Twitxy Saved

Igor Kulakov (MicroFactory)

Verified account This account is verified. Learn more

@ihorbeaver

Making physical products too cheap to meter
San Francisco, CA microfactory.com Joined November 2007
760 Following 10K Followers 1.4K Posts

Latest posts 5

Igor Kulakov (MicroFactory) Verified account @ihorbeaver
Follow me for tips on how to use it in production, since we’ve had it deployed for a month now: x.com/ihorbeaver/sta…
Claude Verified account @claudeai Nov 9
Fable 5.2 shows substantially improved skills in controlling physical robots, even on tasks it has never explicitly been trained to do.

On RoboDojo, a benchmark with a simulated two-arm robot, it outperformed other models and most AI systems built to control robots.
View quoted post on X
1 replies 3 reposts 63 likes 3.6K views View on X
Igor Kulakov (MicroFactory) Verified account @ihorbeaver
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.
1:37