Anton Akhmerov
@antonakhmerov.org
Condensed matter theorist and a fan of open source and open science. https://antonakhmerov.org/
Reposted by Anton Akhmerov
My personal no-go property of a platform is the login wall: I don't want to demand a platform account from my potential readers.
Twitter had many last drops for different people, but the login wall a couple of years ago was mine. I've been avoiding LinkedIn for the same reason.
Twitter had many last drops for different people, but the login wall a couple of years ago was mine. I've been avoiding LinkedIn for the same reason.
January 7, 2026 at 11:33 AM
My personal no-go property of a platform is the login wall: I don't want to demand a platform account from my potential readers.
Twitter had many last drops for different people, but the login wall a couple of years ago was mine. I've been avoiding LinkedIn for the same reason.
Twitter had many last drops for different people, but the login wall a couple of years ago was mine. I've been avoiding LinkedIn for the same reason.
Okie dokie, in arxiv.org/abs/2601.00784 we invented a format for the AI disclosure statement.
January 5, 2026 at 7:59 PM
Okie dokie, in arxiv.org/abs/2601.00784 we invented a format for the AI disclosure statement.
Reposted by Anton Akhmerov
December 31, 2025 at 7:32 PM
Educators wonder how to teach being critical about LLM outputs.
What if we take a mid-range model, raise the temperature until its success rate on some standard task (think computing an integral) is 50–70% and let students solve the task with this LLM.
Teaches validating LLM in real conditions.
What if we take a mid-range model, raise the temperature until its success rate on some standard task (think computing an integral) is 50–70% and let students solve the task with this LLM.
Teaches validating LLM in real conditions.
December 19, 2025 at 12:31 PM
Educators wonder how to teach being critical about LLM outputs.
What if we take a mid-range model, raise the temperature until its success rate on some standard task (think computing an integral) is 50–70% and let students solve the task with this LLM.
Teaches validating LLM in real conditions.
What if we take a mid-range model, raise the temperature until its success rate on some standard task (think computing an integral) is 50–70% and let students solve the task with this LLM.
Teaches validating LLM in real conditions.
Thanks, but no thanks.
December 18, 2025 at 8:12 PM
Thanks, but no thanks.
Psst kids, want to buy some research projects?
December 2, 2025 at 12:13 AM
Psst kids, want to buy some research projects?
A lesson I learned from software development: changes don't need to be perfect, and incremental improvements are often good enough. For example, TU Delft changed its motto from "Challenge the future" (!?) to "Impact for a better society". Perfect? Maybe not. Better? Very much so! :)
November 28, 2025 at 10:36 AM
A lesson I learned from software development: changes don't need to be perfect, and incremental improvements are often good enough. For example, TU Delft changed its motto from "Challenge the future" (!?) to "Impact for a better society". Perfect? Maybe not. Better? Very much so! :)
What are the best examples of AI usage disclosure forms in coding and academia? I've seen a handful of "tell me everything", which seems fairly impractical.
I am by now convinced that disclosing genAI usage should become a norm in academia and education. I was initially skeptical—it's just another tool—but I am now convinced that the extra transparency is helpful. After all, I am all for sharing code, what's different here?
November 27, 2025 at 2:02 PM
What are the best examples of AI usage disclosure forms in coding and academia? I've seen a handful of "tell me everything", which seems fairly impractical.
I am by now convinced that disclosing genAI usage should become a norm in academia and education. I was initially skeptical—it's just another tool—but I am now convinced that the extra transparency is helpful. After all, I am all for sharing code, what's different here?
November 25, 2025 at 10:16 PM
I am by now convinced that disclosing genAI usage should become a norm in academia and education. I was initially skeptical—it's just another tool—but I am now convinced that the extra transparency is helpful. After all, I am all for sharing code, what's different here?
Things language models apparently can't do yet: optimize a website. This is chatgpt.com eating half a gigabyte of RAM and 100% CPU to respond to a text-only query.
November 13, 2025 at 11:51 AM
Things language models apparently can't do yet: optimize a website. This is chatgpt.com eating half a gigabyte of RAM and 100% CPU to respond to a text-only query.
