Amjad Masad
4 tracked predictions · 0 hit · 0 partial · 0 miss · 4 still open
open Amjad Masad: Replit will ship 'Agent 4' roughly six months after Agent 3, with each new agent version timed to the next jump in underlying LLM capability.
“we already know what agent 4 is going to look like”
Your call:
open Amjad Masad: As AI agents make it cheap for an agent to hire a human or another agent directly, large companies will shrink and society will see much faster boom-and-bust cycles of company formation and death.
“if an agent can go hire human or can go hire another agent, then we're going to see companies shrink”
Your call:
open Amjad Masad: The most capable future LLM will likely need only about 1 billion parameters, far smaller than today's frontier models, as AI continues compressing most of human knowledge into an efficient core.
“I think the ultimate LM will probably be about a billion parameters”
Your call:
open Amjad Masad: Replit will likely resume training its own foundation models, as it did with its 3-billion-parameter Replit Code model in 2022-23, now that it has a large unspent funding war chest and specific agent-architecture needs.
“I think we're reaching a point where it makes sense to go back into training”
Your call: