8 ms·
It’s funny, because while technically correct, you may end up fundamentally wrong that there was a limit to the value we might find in parallel compute. After a
by dougmwne 3y ago
It’s funny, because while technically correct, you may end up fundamentally wrong that there was a limit to the value we might find in parallel compute. After all, we may have lie of sight to AGI through parallel scale alone.
- eslaught 3y agoWhat you've identified is known as Gustafson's Law. If you can keep solving larger and larger problems, your solution time doesn't necessarily go down, but "speedup" is potentially uncapped. Of course there may be some practical limit to Gustafson's Law, but I don't think we've found it yet, at least in many scientific domains.