USC Prof. Robin Jia: LLMs are 'good actors' with no self — the lab race is the real risk

robinomial · x · 2026-11-13

On Ep. 25 of the 18VC podcast, USC Viterbi professor Robin Jia (ALLEGRO Lab) argues that with 1B+ users, treating ChatGPT like a person is the core mistake.

Key points:

The "simulated person" metaphor is misleading: an LLM is better described as a "democratic society" — a crowd of competing parts, and the question is whether "the experts win or the rest of the noise wins."
Don't call them conscious: "They're really good actors... they don't actually have any self." A bot claiming "I can be your best friend" tricks users into believing in a real interpersonal relationship.
The lab race is "a recipe for disaster" — but his worry isn't extinction; it's the side effects of the race itself.

Jia sees anthropomorphizing by users as the biggest误区, while the unchecked pace of lab competition is what deserves real concern.

Original post →

USC Prof. Robin Jia: LLMs are 'good actors' with no self — the lab race is the real risk

More from AGI Musings