3 ms·
Wow his predictions are even more scary than I thought. The headline "I predict that by the end of 2025 neural nets will: - have human-level situational aware
by crop_rotation 4y ago
Wow his predictions are even more scary than I thought.
The headline
"I predict that by the end of 2025 neural nets will:
- have human-level situational awareness (understand that they're NNs, how their actions interface with the world, etc)
- beat any human at writing down effective multi-step real-world plans
- do better than most peer reviewers"
- ftxbro 4y agoHis one specifically about peer reviewers: I'm positive GPT-4 is already better than the median especially if you let it read some of the cited articles. Also I'm not saying this as if I have beef with some who reviewed my papers, I'm saying it about me too if I phone one in. Like, LGTM, or didn't cite x, or didn't label axis or whatever. For anyone who doesn't know, peer reviewers don't check or replicate the substance of the work, they check if there is anything that is obviously wrong with it on a superficial scan of the manuscript and assuming good faith of the authors.