Lightnews — Scholar-powered news

Anton Akhmerov @antonakhmerov.org · 1d

Are there reliable ways to ensure that coding agents don't use ridiculously defensive coding? I often run into

try:
data = get(nonexistent_url).json()
except Exception:
data = placeholder_data()

I think excessive defensiveness is the biggest pitfall of current tools.

2

Anton Akhmerov @antonakhmerov.org · 3d

Firstly, amazing! I've seen that builds against pre-release version started much earlier, and that's awesome!

NB, there's draft text remaining in the post: "With yesterday's release of Python 3.14 (insert release link)"

1

Anton Akhmerov @antonakhmerov.org · 5d

I don't think I'd manage that long scholar.google.com/scholar?q=%2...

scholar.google.com

Anton Akhmerov @antonakhmerov.org · 5d

I didn't realize this before, but chatGPT logo has deep symbolism to it:

- It's cool
- It's simple
- It's visually appealing
- If you try to study it carefully, things don't quite make sense nor match

(Source: the SVG form the app, so this is *the* logo)

2

Anton Akhmerov @antonakhmerov.org · 11d

A part of an email showing a lot of inhomogeneously collapsed and indented text hidden behind ellipses, and ending with "Dear Anton,"

1 1

Anton Akhmerov @antonakhmerov.org · 14d

ChatGPT is feeling mighty enabled in the last couple of days.

A screenshot from ChatGPT saying "If you want me to, I can (A) open an issue in the emscripten-forge/recipes repo requesting wasm64 support, or (B) monitor upstream Emscripten / browser runtime status and report back on milestones. Tell me which action to take or I will proceed with (A)."

A screenshot from ChatGPT saying "If you want, I can dig deeper (e.g. search internal Google filings in non-public repositories, or see if there are conference talks) to check whether there's non-obvious public disclosure.

1

Anton Akhmerov @antonakhmerov.org · 17d

We needed to demonstrate how different symmetries look, and I think the result is neat.

An image with 4 panels titled "none"/"mirror"/"magnetic_translation"/"glide", showing a random signal which is symmetrized according to the corresponding symmetry.

1

Anton Akhmerov @antonakhmerov.org · 28d

The decision seems to strongly depend on the review quality, difficulty of prompt design, and the accuracy of the genAI result.

- How likely is a researcher to be able to generate an appropriate high quality prompt?
- How would one establish validity of the genAI output?

Anton Akhmerov @antonakhmerov.org · Sep 10

I made an app for raising hands. Make a floor plan of your classroom, define tables, share QR codes with students, and you're set. They raise hands in a queue, you and/or your fellow TAs get notifications. Try it out here (if it's your kind of thing) handraiser.quantumtinkerer.tudelft.nl

Hand Raiser

handraiser.quantumtinkerer.tudelft.nl

1

Anton Akhmerov @antonakhmerov.org · Aug 31

Even the best general purpose LLMs struggle with niche knowledge. This is why I think such a tool can make a big difference. At the same time, the available open AI tools are already impressive, and I expect that they will prove amazingly useful.

1

Anton Akhmerov @antonakhmerov.org · Aug 31

I’m working on Kwantsplain, an AI tool that turns natural language descriptions into Kwant simulations.

Your input can shape it! What simulations should it handle?
Submit ideas: forms.gle/z6BpHNDKJfRN... #opensource #quantumtransport

Kwantsplain query

I’m developing Kwantsplain, an open-source AI assistant that turns natural language queries into Kwant simulations. I’d love your input: what would you ask from this tool? Your feedback will help guid...

forms.gle

1 1 6

Anton Akhmerov @antonakhmerov.org · Aug 25

I have to ask: did the agent come up with this blackboard? Seems only tangentially related to arachnophobia, but definitely anxiety-inducing.

A blackboard with text "CORPSE #1 #2 FORMER CADAVERIC DRAIN BODY F..."

