4 ms·
> Is there a physical law which will limit AI progress? Probably the same law that limits compute power wrt to space and energy. Look at how much energy goes
by dartos 2y ago
> Is there a physical law which will limit AI progress?
Probably the same law that limits compute power wrt to space and energy.
Look at how much energy goes into training GPT 4.5 vs it’s improvements (on our own potentially bunk benchmarks, granted)
Also like… literally nothing in the universe advances exponentially forever.
Whether or not any of that is valid, I’d like an answer to my question.
Is the only reason you think that a blind belief that AI will advance exponentially in all domains?
- nthingtohide 2y agoIf AI is the general purpose pattern recognition algorithm and real world has patterns then as long as data i.e. real world has juice, AI will do better. Recursive self improvement is included in the domain. The only constraint will be how long it takes to gather feedback from the real world when AI does experiments.
- dartos 2y agoI’m sorry, but I didn’t understand what you meant on and after the word “juice” I’m not sure any of that was relevant to what I was saying. And you still didn’t answer my question.
- nthingtohide 2y agoI am only doubtful about the exponential part. I am hopeful that yes all fields will ultimately subsumed by AI.
- dartos 2y agoI believe exponential (as opposed to logarithmic) advancement would be needed for that to happen. If the improvement curve is linear, it would be prohibitively expensive as and if logarithmic, it’d be impossible.
- nthingtohide 2y agoAnother point in your favour is the problem space itself will become exponentially large. Imagine getting a requirement to synthesize a material with 40 properties rather than 1 or 2.