HackerLangs
TopNewTrendsCommentsPastAskShowJobs

lucrbvi

91 karmajoined 6 mesi fa
https://lucrbvi.com

Submissions

Building a Latent World Model of Doom

lucrbvi.com
1 points·by lucrbvi·2 mesi fa·0 comments

World Models will push the frontier for LLMs

lucrbvi.bearblog.dev
3 points·by lucrbvi·3 mesi fa·0 comments

Why Claude Mythos is not AGI

lucrbvi.bearblog.dev
2 points·by lucrbvi·3 mesi fa·1 comments

Next Embedding Prediction Makes World Models Stronger

arxiv.org
1 points·by lucrbvi·4 mesi fa·0 comments

Ask HN: Maintainers, do LLM-only users often clutter your issues/PRs?

9 points·by lucrbvi·4 mesi fa·9 comments

comments

lucrbvi
·4 giorni fa·discuss
Anthropic theorize that middle layers in an LLM is a "J-Space" used to "think" about the future answer or about abstract concepts.

Their method is used to identify which tokens can appears in which layers of the model.
lucrbvi
·mese scorso·discuss
http://archive.today/1YRCE
lucrbvi
·mese scorso·discuss
https://xkcd.com/927/
lucrbvi
·2 mesi fa·discuss
In the article the author says they are doing reinforcement learning with LLMs.
lucrbvi
·2 mesi fa·discuss
How does this differ from dottxt's Outlines[0] on the technical level? Are you using some JSON grammar to force the LM head distribution to follow it?

[0]: https://github.com/dottxt-ai/outlines
lucrbvi
·2 mesi fa·discuss
it's using webgpu kernels, probably a false positive
lucrbvi
·2 mesi fa·discuss
I really like QBE but I hope they will make it a true library because I don't want to launch a subprocess to compile a program.
lucrbvi
·2 mesi fa·discuss
Cool but why would anyone write an agent in C? Since the language is verbose it seems to be a painful choice to use C for this task.
lucrbvi
·3 mesi fa·discuss
They have added a lot of optimization focussing on the KV-cache, so they can have a much larger window without eating all the VRAM.

The 1M window might be usable, but it will probably underperform against a smaller window of course.
lucrbvi
·3 mesi fa·discuss
Mollie seems to only provide services to business based in European Economic Area, Switzerland and the UK [0], so yes?

[0]: https://help.mollie.com/hc/en-us/articles/115002116105-Can-I...
lucrbvi
·3 mesi fa·discuss
Mollie might be a direct competitor
lucrbvi
·3 mesi fa·discuss
Is there anyone that feels that LLMs are wrong for computer use? It's like robotic, if find LLMs alone are really slow for this task
lucrbvi
·3 mesi fa·discuss
Some people are speculating that Opus 4.7 is distilled from Mythos due to the new tokenizer (it means Opus 4.7 is a new base model, not just an improved Opus 4.6)
lucrbvi
·3 mesi fa·discuss
Hi! I share my first blog post ever about the definition on AGI following the annoucement of Claude Myhtos Preview.

I hope some of you might found this interesting.
lucrbvi
·4 mesi fa·discuss
Sounds like Multi-Head Latent Attention (MLA) from DeepSeek
lucrbvi
·4 mesi fa·discuss
I wouldn't be surprised if Vercel were bought by Anthropic/OAI (but maybe it would be too expensive?)
lucrbvi
·4 mesi fa·discuss
This is a weird pattern accross OpenAI/Anthropic to buy startups building better toolings.

I don't really see the value for OAI/Anthropic, but it's nice to know that uv (+ ty and many others) and Bun will stay maintained!
lucrbvi
·4 mesi fa·discuss
I share the feeling; but people using it are mostly non-technicals (despite the 50+ config files lol) and are just runing it constantly to do random things.

But a message bot + Claude Code/Codex would be the better version
lucrbvi
·4 mesi fa·discuss
The whole article feels like an ad
lucrbvi
·4 mesi fa·discuss
The "egg" system seems really good!

It will maybe solved soon if we train yet another neural network on scanning GitHub activities; but by also adding other forges like codeberg, gitlab, self-hosted forgejo, etc... to not lock non-github users out

Still really good idea!