Hacker News Reader

redman25on The path to ubiquitous AI (17k tokens/sec)

Many older models are still better at "creative" tasks because new models have been benchmarking for code and reasoning. Pre-training is what gives a model its creativity and layering SFT and RL on top tends to remove some of it in order to have instruction following.

redman25on Monty: A minimal, secure Python interpreter written in Rust for use by AI

They’re good at it because they’ve learned from the existing mountains of python and javascript.

redman25on In 6 violent encounters, evidence contradicts immigration officials' narratives

Why are they more inclined to be non-violent?

redman25on Bugs Apple loves

Not to mention not resolving local domains (and attempting to search instead) unless they specifically contain the `http://`.

redman25on The challenges of soft delete

Exactly, partition the table vertically by month. Surprised no one else seems to be mentioning this.

redman25on Why we built our own background agent

It's really a shame because anthropic had a lot of opportunity to show good will by open sourcing claude code.

redman25on TimeCapsuleLLM: LLM trained only on data from 1800-1875

It's a base model. It hasn't been instruction tuned to "solve problems" necessarily. All it can do is attempt to complete text given some starting text.

redman25on A 30B Qwen model walks into a Raspberry Pi and runs in real time

That's interesting. For what projects would you want the "sheen of natural language" though?

redman25on MiniMax M2.1: Built for Real-World Complex Tasks, Multi-Language Programming

https://www.swebench.com

https://swe-rebench.com

https://livebench.ai/#/

https://eqbench.com/#

https://contextarena.ai/?needles=8

https://metr.org/blog/2025-03-19-measuring-ai-ability-to-com...

https://artificialanalysis.ai/leaderboards/models

https://gorilla.cs.berkeley.edu/leaderboard.html

https://github.com/lechmazur/confabulations

https://dubesor.de/benchtable

https://help.kagi.com/kagi/ai/llm-benchmark.html

https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard

redman25on The Easiest Way to Build a Type Checker

Since Haskell is statically compiled, wouldn't it not compile at all?

redman25on Tell HN: X is opening any tweet link in a webview whether you press it or not

A whole lot of machine learning practitioners use X. Makes it difficult to avoid if you're interested in the news. It's definitely a network effect issue.

redman25on Ask HN: Who wants to be hired? (November 2025)

  Location: Chicago
  Remote: yes, hybrid ok
  Willing to relocate: no
  Technologies: rust, python, postgres, PEG, terraform, google cloud platform (GCP), flask, react, redis, sqlalchemy
  Resume: joshvoigts.com/resume.pdf
  Email: joshvoigts@gmail.com

I'm Josh, a fullstack developer of more than 10 years with specific experience writing APIs and parsers for backend data ingestion. I write rust and python but have used many languages and frameworks. I'm quick to learn and flexible.

redman25on Futurelock: A subtle risk in async Rust

It's a good idea in concept but tons of popular libraries use async which makes it difficult to avoid. Want to do anything with a web server or sending requests, most likely async for popular libraries.

redman25on Run interactive commands in Gemini CLI

I agree, gemini pro is a great model for coding if you don't need to do agentic work. I've found that it's a lot less "wordy" when editing, debugging, reviewing, etc. It gets to the point whereas other models can provide long useless explanations. It's also very smart and great with long context.

redman25on Programming in the Sun: A Year with the Daylight Computer

I've found the opposite, I suppose. I've found I'm more likely to run across apps or sites that don't support dark mode than sites that are natively dark that invert makes worse (because they turn light).

redman25on Programming in the Sun: A Year with the Daylight Computer

I use lightmode full time for that exact reason. Other developers tend to think it's weird when pair programming or whatnot.

At night, I just use macos built-in accessibility functions to invert the screen. Work pretty well but sometimes you have to un-invert to view photos.

redman25on Gemini 2.5 Computer Use model

They could literally run 24/7 overnight assuming they eventually become good enough to not need hand holding.

redman25on OpenAI's H1 2025: $4.3B in income, $13.5B in loss

It is in terms of users. There's a lot of sticking power to "the thing you already know and use".

redman25on I built ChatGPT with Minecraft redstone [video]

I guess redstone still somewhat follows the physics of electricity. Maybe a quantum simulator?

redman25on A postmortem of three recent issues

I wish I had 100 upvotes to give you. Weak, heavily mocked tests are my biggest pet peave. Test “quality” is important and not something a lot of devs pay attention to.

I’ve found myself preferring integration tests or unit tests with a “real” database set up because the tests are much more effective. If you design them right, they don’t even need to be slower.

redman25on Rust: A quest for performant, reliable software [video]

I think I disagree with this sentiment. I’ve found AI to be poor at one-shotting rust but give it a short agentic loop running `cargo check` and it can often produce code more solid than that created with a dynamically typed language.

redman25on Bear is now source-available

Did you reply to the wrong comment? The posters above never mention GPL.

redman25on AI adoption linked to 13% decline in jobs for young U.S. workers: study

Do you really believe that all technological progress has bettered humanity? Where’s the four day work week we were promised? I thought automation was supposed to free us from labor.

redman25on Qodo CLI agent scores 71.2% on SWE-bench Verified

That sounds like an interesting idea to me. It would at least resolve the problem of companies gaming the metric.

Another approach might be the LiveBench approach where new tests are released on a regular basis.

redman25on Qwen3-4B-Thinking-2507

I’m American. Just giving some background to the feeling. There’s some discontent with some western communities (localllama) that Chinese developers have been open weighting all of their models while most western models have been closed weights.

redman25on FoundationDB: From idea to Apple acquisition [video]

Arroyo uses it for their streaming database. Think streaming stateful aggregations. It’s pretty cool tech. Looks like they were recently acquired by cloudflare.

redman25on RisingWave: An Open‑Source Stream‑Processing and Management Platform

Solutions like this can help with complex transformations that rely on intermediate state like streaming aggregations.

They often have watermarking, windowing, and all that good stuff built in whereas with redis you would have to build all of that in your application.

redman25on Upgrading an M4 Pro Mac mini's storage for half the price

It can be good to know that SSDs are fast until you exhaust the cache by copying gigs of files around.

It doesn’t hurt to be aware of the limitations even if for the common case things are much better.

redman25on Sam Altman Slams Meta’s AI Talent Poaching: 'Missionaries Will Beat Mercenaries'

Sad to say but it would have put US companies at a major disadvantage if they were not allowed to.

redman25on MCP in LM Studio

Qwen is great but for creative writing I think Gemma is a good choice. It has better EQ than Qwen IMO.