Vitalik Buterin
@vitalik.eth
Grok 3 is super impressive, gave quite a detailed answer to this question. Much better than GPT o3-mini-high
36 replies
121 recasts
585 reactions
lotu
@thelotu
Nope. I definitely would not trust the data provenance, training or guardrails of this LLM.
1 reply
0 recast
0 reaction
Vitalik Buterin
@vitalik.eth
You might be surprised by the ideological leaning that it ends up having! (My hypothesis is that, like deepseek, it was trained on gpt outputs)
1 reply
0 recast
2 reactions
lotu
@thelotu
Yeah. I was somewhat surprised at some of the responses I’ve seen. But given the source, I would need more rigor & transparency. Sound hypothesis! I wish it was Claude.
1 reply
0 recast
0 reaction