3 ms·
Interview with Sebastien Bubeck: Sparks of AGI: Early Experiments with GPT-4
- ftxbro 4y agomy guy's last name was first alphabetically on his paper is that why he was invited to speak #alphabetism https://news.ycombinator.com/item?id=35550591 https://news.ycombinator.com/item?id=35550591
- ftxbro 4y agoIt's interesting that his first slide is titled "First Contact" and the commented out title in the tex source of his 'sparks of agi' paper was "First Contact with an AGI System" https://news.ycombinator.com/item?id=35304090 https://news.ycombinator.com/item?id=35304090
- pulvinar 4y agoIn the section "GPT4 cannot do true planning" (42:40), he asks it to change a math identity in a way that it fails to do, whereas he shows how he can do it with a little planning ahead. But this really isn't a fair comparison. If you tell GPT4 to show its reasoning before giving the final answer (i.e., to plan ahead), it also gives the correct answer. It seems that the major difference is that we usually "don't show our work"-- our inner monologue is private.