farhan

@columbia
@awscloud @janestreetgroup

[Brian Armstrong @brian_armstrong]

We were able to fine tune an open weight model (Qwen3.5-9B) on proprietary internal data at @coinbase, and it out performed frontier models, at much lower cost and speed, for a specialized task (fraud prevention in this case).

I think we'll see more of this pattern inside companies. The frontier models are amazing generalists. But increasingly you'll see specialist models out perform on more narrow tasks, either working on teams, or by themselves.

We're making it easy to fine tune models internally for our employees. We even have a skill for it in our internal harness that makes it easy to spin these up and start training them:

"/fine-tune-and-serve-llm I want to fine-tune Qwen for payments risk review. help me get started"

I imagine we'll have some backend services that could be replaced entirely by this new architecture.

[farhan @farhansadeek]

@AdamHoltererer I think the Claude $20 plan usage limit is too low for me. I would take the OpenAI $20 over Claude $20 plan.

[farhan @farhansadeek]

@askalphaxiv Finally

[farhan @farhansadeek]

@OpenAI @RasmusLindahl6 time to add this to https://t.co/b9EwYpOuXC?

[farhan @farhansadeek]

@josue_gdm i started to write a good amount of code by hand recently, there is a different level of satisfaction in that can't lie

[farhan @farhansadeek]

@ranvirdeshmukh_ @ycombinator Good luck!

[farhan @farhansadeek]

@ranvirdeshmukh_ @RealPact damn cortisolmaxxing @wongm6216 and @ranvirdeshmukh_

[farhan @farhansadeek]

@deepgrove_ai incredible work team!

[DeepGrove @deepgrove_ai]

We introduce Maple-Preview, an open-source 20B-A1B ternary-weight reasoning LLM, SOTA in its weight class.

It solves IMO-level problems and runs at 200+ tokens/s on a Mac Mini M4, 5–16× faster than efficient models like Gemma 4, Qwen3.5, and gpt-oss.

https://t.co/l2y5eZTVS4

[farhan @farhansadeek]

@FactualCentrist @alexandr_wang I agree output token wise it's definitely a point. I meant from a solely intelligence perspective.

[farhan @farhansadeek]

@ESPNFC I'm sorry but Messi, Mbappe and Bellingham were far better than him. Just because his team won the WC doesn't mean he deserves the golden ball.

[Eddie Bae @EddieBae7]

Excited to have won @OwkinScience Rewire Biology Hackathon this weekend in SF. We built K Scope an MCP server on top of Owkin’s AI scientist platform K Pro. We allow K Pro’s orchestrating LLM to access model internals through mechanistic interpretability tools like probes, sparse auto encoders, and activation patching. This allows these AI scientists to not only get inputs/outputs from these ML models but truly understand the causality of how these black box models reason. We incorporated Owkins open source models like their pathology ones: H Optimus and Phikon. Was a ton of fun and very thankful for my teammates and all the passionate AI x bio people I met.

Below is the repo link and our demo. Please check it out!

https://t.co/9tPGPuelc9

[farhan @farhansadeek]

@KollaikalRupesh @treenalabs @camden_hulse @ruomengsyn coding 24/7 is no longer a dream, it's a reality lol

[farhan @farhansadeek]

@founderleague @erik_petersons @wongm6216 @issaodeh23 Cortisonmaxxing at finest @wongm6216

[farhan @farhansadeek]

@carpetxie Where are the Labubus?

[farhan @farhansadeek]

@connor1 https://t.co/TMbTyRg6xy

[Emir Karabeg @emkara]

Fullstack comes to an end. The first ever @simdotai hackathon with @Lovable, @supabase, @stripe, @brexHQ was a major success.

@aaron_epstein ended up giving interviews to the top 3 teams! Thanks @ycombinator and congrats to the winners. Keep shipping 🚀

[farhan @farhansadeek]

@deedydas It doesn't say if they went for Undergrad or Graduate School. Also school size matters a lot; a place like Berkeley would has a ton more students than say Princeton so comparing them won't be a good measure of how good the CS programs are (both are world-class).