Hacker News: jmpeax

New comment by jmpeax in "Jeff – Jev-compatible 0.8B decision models, trained at home, ~30 ms"

jmpeax — Tue, 29 Sep 2026 06:43:32 +0000

Then the probabilities are not calibrated. Jev is basically the LLM version of the "is-even" library.

New comment by jmpeax in "The safest job from AI may be writing"

jmpeax — Tue, 01 Sep 2026 00:05:01 +0000

> A prompt complex enough where it won't get any wrong ideas has to contain as much information as the output would have.. so just write it.

This is only if the output is fully compressed. Writing is not just about encoding the writer's ideas but also about how the reader will ingest those ideas. The writer needs to consider when to put in rests in between complex ideas to help the reader flow through the text. This suggests the LLM could be prompted by a dense complex idea to be presented with the boilerplate needed for the human mind read smoothly and without unnecessary effort.

New comment by jmpeax in "Parallel development without the headaches using Git worktree"

jmpeax — Sun, 23 Aug 2026 23:56:49 +0000

What a peculiar title. The title suggest parallel dev without using worktrees, but the article is about using worktrees.

New comment by jmpeax in "HTML Can Do That"

jmpeax — Fri, 21 Aug 2026 00:22:10 +0000

> LLMs are also terrible at these new standards.

Skill issue? I mean, I wonder if adding a skill to use these newer standards can push it over the edge from the crappy JS it was trained on.

New comment by jmpeax in "US hires over 2k video gamers as air traffic controllers"

jmpeax — Wed, 12 Aug 2026 01:51:42 +0000

The types of skills from FPS games and real fighting are much further apart than gaming and ATC. A closer analogy would be a coworker who knows Judo is needed for Greco-Roman wrestling. Very different sports with different rules, but the skills in building and navigating a mental map of the environment carry over.

New comment by jmpeax in "AI agent bankrupted their operator while trying to scan DN42"

jmpeax — Fri, 12 Jun 2026 09:07:19 +0000

This whole fiasco could have been prevented had the operator included "Make no mistakes" in the prompt.

New comment by jmpeax in "Gemma 4 12B: A unified, encoder-free multimodal model"

jmpeax — Thu, 04 Jun 2026 02:26:43 +0000

> nowhere in the announcement is coding mentioned

It's right there in the middle benchmark bar "LiveCode Bench" 72%.

New comment by jmpeax in "Learnings from 100K lines of Rust with AI (2025)"

jmpeax — Wed, 20 May 2026 13:13:12 +0000

All unit tests can pass if you don't assert anything. Just have to make sure to read through all 130k lines of code to check.

New comment by jmpeax in "Write some software, give it away for free"

jmpeax — Wed, 06 May 2026 07:48:29 +0000

This post misses that the people selling their "I built" SaaS advertised as a hobby/passion-project are just marketing something as indie that they fully intended to get rich from the beginning. It was never even a hobby worth giving away for free, it was always an attempt at making an easy fortune because they bought into the dream that infra was selling.

New comment by jmpeax in "State of the Windows: What is going on with Windows 11?"

jmpeax — Mon, 26 Jan 2026 22:32:09 +0000

it's probably been thrown out the window compared to office and chatgpt

New comment by jmpeax in "How I estimate work"

jmpeax — Sat, 24 Jan 2026 01:47:49 +0000

"If you refuse to estimate, you’re forcing someone less technical to estimate for you."

This is the perfect approach, given that estimates are top down and work to fill the estimate is bottom up.

"When I estimate, I extract the range my manager is looking for, and only then do I go through the code and figure out what can be done in that time."

New comment by jmpeax in "If you put Apple icons in reverse it looks like someone getting good at design"

jmpeax — Sun, 18 Jan 2026 01:27:37 +0000

Future icon will just be this: ∠

New comment by jmpeax in "Spherical Snake"

jmpeax — Tue, 06 Jan 2026 23:08:33 +0000

I like the surface dots like it is. It gives me two points of reference at the poles, and adds intuition for how long it takes to go around the sphere.

New comment by jmpeax in "Why does a least squares fit appear to have a bias when applied to simple data?"

jmpeax — Sun, 04 Jan 2026 23:32:58 +0000

From that wikipedia article, delta is the ratio of y variance to x variance. If x variance is tiny compared to y variance (often the case in practice) then will we not get an ill-conditioned model due to the large delta?

New comment by jmpeax in "Your job is to deliver code you have proven to work"

jmpeax — Fri, 19 Dec 2025 11:17:15 +0000

Actual architects design buildings.

New comment by jmpeax in "Your job is to deliver code you have proven to work"

jmpeax — Thu, 18 Dec 2025 20:31:54 +0000

Don't get me started on "software architect".

New comment by jmpeax in "Is Mozilla trying hard to kill itself?"

jmpeax — Wed, 17 Dec 2025 13:22:50 +0000

How is blocking ad blockers going to make them $150m?

New comment by jmpeax in "A linear-time alternative for Dimensionality Reduction and fast visualisation"

jmpeax — Tue, 16 Dec 2025 08:06:09 +0000

> They typically need to compare many or all points to each other, leading to O(N²) complexity.

UMAP is not O(n^2) it is O(n log n).

New comment by jmpeax in "Python Data Science Handbook"

jmpeax — Wed, 03 Dec 2025 08:50:24 +0000

Polars made the mistake of not maintaining row order for all operations, via the False-by-default argument of maintain_order. This is basically the billion-dollar null mistake for data frames.

New comment by jmpeax in "Every mathematician has only a few tricks (2020)"

jmpeax — Sat, 29 Nov 2025 23:01:58 +0000

> always respect human dignity even when nasty players try to make a dirty move against you

What a gem of a quote. A great way to avoid becoming a bitter person.