Jacob Coxon 在 Y Combinator 的辯論中提到,Anthropic 研究人員回報,AI 已讓研究工作加快約 4 到 5 倍;他也提醒,這是人員自行回報的數字,可能不太準確,甚至可能低估實際效果。
- 照目前約每 3 個月加快一倍的趨勢推算,到 2027 年可能達到 100 倍。他另提到,Anthropic 目前約有 25% 的研究透過 Claude 進行。
- 若趨勢持續,Claude 到 2027 年底可能負責全部研究。這些都是依趨勢推算的可能發展,顯示 AI 可能越來越多地參與 AI 研究。
AI 伴隨問答
針對這篇報導有疑問?可直接點選快捷問題或向 AI 追問背景脈絡與技術細節:
> Anthropic researchers are gaining a 4 to 5x AI speedup, doubling every 3 months, which puts 100x in 2027
-Jacob Coxon, ex-Anthropic and OpenAI researcher
> Claude will be doing all of Anthropic’s AI research by late 2027
> 25% of Anthropic’s research already runs through Claude. On trend, it’s 100% by late 2027
~ Jacob was at Y Combinator’s first YC Debate, "Should We Slow Down AI?"
"You can also look at the trend of self-reported speed up. Humans can report how much they're getting a speed up from AI. It's currently around 4 to 5x reported at Anthropic. That's maybe an underestimate. The report is maybe untrustworthy, but it's doubling every 3 months roughly. If you extrapolate that forward, you get a 100x 2027, which sounds a bit like an automated AI researcher."
----
Full video on "Y Combinator" YouTube channel, (link in comment)