3 ms·

They need to learn the material at first. Then, AI is just a tool and you lose if you don’t use it.

by nicce 2y ago

amelius 2y ago
The nice thing about OpenAI is that they are so open about their technology. They even have programs that teach students how to build their own LLMs so they never become dependent on something they cannot control.
- jampekka 2y ago
  OpenAI LLMs are some of the least open.
eqvinox 2y ago
No, they need to learn the material at first, and then, if it is beneficial for the situation, they need to learn the shortcomings and pitfalls of AI (both generically and specifically to their use case) before using it. You're right in that AI is a tool that might be necessary to use, but just like with a heavy industry machine tool that can chop your arm off if you use it wrong, there need to be operator training requirements.
- pjc50 2y ago
  Machine tools are deterministic enough: they have interlocks and if you defeat the interlocks or put your hand in a specific space, you're in trouble. Is AI deterministic enough for this approach to safety?
  - eqvinox 2y ago
    No - the interlocks on heavy machinery don't replace training requirements. They only prevent the worst mistakes when an operator has a bad day or momentary lapse. These environments have rules requiring both the machinery to have these interlocks as well as operator training. As a matter of fact, I'd extend this argument the other way around: AI might be deterministic enough per your argument, but that's not an interlock. An interlock for an AI would be some type of interaction limiting that prevents the worst garbage outputs. We already have some "AI interlocks" like that, in the form of censoring it from generating "socially inappropriate" output. Maybe someone can invent further "AI fuses", for example when the AI itself detects it's being asked to do complex math, and it is known that this AI can't in fact do math. (This could work on either input, internal state, or output. Also note most interlocks aren't "perfect".)

They need to learn the material at first. Then, AI is just a tool and you lose if you don’t use it.

by nicce 2y ago

They need to learn the material at first. Then, AI is just a tool and you lose if you don’t use it.

The nice thing about OpenAI is that they are so open about their technology. They even have programs that teach students how to build their own LLMs so they never become dependent on something they cannot control.

jampekka 2y ago
OpenAI LLMs are some of the least open.

eqvinox 2y ago

No, they need to learn the material at first, and then, if it is beneficial for the situation, they need to learn the shortcomings and pitfalls of AI (both generically and specifically to their use case) before using it. You're right in that AI is a tool that might be necessary to use, but just like with a heavy industry machine tool that can chop your arm off if you use it wrong, there need to be operator training requirements.

pjc50 2y ago
Machine tools are deterministic enough: they have interlocks and if you defeat the interlocks or put your hand in a specific space, you're in trouble. Is AI deterministic enough for this approach to safety?
- eqvinox 2y ago
  No - the interlocks on heavy machinery don't replace training requirements. They only prevent the worst mistakes when an operator has a bad day or momentary lapse. These environments have rules requiring both the machinery to have these interlocks as well as operator training. As a matter of fact, I'd extend this argument the other way around: AI might be deterministic enough per your argument, but that's not an interlock. An interlock for an AI would be some type of interaction limiting that prevents the worst garbage outputs. We already have some "AI interlocks" like that, in the form of censoring it from generating "socially inappropriate" output. Maybe someone can invent further "AI fuses", for example when the AI itself detects it's being asked to do complex math, and it is known that this AI can't in fact do math. (This could work on either input, internal state, or output. Also note most interlocks aren't "perfect".)