occamschainsaw

There’s a fascinating story about S Chandrasekhar (of Chandrasekhar limit fame) driving 100 miles to teach him every week. Teaching two students, the professor got a Nobel prize and the two students got a Nobel prize.

“ One story in particular illustrates Chandrasekhar's devotion to his science and his students. In the 1940s, while he was based at the University's Yerkes Observatory in Williams Bay, Wis., he drove more than 100 miles round-trip each week to teach a class of just two registered students. Any concern about the cost-effectiveness of such a commitment was erased in 1957, when the entire class -- T.D. Lee and C.N. Yang -- won the Nobel Prize in physics.”

Source: https://chronicle.uchicago.edu/951012/chandra.shtml

There’s already some work looking into this[1]. The authors add invisible prompts in papers/grants to embed watermarks in reviews and then show that they can detect LLM generated reviews with reasonable accuracy (more than chance, but there’s no 100% detection yet).

[1] Rao et al., Detecting LLM-Generated Peer Reviews https://arxiv.org/pdf/2503.15772

Even o3 can't figure out the naming scheme. When asked to generate names that would be easier to use by non-technical people, it assumed o3 and o4 are the smallest models:

``` Below is one straightforward, user-friendly approach you could adopt. It keeps two dimensions only—generation and tier—and reserves an optional “optimisation” suffix for special-purpose variants (e.g. vision, coding, long-context).

⸻

1. Core conventions

Element Purpose Example values Generation Major architectural release. Keep a whole number; use “.1”, “.2”… for mid-cycle improvements. 4, 4.1, 4.5 Tier Rough capability / cost band, easy to interpret. Lite, Standard, Pro, Ultra Suffix (optional) Special optimisation or domain specialisation. -LongCtx, -Vision, -Code

Why this works • No ambiguous letters or numerics – “o3” can be read as “03” or “oz”; avoid that entirely. • Self-explanatory language – non-technical users recognise “Lite” versus “Pro” instantly. • Scalable – new minor rev? bump the generation (4.2). Need a cheaper size? add a Nano tier without disturbing the rest.

⸻

2. Applying it to your current list

Current name Proposed new name Rationale GPT-4o GPT-4 Standard Baseline flagship of the 4-series. o3 GPT-4 Lite Same generation, lowest tier. o4-mini GPT-4 Lite+ (or GPT-4 Lite LongCtx if that’s the point) Indicates “Lite” family but a bit more capable; “+” or a suffix clarifies how. o4-mini-high GPT-4 Standard LongCtx (or GPT-4 Lite Pro) Pick one dimension: either it’s still “Lite” but higher context, or it has moved into “Standard”. GPT-4.5 GPT-4.5 Standard Mid-cycle architectural upgrade, default tier. GPT-4.1 GPT-4.1 Standard Ditto. GPT-4.1-mini GPT-4.1 Lite Same generation, smaller/cheaper option.

⸻

3. Quick style guide for future models 1. Stick to two words (or two words + optional suffix) GPT-5 Pro, GPT-5 Lite-Vision – still readable at a glance. 2. Reserve extra punctuation for special cases only Hyphens or the “+” symbol should signal meaning, not be decorative. 3. Publish a public matrix A small table in docs or the dashboard that maps Generation × Tier → context length, cost, latency eliminates guesswork.

⸻

One-line summary

GPT- [-Specialisation] keeps names short, descriptive and future-proof—so even non-technical users can tell instantly which model suits their needs. ```

o4-mini? It is a rumored model building up on o3, and different from 4o-mini. Sorry, their naming scheme is a mess. Would be cool if you do have o4-mini access though.

https://www.theverge.com/news/646458/openai-gpt-4-1-ai-model

I have been using the Obsidian web clipper to export chats from ChatGPT and Claude web versions to nicely-formatted md files. You can save md to Obsidian or download it as a standalone file. It doesn’t support Gemini yet though.

https://github.com/obsidianmd/obsidian-clipper

Interesting discussion on the merits of the initial plan of a paid link shortening service and the opposite approach (easy-to-setup paid access to links).They were discussing adding Bitcoin as a payment method when it was 0.7 cents a pop.

Is it just me or does Gemini fail the 4D tesseract spinning challenge? That solution looks like a 3D object spinning in 3D space. It seems Claude's solution is better (still difficult to interpret). For reference, this is what a 4D rotation projected to 3D should look like: https://en.wikipedia.org/wiki/Tesseract

Did they time it with the Gemini 2.5 launch? https://news.ycombinator.com/item?id=43473489

Was it public information when Google was going to launch their new models? Interesting timing.

Cellular automata in general can exhibit chaos

https://en.wikipedia.org/wiki/Rule_30

Hi, I have a latex resume/cv too and am curious about "some protections against recruiters trying to copy and paste and change things". What does that mean?

Wonderful vizualization and library! Many years ago I made a Chua's circuit PCB for an electronics course[0]. We had to dust off an old Cold War era analog oscilloscope to actually see the attractors because the resolution of the newer digital ones induced too many artefacts.

I want to use this library to vizualize this using the electronic component values now.

[0] https://karan.sh/res/Chua_Circuit_report.pdf

This actually happened much earlier in the first half of the 20th century. It was an international cartel with household names (GE, Osram, Phillips etc).

"The cartel tested their bulbs and fined manufacturers for bulbs that lasted more than 1,000 hours."

https://en.wikipedia.org/wiki/Phoebus_cartel

https://spectrum.ieee.org/the-great-lightbulb-conspiracy

Probably the Turing award for discovering a breakthrough compression scheme.

I am currently working on physics-informed ML models for accelerating DFT calculations and am broadly interested in ML PDE solvers. Overall, I think physics-informed ML (not just PINNs) will be very impactful for computationally heavy science and engineering simulations. Nvidia and Ansys already have "AI" acceleration for their sims.

https://developer.nvidia.com/modulus

https://www.ansys.com/ai

The Milky Way is also the only known galaxy to have ever contained Magellan.

