Hacker News: belluxx

New comment by belluxx in "DeepSeek-v4.1-Flash-1017"

belluxx — Sat, 17 Oct 2026 13:16:45 +0000

Pi is very good with open weight models (proprietary ones too)

New comment by belluxx in "Watch AI materials-science and bioscience abilities closely"

belluxx — Mon, 14 Sep 2026 13:36:56 +0000

Lately "one-shotting" means "without human steering / feedback"

I think its a consequence of the improvements in harnesses and LLMs of the last year. Before they had a hard time self-reviewing the results and proceeding to a second iteration, now they are quite good at it although not perfect.

Show HN: Nanointerpret – LLM Interpretability Playground

belluxx — Thu, 27 Aug 2026 13:52:44 +0000

I find LLM interpretability extremely interesting and wanted to create a minimal repo for: - SAE training - Automatic feature interpretation - Visualizing features and running interventions through a GUI

You can try it here: https://nanointerpret.pages.dev/ Or check the repo: https://github.com/Belluxx/nanointerpret

Comments URL: https://news.ycombinator.com/item?id=49464942

Points: 5

# Comments: 1

Show HN: TransFire – Chat with your local LLMs without port-forwarding

belluxx — Mon, 06 Oct 2025 13:22:21 +0000

Article URL: https://github.com/Belluxx/TransFire

Comments URL: https://news.ycombinator.com/item?id=45491133

Points: 1

# Comments: 0

Show HN: Local LLM AIME benchmarking tool

belluxx — Sat, 31 May 2025 12:11:21 +0000

I made this simple tool to compare local LLMs. Any provider that supports OpenAI-like APIs can be used (LMStudio, Llama.cpp, Ollama) but you can also use Openrouter/OpenAI if you change the base URL accordingly.

In my opinion it is not particularly useful for comparing different models from different companies since some models are optimized heavily on math or even trained on AIME problems.

However it is really useful for testing different quantizations of the same model or the same quantization from different providers.