lbr

Enterprise AI - AI agents harness - Local LLMs

[lbr @teleogen]

@javierbits https://t.co/wdzkcQyN1G

[lbr @teleogen]

@tysn_dev Maki https://t.co/wdzkcQzkRe

[lbr @teleogen]

@joelniklaus Great job. Can you please try Maki which is designed to be fast and token efficient: https://t.co/JjY4DlFLDT

[lbr @teleogen]

The most bullish thing for the AI sector is to actually cause measurable deflation.

[lbr @teleogen]

Generative AI is great at writing code.
It’s terrible at deleting it.
The next frontier is Subtractive AI: systems rewarded for what they remove and simplify — not what they add.

[lbr @teleogen]

8/8 TL;DR:

Looped transformers = better quality per token, but 3-5× more compute.

Smaller models. Higher latency. More GPUs needed for same throughput.

The frontier is getting deeper, not wider.

And the hardware bill is coming.

[lbr @teleogen]

1/8 Looped transformers are coming.

OpenMythos, recurrent-depth architectures, weight recycling — the frontier labs are all exploring this.

But here's what nobody's talking about: the hardware implications are massive.

Let me break it down. 🧵

[Kye Gomez (swarms) @KyeGomezB]

Introducing OpenMythos

An open-source, first-principles theoretical reconstruction of Claude Mythos, implemented in PyTorch.

The architecture instantiates a looped transformer with a Mixture-of-Experts (MoE) routing mechanism, enabling iterative depth via weight sharing and conditional computation across experts.

My implementation explores the hypothesis that recursive application of a fixed parameterized block, coupled with sparse expert activation, can yield improved efficiency–performance tradeoffs and emergent multi-step reasoning.

Learn more ⬇️🧵

[lbr @teleogen]

7/8 Investment takeaway:

Don't just watch VRAM demand. Watch:
• Compute density per token
• Cache efficiency
• Interconnect tech
• Specialized recurrent hardware

The next hardware cycle won't look like the last one.

[lbr @teleogen]

This is how vLLM got started. The best explanation by far!

[Avi Chawla @_avichawla]

https://t.co/zu8HupqrL5

[lbr @teleogen]

Such a big deal. Alternatives are @UnslothAI and @axolotl_ai

[Workshop Labs @WorkshopLabs]

Introducing Trellis for Kimi K2 Thinking.

It's post-training code that's 50x faster than the best single-node open-source version and 2x cheaper than training APIs.

After safety testing, we're open-sourcing it, giving builders the best tools to customize a frontier model.

🧵

[lbr @teleogen]

@__tinygrad__ https://t.co/qaEzrMcenW

[lbr @teleogen]

@__tinygrad__ Reminds me of Sun modular data centres in 2006 https://t.co/qaEzrMbGyo

[lbr @teleogen]

@steipete @nakulsinglla Try using the latest OpenAI Symphony with Codex and others like Gemini CLI and Claude Code behind a Codex gateway over WebSocket https://t.co/w4FKCEhB1Y

[lbr @teleogen]

@0xSero May be related to the disruption of the AWS data centre in UAE

[lbr @teleogen]

@xeophon The performance is substantially below the medium models (unsurprisingly at less than 10% size) but can be useful for fine-tuning and further domain specific RL for edge devices