anotherpaulg

Clanki is a great name.

I’ve had a lot of success using AI to generate memorable images to show alongside vocabulary cards.

If you have CAT5/6 between the locations, USB-over-CAT5/6 adapters are inexpensive and work quite well. I've used OREI's successfully over >60m. But there are many options, with varying limits on cable length.

https://www.orei.com/products/usb-over-ethernet-extender-upt...

FWIW, I have had great success asking AI coding tools to generate/edit tikz code. As with all AI coding, it helps to steer the agent to structure the code sensibly, etc. But frontier models seem to know how to write tikz.

Parallel Lives: A Local-Realistic Interpretation of “Nonlocal” Boxes by Gilles Brassard and Paul Raymond-Robic.

Abstract: We carry out a thought experiment in an imaginary world. Our world is both local and realistic, yet it violates a Bell inequality more than does quantum theory. This serves to debunk the myth that equates local realism with local hidden variables in the simplest possible manner. Along the way, we reinterpret the celebrated 1935 argument of Einstein, Podolsky and Rosen, and come to the conclusion that they were right in their questioning the completeness of the Copenhagen version of quantum theory, provided one believes in a local-realistic universe. Throughout our journey, we strive to explain our views from first principles, without expecting mathematical sophistication nor specialized prior knowledge from the reader.

I’ve been building quantum photonics experiments. Repeating the Bell inequality tests that won the 2022 Nobel, quantum erasers, etc.

I just published a fun interactive 3D demo of SPDC, one of the most common and accessible ways to create entangled pairs of photons. I'm hoping to publish a series of articles on other cool learnings about doing quantum photonics in the lab.

https://paulg.info/2026/04/10/spdc/

I’ve been building quantum photonics experiments. Repeating the Bell inequality tests that won the 2022 Nobel, quantum erasers, etc.

Probably the coolest part has been automating the optomechanical equipment and optimizing physical experiments with Bayesian optimization. Similar to hyperparameter tuning in ML, but with lasers.

Also, Thorlabs sells some really fun toys.

Recorded 10 February 2026. Terence Tao of the University of California, Los Angeles, presents "Machine assistance and the future of research mathematics" at IPAM's AI for Science Kickoff.

I appreciate the ability to rapidly capture a note/thought/todo without friction and context switching.

I solved this problem with a twilio sms number. When I send a text to it, the content gets prepended to my obsidian todo.md file. This was easy to arrange with a few lines of Python glue.

iOS makes it easy to text or share to sms from almost any context.

The Earth is generally expected to spin more slowly over time, due to tidal friction. But it has been spinning faster and faster since the 1960s. As shown in the figure in the wikipedia article [0].

I have read numerous explanations, but haven't found a really authoritative discussion.

[0] https://en.wikipedia.org/wiki/Leap_second#Rationale

He just discussed this on Robinson’s podcast, in conversation with Tim Maudlin.

Zurek published a book about Quantum Darwinism about a year ago. It is a text book, not a popular treatment, but it is quite a good read.

https://www.cambridge.org/core/books/decoherence-and-quantum...

Aider actually prompts the model to say if it needs to see additional files. Whenever the model mentions file names, aider asks the user if they should be added to context.

As well, any files or symbols mentioned by the model are noted. They influence the repomap ranking algorithm, so subsequent requests have even more relevant repository context.

This is designed as a sort of implicit search and ranking flow. The blog article doesn’t get into any of this detail, but much of this has been around and working well since 2023.

Once you’ve confirmed when your target is visible, this site provides a handy forecast of atmospheric viewing conditions.

https://www.cleardarksky.com/csk/

I’m building a quantum photonics experiment that is a variation of the quantum eraser.

One aspect that HN may find interesting is my use of Bayesian optimization to control and perfect key experimental settings. About a dozen of the wave plates and other optical components are motorized and under computer control.

Given a goal metric like "maximally entangle the photon pairs" the optimizer will run the experiment 50-100 times, tweaking the angles of various optics and collecting data. Ultimately it will learn to maximize the given cost function.

This sort of thing is commonly done with tools like Optuna during NN/LLM training to optimize hyper-parameters, but seems less common in physics especially quantum photonics. I'm using a great tool called M-loop to drive the optimization, which was originally developed for creating Bose-Einstein condensates.

https://github.com/michaelhush/M-LOOP

I like the unconventional approach. A few minutes with GPT raises two issues:

1. We've raised CO2 from 280ppm to 420ppm, about a 50% increase. To dilute it back down would require 50% more total atmosphere. This would also raise the surface air pressure 1.5x.

2. How much heat is trapped is related to the absolute amount of CO2 in the atmosphere, not the fraction. So the diluted atmosphere would retain just as much heat.

It would be interesting to see how hard it would be to walk these models towards general relativity and quantum mechanics.

Einstein’s paper “On the Electrodynamics of Moving Bodies” with special relativity was published in 1905. His work on general relativity was published 10 years later in 1915. The earliest knowledge cuttoff of these models is 1913, in between the relativity papers.

