linearly at first, corresponding to about 15 submissions per year. Around June 2026, the slope increases sharply; a fitted rate changes to about 51 submissions per year. The underlying data are shown as monthly cumulative totals, with a dashed piecewise-linear rate fit and a vertical marker at approximately June 2026.

030

Anton Akhmerov @antonakhmerov.org · 29/08/2026

If it's really "lean", why does it take 7837 lines of code and counting?

010

Anton Akhmerov @antonakhmerov.org · 16/08/2026

Pretty much this. If you can also not caring about getting "enough credit" is also helpful.

010

Anton Akhmerov @antonakhmerov.org · 14/08/2026

Coming back to the profile: shouldn't researcher verification be automatic from a linked orcid? About adding papers to the profile: consider reading data from crossref API: browser can do this, and it saves typing. See here for example antonakhmerov.org/misc/reasonableda…

100

Anton Akhmerov @antonakhmerov.org · 12/08/2026

Yup! At least not clearly broken 😅 Thanks!

010

Reposted by Anton Akhmerov

Martin Donath @squidfunk.com · 12/08/2026

I've changed my mind on AI-generated bug reports. They got genuinely better and helpful. Open Source maintainers: do you share prompts with your users to help an LLM gather data and debug before filing an issue?

161

Anton Akhmerov @antonakhmerov.org · 12/08/2026

Another confusing bit is that I can't add OpenAlex ID because I'm not verified... except I am verified.

100

Anton Akhmerov @antonakhmerov.org · 12/08/2026

ORCID succeeded, but it didn't suffice to verify me as a researcher because I lack 3 recognized works it seems? (I don't lack 3 works haha.)

210

Anton Akhmerov @antonakhmerov.org · 12/08/2026

Thanks, that part is fixed! I'm moving on to the next guinea pig step: somehow Lea is very selective in what venues are recognized. It accepts Library of the Hungarian Academy of Sciences, but less established outlets like American Physics Society, Nature, Science, Institute of Physics etc.

110

Anton Akhmerov @antonakhmerov.org · 12/08/2026

Ah, that sounds like a sensible design, thanks. Indeed, that's not a given from OAuth.

010

Anton Akhmerov @antonakhmerov.org · 12/08/2026

Trying to connect ORCID (link orcid from profile -> link -> error). Source code: I am asking because granting full account control to a closed source side project of a few researchers is a big trust investment. Source is one extra trust mechanism.

150

Anton Akhmerov @antonakhmerov.org · 12/08/2026

Almost, but not yet completely. "For now" meaning you plan to release the source code?

110

Anton Akhmerov @antonakhmerov.org · 12/08/2026

Interesting! I'm confused: adding ORCID only available to verified researchers? Should it not be the other way around? Also is the source code available? It was for a short time but now it isn't, is that right?

110

Anton Akhmerov @antonakhmerov.org · 11/08/2026

Both Anthropic and OpenAI now oppose the Leiden declaration on AI in mathematics, signing mathematical results “GPT” or “Claude”. Why not sign them “NVIDIA”? Or “silicon”, for its abundance and band gap. I agree with the mockery of authorship; I just don’t want a corporate logo on it.

020

Anton Akhmerov @antonakhmerov.org · 04/08/2026

Because you already tried ClaudeGPT Max Pro and it failed 😆

010

Anton Akhmerov @antonakhmerov.org · 04/08/2026

I think it's straightforward to value the result according to its impact and the author contribution according to the effort invested. The latter is neither zero nor outstanding. And the most expensive ChatGPT is same scale as Mathematica license.

000

Anton Akhmerov @antonakhmerov.org · 31/07/2026

Anthropic: other labs and open models can't be trusted, we're the ones that care about AI safety. Also Anthropic: forgets to restrict internet access in a model security evaluation.

030

Anton Akhmerov @antonakhmerov.org · 31/07/2026

I imagine changing message history invalidates cache and therefore is expensive, and deleting fresh but useless stuff sounds like gaslighting the model: hard to train for, hard to apply, mistakes are costly.

010

Anton Akhmerov @antonakhmerov.org · 30/07/2026

I think the distinction is useful to keep though, and therefore I propose to introduce "tradcoding". Look, I tradcoded an app!

020

Anton Akhmerov @antonakhmerov.org · 29/07/2026

Fight fire with fire 😎 chatgpt.com/share/6a6a63... To be honest I'm surprised to see hallucinated references this time. I thought any modern agent would do a better job.

030

Anton Akhmerov @antonakhmerov.org · 29/07/2026

It's a pity it isn't. Although arXiv's existing policies allow them to do pretty much anything, so it wouldn't be against what is published if they followed the announced 1 year ban. With around 10 out of 26 citations hallucinated, this paper would be a prime candidate.

150

Reposted by Anton Akhmerov

Victor V. Albert @vva.bsky.social · 22/07/2026

We're at a strange, specific moment. Takes range from "publications are dead by fall" to "I tried LLMs in November, they were bad." If you have an LLM-assisted paper, take full responsibility as usual. Do everything: arXiv, GitHub, website, and integrate with databases if appropriate.

