Frankie Li

Build your own in-house AI lab with @bryel_labs (YC F26) | @fundamental @harvard @disney backed ex-founder

[Frankie Li @frankieli_]

@ChandlerSupple Not to mention all the other deals that we get, the value is a lot more than just the initial investment

[Frankie Li @frankieli_]

@janaksunil this is one of the most important benchmarks

[Frankie Li @frankieli_]

@unscriptedmanoj @KJHMiao @bryel_labs it's deployed

[Frankie Li @frankieli_]

@unscriptedmanoj @KJHMiao @bryel_labs have a fix for this, should be deployed later tonight/tmmw

[Frankie Li @frankieli_]

@BenjaminBekken the taste is immaculate

[Frankie Li @frankieli_]

@runtimesi congrats

[Frankie Li @frankieli_]

@dotpem The infra team at Typesafe

[Frankie Li @frankieli_]

@unscriptedmanoj @KJHMiao @bryel_labs weird, this is what I’m getting

can you try going to it in your regular browser?

[Frankie Li @frankieli_]

@zsakib_ @KJHMiao @bryel_labs 👀

[Frankie Li @frankieli_]

@iabdul_wasey @KJHMiao evals and scoping are so important

[Frankie Li @frankieli_]

@robdavitt who knows? something something Jevons paradox?

[Frankie Li @frankieli_]

Coinbase fine tuned a 9B model for under $100, and it beat Opus. A 9B model is so small, it fits on a $10 USB

[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.

[Frankie Li @frankieli_]

@Gabriele_333 @dangaron those will all be solved. Coinbase's first model cost <$100 to train.

So will Data/Evals/Envs one day 🥲

[Frankie Li @frankieli_]

@KJHMiao @bryel_labs 🤔

https://t.co/ltfzBm5lgE

[Brendan (can/do) @BrendanFoody]

Distillation as a service (DaaS) is the fastest-growing market in Silicon Valley

[Kevin Miao @KJHMiao]

What if you could train a model with a single prompt?

This week, Coinbase announced that they trained a 9B model that beat the frontier models in their domain while being significantly cheaper and 55% faster. It cost under $100 to train.

Today, we’re opening the same capabilities to you. Coming to you from the Bryel Foundry, our in-house tooling that we’ve used to train models for unicorns.

Just one prompt, a teacher, and within 7 days, you will receive a custom model that is up to 2x cheaper in a specific domain of your choice.

Book a demo at https://t.co/v3RDWb2MtZ

[Frankie Li @frankieli_]

@kartiksmath I think founders should be more concerned that customers are just going to build their own solutions vs frontier labs doing it

[Frankie Li @frankieli_]

@emollick had the same impression with Muse Wonder

[Frankie Li @frankieli_]

@OpenAINewsroom Detective Astra

[Frankie Li @frankieli_]

@julius_brussee @ycombinator perhaps the most epic video for the most epic product I’ve ever seen

[Julius Brussee 🪨 @julius_brussee]

Why use many token when few do trick?

Caveman is backed by @ycombinator (F26).

It started as a joke to get my AI bill under my grocery bill by having Claude speak like a Caveman. 111k GitHub stars later, it's a company.

Today: Caveman 4.0 and Caveman Cloud.

[Frankie Li @frankieli_]

@alokbishoyi97 @useRouterPlus extremely important problem