Will AI Make Humans Extinct? Did GPT-4 Get Worse? & OpenAI's GPT-4 Roadmap Revealed | E17
This week we cover the Center for AI Safety's Statement and ask, "Will AI Really Make Humans Extinct?" We cover the discussion on GPT-4 Deteriorating: is it really getting worse? And discuss some likely causes.
We also use the Internet Archive to discuss OpenAI's GPT-4 roadmap including a stateful API, giant context sizes, plugins having no product market fit and why OpenAI is constrained by GPUs.
We also cover the lawyer who got caught using ChatGPT and the problems with hallucinations. Will step-by-step rewards help solve these problems? Do we need better warnings for stupid lawyers?
If you like this episode please consider subscribing, comment and liking to help others find our podcast. We appreciate your support.
00:00 - Is GPT-4 Getting Worse?
00:20 - Center for AI Safety Statement: Will AI Make Humans Extinct?
12:09 - Is GPT-4 Getting Worse? Is Alignment the Problem?
29:27 - GPT-4 Roadmap: 1M Context? Multi-modal?
36:11 - Vision of ChatGPT: Have Plugins Failed?
38:02 - Will OpenAI API's Have Too Much Competition?
40:21 - Giant AI Models Aren't Over: Is GPT-4 Much More Capable?
44:39 - ChatGPT for Lawyers: What Could Possibly go Wrong?
48:17 - The Hallucination Problem
53:22 - Process Supervision: Solution to Hallucination?
55:59 - Japan Removes Copyright Laws for AI Progress
57:53 - UAE's Falcon 40B Open Source & Royalty Free!
1:01:04 - Will We Value Written Language?