alt Hacker NewsMCP apparently reduces probabilistic failures - aka the common fatal flaw of all of these robots (hallucinations, missing stuff in the API doc, etc).
This makes it a little more interesting to me, knowing those results.
It definitely underlines what we already know about the specific weaknesses of LLMs replies/results.
Replies
latchkey • today at 1:23 AM
I think that was a problem 6 months ago, but GPT 5.6 Sol on xhigh doesn't have those sorts of issues. I don't think it'll last. Things are moving fast.
➕ show 2 replies
>MCP apparently reduces probabilistic failures
Does it really do that much difference? I mean in the end robots process MCP output the same way they would process API docs.