While the world was focused on Jensen’s life, AMD secured itself a deal to put literal hardcoded AI models in every phone, laptop and data center. An LLM becomes actual hardware and soon we’ll be talking about upgrading your laptop’s built in AI model the same way you swap RAM.
1
63
Taalas’ “brain on a chip” model works for two very good reasons:
1. We’re hitting diminishing returns on bigger models, the jump from one version to the next is no longer “felt”
2. Most tasks, especially for embedded hardware don’t require a crazy amount of intelligence to start with
and so faster inference on a single really good model is exactly the metric to be optimizing for.
42
Rescript is now at ★ 1,000+ Github stars 🤯
I built this as an open-source Descript alternative, it took a single weekend with Fable for the initial version!
Github: github.com/wassgha/rescript
Try it: getrescript.com
Rescript is a free, open-source podcast editor that lets you edit videos the same way you edit text. Unlike Descript, it runs fully offline, right on your device with nothing sent to the cloud.
1
8
292
starting over >>>> adapting pre-ai software
i respectfully disagree
i am actually quite bullish on underdog software startups, and feel fortunate to get to serve thousands of them at greptile.
here is what we see:
- ambitious startups have never been more leveraged. with coding agents they can punch far above their headcount. labs are competing to provide increasingly more value at lower prices.
- large software companies are being forced to step up their game in response. the best ones are painstakingly refactoring their engg orgs to adapt to a new way of building software
like many of the best tech disruptions, this one shifts power from incumbents to upstarts
100
Robots are now an API call. We are about to witness the physical equivalent of the “chatgpt wrapper” startup archetype, hundreds of companies will be specializing robots to various tasks and it’s freaking exciting.
Over the past month, we completely switched to frontier LLMs to control a robot in a customer deployment.
By doing this, we have a classic System 1/System 2 architecture, but on steroids.
For the high-level reasoning layer, we use an agent based on Opus and Astra, which are many generations more capable than any VLA. For the fast execution layer, we use computer vision code written by the model, which is fast, editable, and reliable.
As a bonus, we got in-context learning, which is just a basic feature of LLMs, and continual learning, since whenever a new situation arises, the large model steps in and adapts the pipeline.
And it just feels better to be on the right side of the Bitter Lesson, benefiting from the billions being invested in frontier models around the world.
2
3
325
Wassim retweeted
Over the past month, we completely switched to frontier LLMs to control a robot in a customer deployment.
By doing this, we have a classic System 1/System 2 architecture, but on steroids.
For the high-level reasoning layer, we use an agent based on Opus and Astra, which are many generations more capable than any VLA. For the fast execution layer, we use computer vision code written by the model, which is fast, editable, and reliable.
As a bonus, we got in-context learning, which is just a basic feature of LLMs, and continual learning, since whenever a new situation arises, the large model steps in and adapts the pipeline.
And it just feels better to be on the right side of the Bitter Lesson, benefiting from the billions being invested in frontier models around the world.











