Hacker News: lucrbvi

New comment by lucrbvi in "I've operated petabyte-scale ClickHouse clusters for 5 years"

lucrbvi — Fri, 11 Sep 2026 15:11:44 +0000

That's a lot of ®, curious how ClickHouse® Inc. is treating the use of its name by others ... Hopes it's not like Oracle with JavaScript

New comment by lucrbvi in "Speculative Decoding in vLLM on AMD GPUs"

lucrbvi — Mon, 07 Sep 2026 13:30:30 +0000

The target model is the original LLM that is large and expensive. It can verify candidate tokens in a single forward pass. It means you give all the context + candidate tokens that passes in parallel in the backbone, then you pass the language head (a matmul transformation to produce the token distribution) on all the candidate tokens and you can keep or drop tokens based on how many "quality" you want.

New comment by lucrbvi in "Laion Big Video Dataset"

lucrbvi — Thu, 27 Aug 2026 07:13:19 +0000

Under the EU's AI Act non-profits and universities can just bypass these constents to make non-commercial datasets btw

New comment by lucrbvi in "uBlock Origin Is Giving Up the Fight to Keep Ads Off Facebook"

lucrbvi — Wed, 12 Aug 2026 13:09:27 +0000

It would be fun to train a small vision neural network to recognize ads instead of relying on HTML or network requests. I don't know if anyone has done this yet.

New comment by lucrbvi in "Building a Rust Inference Engine That Matches Llama.cpp"

lucrbvi — Sat, 08 Aug 2026 17:09:29 +0000

https://www.pangram.com/history/c2abdcf5-c52b-41cc-a8fb-0c59...

New comment by lucrbvi in "Mistral's Shieldstral: 3B open-weights model for multimodal moderation"

lucrbvi — Tue, 04 Aug 2026 21:08:14 +0000

Mistral has the capability of training such models. Take a look at Poolside[1], they are claiming to pre-train their Laguna series of models on 4,096 NVIDIA H200 GPUs[2]. Mistral has approximately 13,800 NVIDIA GB300 GPUs, which are nearly 2x more efficient for training.

The problem with Mistral is that they do not seem to have aligned incentives to train big open-weight models, even if the teams would like to.

[1]: https://poolside.ai/ [2]: https://poolside.ai/blog/introducing-laguna-s-2-1

New comment by lucrbvi in "GPT-5.6 vs. Claude Fable 5 for Physical AI, which performs best?"

lucrbvi — Wed, 29 Jul 2026 18:13:26 +0000

"world model" is, sadly, a new buzzword; most of the world models we are seeing are video generative models (like GPT-Image), they receive an action vector and previous frames and must generate the next frame. It's working of course.

The whole field of world models also have lighter approaches based on using latent representations as the target (and not pixel frames). LeWorldModel[0] and DINO-WM[1] are a good way to learn how these works.

Strangely, most of the startups in the space of world models are focused on video-generative world models.

A good application of world models is in robotics where an action policy generate a distribution of actions, you can then predict the future state of the world and use some reward policies to choose the "best" future state.

New comment by lucrbvi in "MAI-Cyber 1"

lucrbvi — Mon, 27 Jul 2026 17:51:29 +0000

Seems like MAI models from Microsoft are not going to be open-weight soon, but they are sharing a lot of details in the making of these models, which is a weird position.

New comment by lucrbvi in "LeMario: Training a JEPA World Model on Super Mario Bros"

lucrbvi — Wed, 15 Jul 2026 00:03:51 +0000

Such a gem, thanks to the author for sharing it's findings :)

The only problem I have with planing in latent space is that it can be really noisy and not representative of the positions in the game (the latent are trained for semantic, so the optimizer can focus a set of specific features and can skip positions, which means it cannot know "where" to go by optimizing on the latents directly).

New comment by lucrbvi in "A global workspace in language models"

lucrbvi — Mon, 06 Jul 2026 19:42:57 +0000

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.

New comment by lucrbvi in "'Sloppenheimer:' Amazon employees mock the company's AI on Slack"

lucrbvi — Tue, 09 Jun 2026 16:59:31 +0000

http://archive.today/1YRCE

New comment by lucrbvi in "Universal Memory Protocol – a shared format for agent memory"

lucrbvi — Sat, 06 Jun 2026 21:28:16 +0000

https://xkcd.com/927/

Building a Latent World Model of Doom

lucrbvi — Mon, 25 May 2026 11:49:01 +0000

Article URL: https://lucrbvi.com/blog/world-model/

Comments URL: https://news.ycombinator.com/item?id=48265766

Points: 1

# Comments: 0

New comment by lucrbvi in "Was my $48K GPU server worth it?"

lucrbvi — Fri, 22 May 2026 09:58:29 +0000

In the article the author says they are doing reinforcement learning with LLMs.

New comment by lucrbvi in "Show HN: Forge – Guardrails take an 8B model from 53% to 99% on agentic tasks"

lucrbvi — Tue, 19 May 2026 21:35:01 +0000

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

New comment by lucrbvi in "Show HN: Watch a neural net learn to play Snake"

lucrbvi — Sat, 16 May 2026 07:20:00 +0000

it's using webgpu kernels, probably a false positive

New comment by lucrbvi in "QBE – Compiler Back End"

lucrbvi — Fri, 08 May 2026 15:12:11 +0000

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.

New comment by lucrbvi in "Show HN: Adam – An embeddable cross-platform AI agent library"

lucrbvi — Wed, 06 May 2026 15:38:12 +0000

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.

New comment by lucrbvi in "DeepSeek v4"

lucrbvi — Fri, 24 Apr 2026 09:12:41 +0000

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.

World Models will push the frontier for LLMs

lucrbvi — Wed, 22 Apr 2026 15:31:04 +0000

Article URL: https://lucrbvi.bearblog.dev/world-models-will-push-the-frontier/

Comments URL: https://news.ycombinator.com/item?id=47865052

Points: 3

# Comments: 0