I recently read about Anthropic’s announcement about Model Hardware Standard, for connecting AI agents to physical hardware.
Today, we're kicking off the first phase of the research preview for Model Hardware Standard (MHS): a new standard for AI agents to safely operate physical equipment in scientific research and advanced manufacturing.
Read more: anthropic.com/news/model-har…
And while the demos are interesting, I am struggling to see what’s so special.
Just like language models, given a set of variable inputs such as microscope settings and a target output like the right focus, obviously an LLM like Claude can predict and iteratively get its way there.
And yes, just like LLMs can and will discover new scientific theorems, drug molecule candidates etc. because of the sheer scale and width of its training knowledge base.
Interesting, but isn’t a a general-purpose model overkill here? Wouldn’t a cheaper, specialized smaller and hopefully open model/standard do better. Kimi-Science anyone?
General-purpose robotic models will be solved on the edge, with native embodied foundation models consuming visual, language, proprioceptive and sensor inputs producing 100 Hz action trajectories like GEN 1-1.5 from Generalist. A general-purpose LLM ? Ermmm.
We've reduced the time it takes to go from physical prompt → robot behavior.
The faster anyone can teach a robot to do something new, the easier it becomes to scale physical work.
Read more about GEN-1.5 in our blog post in the comments below.
Join the conversation