brypto

I’m behind on this AI agent stuff.

What is the bull/bear case? Is there a real benefit to making a verifiable agent? (Eg you can verify output is correct wrt model)

Working on building zkTLS infra/products. A ruthless empiricist and meme connoisseur.

/opacity
/eulerlagrange

I think where verifiability would help is to be able to assert that the agent is acting autonomously, as opposed to each of its actions being guided and/or vetted by a human before actually posting. Some attempts exist at running them in TEEs for this reason but it's not perfect.