The knowledge cutoffs are also right in the middle of the early days of quantum mechanics, as various idiosyncratic experimental results were being rolled up into a coherent theory.

I’ve relied heavily on seamless capture for a couple of decades, ala Getting Things Done.

My solution is a twilio text number that automatically inserts any texts it receives into the top of my todo.md file. Previously todo.org, until about a year ago.

iOS has ubiquitous support to quickly share to SMS from any/everywhere. It’s easy to send a text to this contact from a Home Screen shortcut, but also also from the share sheet in most every app.

I regularly use LLM-as-OCR and find it really helpful to:

1. Minimize the number of PDF pages per context/call. Don't dump a giant document set into one request. Break them into the smallest coherent chunks.

2. In a clean context, re-send the page and the extracted target content and ask the model to proofread/double-check the extracted data.

3. Repeat the extraction and/or the proofreading steps with a different model and compare the results.

4. Iterate until the proofreadings pass without altering the data, or flag proofreading failures for stronger models or human intervention.

I’ve been building proficiency with quantum optics equipment. Repeating classic quantum entanglement experiments like the quantum eraser [0] and violating the CHSH inequality (which won the 2022 Nobel). I’m working towards a novel quantum eraser variant.

[0] https://github.com/paul-gauthier/entangled-pair-quantum-eras...

I really like LLM+sympy for math. I have the LLM write me a sympy program, so I can trust that the symbolic manipulation is done correctly.

The code is also a useful artifact that can be iteratively edited and improved by both the human and LLM, with git history, etc. Running and passing tests/assertions helps to build and maintain confidence that the math remains correct.

I use helper functions to easily render from the sympy code to latex, etc.

A lot of the math behind this quantum eraser experiment was done this way.

https://github.com/paul-gauthier/entangled-pair-quantum-eras...

Below are some great videos on the physics and practicalities of single mode fiber. They are Thorlabs videos, so are slanted more towards the use of SMF in a laser lab rather than a telecom setting. They reference a lot of the theory, but also provide a good intuition about how and why SMF works so well.

https://youtu.be/FbOXRuBQt_U

https://youtu.be/HvJeXakc8Kc

If you like this then you will probably enjoy the book How to Invent Everything: A Survival Guide for the Stranded Time Traveler by cartoonist and computer-scientist Ryan North.

https://www.howtoinventeverything.com/

Scott Aaronson has an interesting take on the 2018 paper being discussed in the article:

https://scottaaronson.blog/?p=3975

Ya, you would think so. But when you connect to it and sshd tries to fork a process to handle the session… you get a privileges error.

Looks very interesting. I was hoping it would solve a problem I’ve had recently:

I want to ssh into a windows box that I only have a normal user account on. So I can’t (and don’t want to) change any admin settings or install anything as admin.

All the obvious approaches hit roadblocks.

Seems like this tool solves the opposite problem: sshing out from a minimally privledged environment.

Aider has had support for Python and shell scripting [0] for a long time. I made a screencast [1] recently that included ad-hoc bash scripting aider as part of the effort to add support for 130 new programming languages. It may give a flavor for how powerful this approach can be.

[0] https://aider.chat/docs/scripting.html

[1] https://aider.chat/docs/recordings/tree-sitter-language-pack...

I wish Anki could dynamically adjust the new cards introduced each day, based on whether you get them right or wrong on first viewing.

Getting 10 new cards that I happen to already know is easy. That's very different than getting 10 new cards that I get wrong, and therefore need to memorize.

I'd love to have 2 settings like:

  - Max number of new card fails, say 4
  - Max number of new cards, say 20

Fancier versions could use a decaying average of new card fails from the last few days. Those recently memorized cards are more challenging during review.

Fun project with a nice compact code base. Agreed that ad-hoc scripting agents can be very powerful.

Aider has had support for scripting [0] in python or via the command line for a long time.

I made a screencast [1] recently that included ad-hoc bash scripting aider as part of the effort to add support for 130 new programming languages. It may give a flavor for how powerful this approach can be.

[0] https://aider.chat/docs/scripting.html

[1] https://aider.chat/docs/recordings/tree-sitter-language-pack...

I've long been fascinated by AI's ability to do the reverse: generate photos with lots of highly relevant content when the prompt includes a location. Terrain, plants, buildings, landmarks, coastlines and lots of details are included.

Here's an example [0] for "Riding e-scooters along the waterfront in Auckland". The iconic spire is correctly included, but so are many small details about the waterfront.

I've been meaning to harness this into a very-low-bandwidth image compression system. Where you take a photo and crunch it to an absurdly low resolution that includes EXIF data with GPS, date/time. You then reconstruct the fine details with AI.

Most photos are taken where lots of photos are taken, so the models have probably been appropriately trained.

[0] https://chatgpt.com/share/680d0008-54a0-8012-91b7-6b1794f485...

In aider, instead of “ultrathink” you would say:

  /thinking-tokens 32k

  /think 32k