I want students to collectively decide on a formula sheet for an exam. How to organize collaborative work of 100-200 people on one document of fixed size?
November 13, 2025 at 12:40 AM
I want students to collectively decide on a formula sheet for an exam. How to organize collaborative work of 100-200 people on one document of fixed size?
There is a persistent view in academia that you learn only by living through an experience. "My students implement everything from scratch so that they understand how code works." "If you don't write all your papers, you never learn." Active learning is great, but applying it mindlessly isn't OK.
October 29, 2025 at 2:18 PM
There is a persistent view in academia that you learn only by living through an experience. "My students implement everything from scratch so that they understand how code works." "If you don't write all your papers, you never learn." Active learning is great, but applying it mindlessly isn't OK.
I've been using @mattermost.bsky.social for almost a decade (since December 2015) for my group, some courses I teach. I ended up providing a chat for a few other research groups. Unfortunately with Mattermost Inc. backpedaling their open source stance, I'll find something else.
October 22, 2025 at 5:50 PM
I've been using @mattermost.bsky.social for almost a decade (since December 2015) for my group, some courses I teach. I ended up providing a chat for a few other research groups. Unfortunately with Mattermost Inc. backpedaling their open source stance, I'll find something else.
And this is how I learned about jupyter-ssh-proxy. Seems handy, and a lot less headache to set up than an actual ssh server into the hub. ✨
Thanks to Andy from OpenScapes for fixing a docs issue in jupyter-sshd-proxy that helps users connect to JupyterHub via SSH. This is the cycle we want to see - communities use open source tools, find issues, and contribute fixes that benefit everyone 🎉
2i2c.org/blog/2025/op...
2i2c.org/blog/2025/op...
A helpful contribution to our JupyterHub SSH README from OpenScapes | 2i2c
We love when collaborators contribute back to the tools we maintain! Andy Teucher from OpenScapes recently fixed a documentation issue in jupyter-sshd-proxy that benefits everyone using the tool. jupyter-sshd-proxy...
2i2c.org
October 21, 2025 at 3:37 PM
And this is how I learned about jupyter-ssh-proxy. Seems handy, and a lot less headache to set up than an actual ssh server into the hub. ✨
Surgeon general warning: quantum computing is not close to being useful.
October 15, 2025 at 10:36 AM
Surgeon general warning: quantum computing is not close to being useful.
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.
try:
data = get(nonexistent_url).json()
except Exception:
data = placeholder_data()
I think excessive defensiveness is the biggest pitfall of current tools.
October 11, 2025 at 10:45 AM
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.
try:
data = get(nonexistent_url).json()
except Exception:
data = placeholder_data()
I think excessive defensiveness is the biggest pitfall of current tools.
I don't think I'd manage that long scholar.google.com/scholar?q=%2...
scholar.google.com
October 7, 2025 at 10:26 PM
I don't think I'd manage that long scholar.google.com/scholar?q=%2...
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)
- 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)
October 7, 2025 at 11:57 AM
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)
- 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)
ChatGPT is feeling mighty enabled in the last couple of days.
September 28, 2025 at 11:04 AM
ChatGPT is feeling mighty enabled in the last couple of days.
We needed to demonstrate how different symmetries look, and I think the result is neat.
September 25, 2025 at 1:12 PM
We needed to demonstrate how different symmetries look, and I think the result is neat.
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
September 10, 2025 at 9:50 PM
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
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
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
August 31, 2025 at 11:49 AM
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
Your input can shape it! What simulations should it handle?
Submit ideas: forms.gle/z6BpHNDKJfRN... #opensource #quantumtransport
How about the editors publish a note "we screwed up and we shouldn't have published the paper" instead of retracting?
Science is retracting the December 2010 Research Article, “A bacterium that can grow by using arsenic instead of phosphorus.” (THREAD 🧵) scim.ag/4lGQ9g7
August 3, 2025 at 1:08 PM
How about the editors publish a note "we screwed up and we shouldn't have published the paper" instead of retracting?