1

Anton Akhmerov @antonakhmerov.org · Aug 21

Hour 3+: ok you've no idea, let me show you how it is actually done.

Anton Akhmerov @antonakhmerov.org · Aug 3

How about the editors publish a note "we screwed up and we shouldn't have published the paper" instead of retracting?

Science Magazine @science.org · Jul 24

Science is retracting the December 2010 Research Article, “A bacterium that can grow by using arsenic instead of phosphorus.” (THREAD 🧵) scim.ag/4lGQ9g7

“If the editors determine that a paper’s reported experiments do not support its key conclusions, even if no fraud or manipulation occurred, a Retraction is considered appropriate.” H. Holden Thorp, Editor-in-chief, Science journals

Reposted by Anton Akhmerov

Daphne Cornelisse @daphne-cornelisse.bsky.social · Apr 19

Overnight runs are the overnight oats of research — prep, forget, and rewarding by morning

4 13

Anton Akhmerov @antonakhmerov.org · Jul 17

To add gravitas and style to your statements, end sentences with "unless polynomial hierarchy collapses".

- Want to grab lunch tomorrow?
- Sounds awesome, I'm in! Unless polynomial hierarchy collapses.

Anton Akhmerov @antonakhmerov.org · Jul 6

I've seen plenty of #arXiv preprints that look like they've been authored years after the actual upload.

1. Authors: remember to replace \date{\today} so that you don't look like you're a decade late with your work.
2. Wouldn't it be amazing if arXiv set the OS date to upload date when compiling?

3

Anton Akhmerov @antonakhmerov.org · Jun 26

Let's collect rules!

What makes a good paper title? What are patterns to avoid, and what approaches work well?

As an exercise and to motivate the question, go through fresh arXiv and see how many titles you like.

1 1

Anton Akhmerov @antonakhmerov.org · Jun 24

I would even go as far as to recommend adding *.ipynb to .gitignore and use jupytext with .py or .md for interactive work. It is possible to have clean notebooks in git, but the extra complexity is a challenge for many people.

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · Jun 23

Friends, please strip your output from jupyter notebooks before committing. You can just add github.com/kynan/nbstri... as a pre-commit hook. Having readable diffs is worth it for 30 seconds of effort

GitHub - kynan/nbstripout: strip output from Jupyter and IPython notebooks

strip output from Jupyter and IPython notebooks. Contribute to kynan/nbstripout development by creating an account on GitHub.

github.com

1 2

Anton Akhmerov @antonakhmerov.org · Jun 15

An important caveat here is that usually we're interested in angular momentum because it is conserved. On a torus the operator L, defined in the preprint does not commute with the Hamiltonian at the boundary, where q is discontinuous.

2

Anton Akhmerov @antonakhmerov.org · Jun 11

I haven't done this in years. I think I'd put it online instead.

1

Anton Akhmerov @antonakhmerov.org · Jun 11

By the way, in order to do 2nd quantized perturbation theory we needed to figure out how to simplify operator expressions, which is quite useful by itself. Check out our docs here: pymablock.readthedocs.io/en/latest/se... if you need to do symbolic 2nd quantized computations.

1 3

Anton Akhmerov @antonakhmerov.org · Jun 10

Two last remaining uses for a USB stick:

- Install Linux.
- Share your slides because the conference organizers asked.

1 2

Anton Akhmerov @antonakhmerov.org · Jun 6

Me, a year ago: "Perturbation theory is always the same, second quantization is just filling in the gaps."

6000 lines of code and docs later: Pymablock v2.2.0 released🎉 pymablock.readthedocs.io/en/v2.2.0/

We have bosons, spins, fermions, and Floquet, and I'm quite happy about it.

Pymablock — pymablock 2.2.0 documentation

pymablock.readthedocs.io

1 9

Anton Akhmerov @antonakhmerov.org · May 19

Learning #manim library to make video lectures for the solid state physics course. So far so good! Any useful links or resources?

2