https://en.wikipedia.org/wiki/Ferdinand_Magellan

I think the parent comment is addressing the Kindle app on iPad rather than the iPadOS. IMO the Kindle app has been buggy on iPad and Mac. They recently overhauled the Mac app to be more like iPad (Catalyst I guess).

There's also the issue of drift [0]

[0] https://en.wikipedia.org/wiki/Drift_(linguistics)

And mummy/pappa would be used be used colloquially in Gujarati, apart from the more formal mata/pita and other terms like ma/bapu.

It's not exactly at the South Pole. It is somewhere around 70 deg latitude. It is lunar day now which will last for about 12 (earth) days. After lunar night falls, the rover and the lander will probably stop working. They might try reviving them in the next lunar day but there are no guarantees given the extreme conditions of the lunar night. Also this might be an interesting tangent (not related to CH3): https://en.wikipedia.org/wiki/Peak_of_eternal_light

Stretching the analogy from the original commenter, the code is “in production “. It is already running, it is the conscious part of you. Reading all the self help books in the world will only reinforce the things that the system already believes and reject uncomfortable things. As for introspection, often by the time people get therapy, the processor load is already high enough that there is little room for introspection. A therapist is a trained debugger, and can help you get out of your own head and see things more “objectively”.

Also knowing is different from doing. What’s stopping you learning karate or bjj from books or even videos? You already have the hardware for it. But you need to learn new patterns, and a coach can often get you there more effectively. Also having a sparring partner is a good way to break old patterns and build new ones. Now replace karate with mental models/peace of mind etc, except now your system is not performing optimally by this point.

In my field, we invoke the name of Oppenheimer (the Born-Oppenheimer approximation[0,1]) in many of our formulations, every paper we write or talk we give. He did this work when he was a 23 year old grad student. He seems to have led a complex life. Either ways, I don't see the point of this article. While building on the shoulder of giants, he managed to lead the Manhattan project to fruition under the most existential of deadlines. We could even argue that this opening of Pandora's box has had the most consequential direct impact on modern society.

[0] https://en.wikipedia.org/wiki/Born–Oppenheimer_approximation [1] https://gallica.bnf.fr/ark:/12148/bpt6k15386r/f475.item - page 457

It is the common central/federal baseline. It is accepted as a high school credential everywhere in India (other boards are accepted in general too, but NCERT has to be accepted).

This insufficiency with weird omission for evolution and periodic table seems to be due to covid after effects or general incompetence rather than malice.

I studied in 3 different boards at different points of schooling and the rule of thumb was state board, NCERT/CBSE and ICSE in increasing order of “rigor”. The last two years of high school are spent in preparing for competitive entrance exams anyways, whose syllabus usually is a superset of the school syllabus in your chosen field (engineering, medicine, sciences etc.)

I use a combination of Tailscale and Nord Meshnet on Raspberry Pis that I have set up at my home and family home in different countries as my personal VPN. Home country does not have a good relationship with VPNs and the commercial VPN services discontinued their servers there. So now I get a clean residential IP from my family home when I want to surf from that country.

I built something similar (Chua's circuit) for an electronics course in college[1]. It was a lot of fun. While testing it, we couldn't get it to display the phase portraits on an oscilloscope for a long time, and then we realised we were using a modern "digital" oscilloscope. I think the (high but finite) sampling rate was not enough. The professor found an old analog one from the 80s in his lab and then we got the pretty phase portraits. This site was super helpful at that time (2016)[2]

[1] https://karan.sh/res/Chua_Circuit_report.pdf [2] http://www.chuacircuits.com

I really want to try the code interpreter. I got access to the plugins early but there was no browsing and interpreter. Then I got access to browsing when they released it widely last week. Does anybody know how to get access to the code interpreter? I have lost track of all the ${LLM_Service} waitlists I have signed up for.

It allows you to install a vpn on iOS that blocks traffic to domains on your blocklist, effectively blocking apps (unless you download things offline like YouTube videos or Podcasts).

I am currently very conflicted about this. I grew up in India but was fortunate to live in the US and now in Europe for undergrad/grad school. I am (slightly) older than the demographic in the article. I am planning on settling down with my girlfriend soon, after I finish some career related stuff. If I want to start a family, I really want to move back to my hometown in India, with the strong social support system. While moving around has been exciting and an eye-opening experience, I am tired of constantly making new friends and leaving them after a few years, and miss my family some times, and eventually the kids would benefit from being around their grandparents. On the other hand, moving back would be career limiting.

Does anybody who is further along on the journey have any advice? How did you make this decision?

We used a fork of Logisim for the computer architecture courses (CS 2110/2200?)at Georgia Tech as well. It was called Brandonsim[1] after the hyper productive Head TA Brandon who adapted it for our courses.

[1] https://github.com/TricksterGuy/Brandonsim

Slight tangent, but have you considered using these new generative models like Dall E 2[1]? So far my experience has been that they generate great oil paintings. Maybe there could also be an automated pipeline that generates oil painting covers using book titles or first sentences.

[1] https://openai.com/dall-e-2/

I agree, the usage of the word “epidemic” has become an epidemic.