0183

Anton Akhmerov @antonakhmerov.org · 18/07/2026

I just think research cost was never a reliable proxy for usefulness of a result, and so I'd rather not use it as a criterion. Of course most of the works in the category 'agent can solve in 2h' aren't useful, but that's more so because most works all around aren't useful.

000

Anton Akhmerov @antonakhmerov.org · 17/07/2026

Depends on how much you want to bother and value the work. Open sourcing is enough for most and a great default. Papers for stuff that you expect will be reused (downscale the expectations though). Patents idk.

010

Anton Akhmerov @antonakhmerov.org · 17/07/2026

Not this argument literally, because a lot of things that "Claude can do in a couple of hours" are worth to be published. Effort cost is not a relevant scale.

100

Anton Akhmerov @antonakhmerov.org · 07/07/2026

At least this one allows unreasonable requests too.

010

Anton Akhmerov @antonakhmerov.org · 30/06/2026

Now I made a physics demo of x-ray crystallography. The reciprocal space and how X-ray works took me a while to understand, and I think this captures the idea well. antonakhmerov.org/misc/xray.html

Interactive X-ray scattering simulation with two main areas. The left side is a crystal editor showing unit-cell projections and a 3D crystal view. The right side shows reciprocal lattice points, a translucent Ewald sphere, an X-ray beam, a detector screen, and amber detector spots where scattering rays intersect the screen. Controls let the user edit atoms and lattice vectors, rotate the crystal, switch powder and top-view modes, and adjust the wavelength.

020

Reposted by Anton Akhmerov

Ran Blekhman @blekhman.bsky.social · 27/06/2026

I wanted to like QED's new 1% preprint ranking. I really did. But the more I looked at the data, the more uncomfortable it made me. Here is my full peer review. open.substack.com/pub/blekhman...

open.substack.com

My Peer Review of The 1%

I wanted to like QED's new 1% ranking. I don't.

1720888

Reposted by Anton Akhmerov

Ted Underwood @tedunderwood.com · 26/06/2026

I think when ppl hear LLMs are good at code some think “whew! I always knew they should be restricted to narrowly technical uses (maybe protein folding?)“ but the thing that makes agentic application different imo is not the code—but the way it leverages context & augments human memory and patience

6727

Anton Akhmerov @antonakhmerov.org · 23/06/2026

Another interactive demo showing the propagation of a wave packet in a tight-binding dispersion relation: antonakhmerov.org/misc/wave_pa...

False-color plot of a localized tight-binding wave packet driven by an electric field E. The packet follows a curved trajectory, illustrating Bloch oscillations, with a wavy path marking its motion through the lattice.

010

Anton Akhmerov @antonakhmerov.org · 12/06/2026

Pretty sure this is what needs to happen in NL. Open weight models aren't competitive at high end of complexity, but they should be quite enough for education.

010

Anton Akhmerov @antonakhmerov.org · 02/06/2026

I made an interactive explainer of carbon nanotube dispersion relations for my class today: antonakhmerov.org/files/carbon..., and I find it amazing and very useful. Naturally it is LLM-made. Making the equivalent by hand requires expertise and time investment only available to extremely few.

Interactive educational diagram showing how a carbon nanotube is formed from graphene. The screen displays a graphene sheet with a highlighted wrapping vector, a 3D nanotube inset, allowed quantized k-vector cuts through repeated graphene Brillouin zones, and the resulting one-dimensional folded band structure.

020

Anton Akhmerov @antonakhmerov.org · 25/05/2026

A zoom ladder perhaps? Vaguely something like this

$A zoom-ladder diagram showing how a 500 mL water estimate shrinks through successive multiplicative corrections. Each row uses the same-width horizontal bar, with a colored fraction of the current bar connected by diagonal zoom lines to the full bar on the next row. The cumulative amounts shown are 500 mL, 333 mL, 222 mL, 37 mL, and finally 1.9–9.3 mL. The correction factors are shown on the right: divide by 1.5 for prompt and response length, divide by 1.5 for longer-output extrapolation, divide by 6 for active Mixture-of-Experts parameters, and divide by 4–20 for optimized production serving. The final result is labeled as roughly a teaspoon of water rather than a bottle.$

040

Anton Akhmerov @antonakhmerov.org · 25/05/2026

Nice read! The bottle visualization is tricky because the space allocated to each factor depends on its place in the multiplication chain but looks like its relative importance. One could consider arranging the factors by "distance from the task", but it's still obscures their relative impact.

120

Anton Akhmerov @antonakhmerov.org · 24/05/2026

Some time in the future a visible and published theoretical condensed matter paper will be discovered to be wrong due to being an AI hallucination. I give it about a year.

160

Anton Akhmerov @antonakhmerov.org · 12/05/2026

No. Because the risks to whistleblowers aren't due to technological limitations.

000

Anton Akhmerov @antonakhmerov.org · 10/05/2026

Curiously, working with other people has the same failure modes.

000

Anton Akhmerov @antonakhmerov.org · 08/05/2026

I'm rooting for no more papers. But also it can lead to both: more and better.

030