Spark of AGI, a new paper from Microsoft, takes us through over 100 pages of GPT-4 experiments to argue that it shows early signs of artificial general intelligence. The results are impressive, but there are also still clear areas for improvement and indications of the limitations of LLMs like GPT-4.
Outline
0:00 - Intro
1:01 - Experiment Speed Run 1
3:51 - ClearML
4:54 - GPT-4 Flaws & Math
8:31 - Tool Usage
11:13 - Experiment Speed Run 2
13:12 - Limitations
19:03 - AGI
ClearML - https://bit.ly/3GtCsj5
Social Media
YouTube - / edanmeyer
Twitter - / ejmejm1
Sources:
Paper - https://arxiv.org/pdf/2303.12712.pdf