Avi Shefi

Consultant | AI & Distributed Systems

[Avi Shefi @shefiavi]

Small beginnings 🚧

Excited to build again, seems like a long time since I dealt with payments processing that it feels like removing rust.

[Avi Shefi @shefiavi]

@hrudolph It was breaking also 6-7 months ago

[Avi Shefi @shefiavi]

@thdxr jqwik to the extreme

[Avi Shefi @shefiavi]

@hwchase17 Barring the other parts of langgraph 😉

[Avi Shefi @shefiavi]

@neogoose_btw Any plans making a cli with cold start?

[Avi Shefi @shefiavi]

@neogoose_btw "Look there, a bird!"

[Avi Shefi @shefiavi]

@sudoingX What bench are you using?

[Avi Shefi @shefiavi]

@bridgemindai No more in the cheap

[Avi Shefi @shefiavi]

Let's go

[Kimi.ai @Kimi_Moonshot]

Introducing Kimi K3: Open Frontier Intelligence

🔹 2.8 Trillion Parameters, 1 Million Context, Native Multimodal
🔹 Kimi Delta Attention enables up to 6.3x faster decoding in million-token contexts
🔹 Attention Residuals deliver ~25% higher training efficiency at <2% additional cost
🔹 Built for long-horizon agentic coding and self-evolving workflows

Kimi K3 is now live on on https://t.co/zrk6zZxZUo, Kimi Work, Kimi Code, and the Kimi API.
Open Weights by July 27, 2026.

🔗 API: https://t.co/XCrgjXAqMw
🔗 Tech blog: https://t.co/YTfiMSNM1f

[Avi Shefi @shefiavi]

That's the most clarifying thing you've said all conversation.

[Avi Shefi @shefiavi]

Been working on graphing out skill relationships, I thought a tree would make sense but the more I think of it the more I realize my personal preferences don't mean much for LLMs.

There's a split between user-facing skills and skills that automated agents utilize, it's just that the line blurs once the same repo hosts both.

[Avi Shefi @shefiavi]

@SamuelBeek nice

[Avi Shefi @shefiavi]

@AbhiCodes15 Well, there is "I" in "AI" after all

[Avi Shefi @shefiavi]

@thdxr Didn't know kilo was an opencode fork

[Avi Shefi @shefiavi]

@liumengxinfly Some slop

[Avi Shefi @shefiavi]

That's Qwen on your phone

[PrismML @PrismML]

Today, we’re announcing Bonsai 27B: the first 27B-class model to run on a phone.

Bonsai 27B is the new multimodal flagship of the Bonsai family. Based on Qwen3.6 27B, it brings a new capability tier to local AI: multi-step reasoning, structured tool use, long-context workflows, and coherent agentic loops.

Until now, models in this class have been impractical to deploy locally. A 27B model occupies roughly 54 GB in 16-bit precision, and even a strong 4-bit build is around 18GB - too large for a phone and for most laptops.
Bonsai 27B changes that.

It comes in two variants:
• Ternary Bonsai 27B: 5.9 GB, 1.71 effective bits per weight, optimized for laptop-class quality.
• 1-bit Bonsai 27B: 3.9 GB, 1.125 effective bits per weight, optimized for phone-class footprint.

Everything is open-sourced today under the Apache 2.0 license.

[Avi Shefi @shefiavi]

@mattpocockuk Combine the three with a router?

[Avi Shefi @shefiavi]

@badlogicgames Does that replace all tools entirely, or added as an additional tool?

[Avi Shefi @shefiavi]

@RaminNasibov American Megatrends for life

[Avi Shefi @shefiavi]

@d4m1n You should build a skill for that

[Avi Shefi @shefiavi]

@mattpocockuk Stay hungry, stay foolish