Hacker News: thomascountz

Teachers decry plan for humanoid robot in New York high school

thomascountz — Sun, 26 Jul 2026 07:32:30 +0000

Article URL: https://www.theguardian.com/us-news/2026/jul/25/new-york-humanoid-robot-teachers-school

Comments URL: https://news.ycombinator.com/item?id=49055628

Points: 6

# Comments: 0

New comment by thomascountz in "Running a 28.9M parameter LLM on an $8 microcontroller"

thomascountz — Sun, 26 Jul 2026 07:22:46 +0000

Many artificial environments are useless to tasks they were not designed for until somebody experiments, tests, redesigns, and iterates.

I'll give a specific example apropos of TFA. Computer vision models were never run on MCUs because they were constrained to the point of being useless for this task, but then someone tried the impractical, and now it's trivial[1]

Regarding MCU shortages, you should be worried about the supply chain, but I don't see the impact really being from running LLMs on ESP-32s, of all things.

[1]: https://github.com/espressif/esp-vision

New comment by thomascountz in "Taylor Farms Called White House to Try to Delay Cyclospora Recall"

thomascountz — Sat, 25 Jul 2026 04:22:26 +0000

http://archive.today/2026.07.25-012850/https://www.wsj.com/h...

New comment by thomascountz in "How to write a quine"

thomascountz — Sat, 25 Jul 2026 04:18:03 +0000

I learned something and along the way, I really enjoyed the author's writing style!

   Phew, it works. But it wasn’t easy, was it? That’s why remember my advice: when writing a quine, avoid using quotes. Once you start escaping them, you’ll never get out to go sledding. At this point, I stop writing the text and go sledding.

New comment by thomascountz in "Show HN: A self-running space economy SIM in Rust and Bevy"

thomascountz — Tue, 21 Jul 2026 20:06:34 +0000

Reminds me of SpaceTraders[1]

[1]: https://spacetraders.io/

New comment by thomascountz in "Claude Fable produced a counterexample to the Jacobian Conjecture"

thomascountz — Tue, 21 Jul 2026 07:22:20 +0000

Not OP, but here's Gemini's reaction: https://aistudio.google.com/app/prompts?state=%7B%22ids%22:%...

New comment by thomascountz in "Funny item co-occurrences in 3.2M Instacart orders"

thomascountz — Sat, 18 Jul 2026 11:02:57 +0000

Baby food is often just puréed fruits and vegetables. Adult diapers are only funny if you've never considered that being able-bodied is temporary.

It is more likely that the person purchasing adult diapers and baby food is the caregiver of an adult. Perhaps of themselves, or an aging parent, or their spouse who is recovering from surgery.

New comment by thomascountz in "Vancouver PD website features Quick Escape button that wipes itself from history"

thomascountz — Wed, 15 Jul 2026 05:55:09 +0000

There are tools which exist to make things such as this accessible to non-CS major perpetrators.

New comment by thomascountz in "Notable Knot Index (2016)"

thomascountz — Tue, 14 Jul 2026 11:24:33 +0000

Animated Knots by Grog[1] and 3D Knots[2] are oft-cited places to learn knots on the internet, but OP's is such a nice throwback. Don't miss the "requested entries" page with 28 more knots![3]

[1]: https://www.animatedknots.com/

[2]: https://knots3d.com/

[3]: https://knots.neocities.org/requested

New comment by thomascountz in "Silent speech with ultrasound"

thomascountz — Sat, 11 Jul 2026 16:16:08 +0000

s/2023/2003/

New comment by thomascountz in "Silent speech with ultrasound"

thomascountz — Sat, 11 Jul 2026 13:04:30 +0000

The use of ultrasound is interesting! The first step (identifying speech) is called subvocalization recognition, and early research (Chuck Jorgensen, et. al (2023) from NASA) use EMG[1].

[1]: https://scispace.com/pdf/sub-auditory-speech-recognition-bas...

New comment by thomascountz in "Every new car sold in the European Union must include a driver monitoring camera"

thomascountz — Wed, 08 Jul 2026 05:40:46 +0000

I with my car had an accessible log of these notices. Another example, my car gives you a safety score at the end of every drive, out of 100. When I get less than 100, there's no information about what I might do to improve my score, and ostensibly, drive more safely.

New comment by thomascountz in "Yes, AI Will Take Your Job"

thomascountz — Tue, 07 Jul 2026 15:23:57 +0000

If AI replaces most software developers, it may be because there is less use for the software (read: mostly SaaS applications) developers are employed to build. Or otherwise, there will be fewer people who will want to buy software (or buy it indirectly from ad conversion).

New comment by thomascountz in "Apple raises prices of MacBooks, iPads"

thomascountz — Thu, 25 Jun 2026 14:22:37 +0000

RAM impacts engineers' machines. We learn to build smaller again. More breakthroughs happen around less-memory intensive local inference. Model provides' bottom lines are impacted. They bail on RAM contracts. The market floods. Private inference becomes flush with resources. The third-wave of local models begins, but RAM trauma keeps things lean. Nature heals?

New comment by thomascountz in "Pledging another $400k to the Zig software foundation"

thomascountz — Mon, 22 Jun 2026 14:32:39 +0000

Money can't buy happiness, but it can buy less unhappiness. There's diminishing returns, of course, but I'd hazard it looks a bit like ln(n), in that the returns are quite significant in the beginning.

New comment by thomascountz in "Good results fine tuning a local LLM like Qwen 3:0.6B to categorize questions"

thomascountz — Mon, 22 Jun 2026 06:40:30 +0000

Yes, you can use constrained decoding like logit masking to force all invalid tokens in the vocabulary to -inf, and effectively be removed from selection. I believe llama.cpp exposes this by accepting a formatted grammar.

New comment by thomascountz in "Rare medieval bookmark exceeds expectations at auction"

thomascountz — Sun, 21 Jun 2026 16:34:16 +0000

I imagined the same; and in that case, it wouldn't need to follow that it was transcriber's bookmark.

New comment by thomascountz in "Commodore announces Linux-based flip phone with 'no social media, no browser'"

thomascountz — Tue, 16 Jun 2026 16:33:38 +0000

New comment by thomascountz in "Getting Creative with Perlin Noise Fields"

thomascountz — Tue, 16 Jun 2026 12:56:13 +0000

If you're into exploring this sort of thing yourself, I can extremely highly recommend Daniel Shiffman's Nature of Code[1]. It's among my top-5 recommended books about computing in general, let alone for algorithmic art.

[1]: https://natureofcode.com/

New comment by thomascountz in "Can Europe train a frontier AI model on the compute it owns?"

thomascountz — Tue, 16 Jun 2026 09:53:52 +0000

Ok, now I understand better the metaphor. I don't understand your conclusion about AI hype, but at at rate, your earlier point is clearer.

The story I'm not readily conceiting to is that trillion-dollar frontier model training capex is the only thing that results in the defense capabilities countries need in relation to information security. From my understanding, the cyber wins being celebrated of late have come just as much (if not more) from how models are wielded than how they were trained.

Maybe the more honest framing of my question is: at what level of capability does sovereign training become the only viable path, and is that level the same as the trillion-dollar pretraining capacity TFA is actually asking about, or is it rather something more tactical/accessible? That is the part I'm genuinely more interested in, since the closer you are to benefiting from the commercial aspects of model training, the more likely it is that the answers will simply follow the money, so to speak.