On Thursday, March 23, 2023, at 9:42 PM, Mike Archbold wrote: > Perhaps the only real test of understanding is whether one can produce new knowledge, such as proving new mathematical theorems, a feat that currently remains out of reach for LLMs."
We have already seen, in posts I shared months back, that DALL-E 2 can uncrop images to see what is around the images fairly accurately in the first inch and somewhat as you keep going an arms length. No matter if its a mathematical problem or an image to uncrop, both depend on context being "recognized" by memories or rules it knows of. It doesn't matter if math needs high precision, or not, what matters is both need something to find the answer, a memory recall. As long as DALL-E 2 can do it somewhat accurately, it's obvious GPT-4 generates new knowledge too, somewhat accurately, to some level of difficulties. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T208c0ae6f6f8ee11-M906eee1cd1ae4c448ef10516 Delivery options: https://agi.topicbox.com/groups/agi/subscription
