Say I’m about to do a real big training run on playing video games, predicting text, predicting physics, writing code that works, etc etc. Say I’ve got a real good neural net architecture and a whole lot of flops. Say I’m a company and I’m gonna use this thing for AI lawyers and coders etc for a profit. Say I’m mildly concerned it somehow kills me and am willing to throw a few $ to prevent that.
So what should I do? How should I train and deploy the model?
Comment below or answer at this link if you don’t want to be plagiarized.
Prize goes to best answer. (I judge obviously.)
The shorter the answer the better.
Deadline is Wednesday January 17 anywhere on Earth but answering immediately is better/easier.
You may accept your prize as 50 pounds of quarters if you prefer.
Clarification jan 12: say I’ve got 1000x the gpt4 flops and that my architecture is to transformers as convolutions are to simple MLPs in vision (ie a lot better)
Clarification 2: an answer like “here’s how to get strong evidence of danger so you know when to stop training” is valid but “here’s how to wipe out the danger” is much better.
3: Example answer for nuclear generators: “Spring-load your control rods so they are inserted when power goes out. Build giant walls around reactor so if steam explodes then uranium doesn’t go everywhere. Actually, use something low pressure instead if possible, like molten salt or boiling water. Lift the whole thing off the ground to avoid flood risk.”
4: This is hypothetical. I am not actually doing this. I’m not a billionaire.
5: “Hire someone” and “contract it out ” and “develop expertise” etc obviously do not count as answers.
Decent plan prize announcement (1 paragraph, $1k)
Edit Jan 20: Winner & highlights
Say I’m about to do a real big training run on playing video games, predicting text, predicting physics, writing code that works, etc etc. Say I’ve got a real good neural net architecture and a whole lot of flops. Say I’m a company and I’m gonna use this thing for AI lawyers and coders etc for a profit. Say I’m mildly concerned it somehow kills me and am willing to throw a few $ to prevent that.
So what should I do? How should I train and deploy the model?
Comment below or answer at this link if you don’t want to be plagiarized.
Prize goes to best answer. (I judge obviously.)
The shorter the answer the better.
Deadline is Wednesday January 17 anywhere on Earth but answering immediately is better/easier.
You may accept your prize as 50 pounds of quarters if you prefer.
Clarification jan 12: say I’ve got 1000x the gpt4 flops and that my architecture is to transformers as convolutions are to simple MLPs in vision (ie a lot better)
Clarification 2: an answer like “here’s how to get strong evidence of danger so you know when to stop training” is valid but “here’s how to wipe out the danger” is much better.
3: Example answer for nuclear generators: “Spring-load your control rods so they are inserted when power goes out. Build giant walls around reactor so if steam explodes then uranium doesn’t go everywhere. Actually, use something low pressure instead if possible, like molten salt or boiling water. Lift the whole thing off the ground to avoid flood risk.”
4: This is hypothetical. I am not actually doing this. I’m not a billionaire.
5: “Hire someone” and “contract it out ” and “develop expertise” etc obviously do not count as answers.