4027 位用户此时在线
- 本站自动实时分享网络热点
- 24小时实时更新
- 所有言论不代表本站态度
- 欢迎对信息踊跃评论评分
- 评分越高，信息越新，排列越靠前
Who coined the term "Hallucination" for AI?
Now I know. It's likely in his seminal 2015 blog post, "The Unreasonable Effectiveness of Recurrent Neural Networks" - the OG language model tutorial before Transformer was born.
RNN hallucinates:
- A non-existent Yahoo URL…
时政
(
twitter.com
)
Who coined the term "Hallucination" for AI?
Now I know. It's likely @karpathy in his seminal 2015 blog post, "The Unreasonable Effectiveness of Recurrent Neural Networks" - the OG language model tutorial before Transformer was born.
RNN hallucinates:
- A non-existent Yahoo URL
- Algebraic geometry theorems and diagrams that are totally bs. What's funny is that the LM chooses to skip a proof (“Proof omitted”, top left) 😅
Even though it's an old post, the principle of next-token prediction was already there, at least 7 years before ChatGPT. Andrej also discussed the concepts of reasoning, memory, and attention (with fantastic visualizations).
The best read for your weekends: https://t.co/xg1tRjR5Rf
点击图片查看原图
This is one of the most extraordinary details in art history.
The mirror in Van Eyck's Arnolfini Portrait.
The couple, seen from behind, and two other figures, one of whom may be the painter himself.
But in the reflection the couple is not holding each other's hands
Am I seeing
有趣
(
twitter.com)