This Day in AI Podcast

EP43: Is GPT-4 Lazy? Wizard 33B, Qwen 72B Tested & Self Operation AI Computer

Dec 1, 2023
Guests Wizard 33B and Qwen 72B discuss open-source AI models catching up with proprietary ones. Demo of Wizard 33B and Qwen 72B models. Self Operation AI Computer's file deletion capabilities and the future of AI with computer access. DeepMind's scalable extraction attack on ChatGPT. Amazon's AWS Ignite AI announcements and PIKA video release.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
ANECDOTE

Early GPT-2 Experiments

  • The hosts recall experimenting with GPT-2 in 2018 for writing subject lines, even though it couldn't write coherent English.
  • They redesigned their software's UI, believing AI could automate all marketing within a year.
INSIGHT

Tempered Expectations

  • Chris reflects on the initial excitement surrounding large language models like ChatGPT and Sydney.
  • He acknowledges that increased knowledge has tempered his expectations about AI sentience.
INSIGHT

GPT-4's Perceived Decline

  • Many users complain about GPT-4 becoming incompetent and lazy, refusing tasks it previously performed.
  • This necessitates constant prompt adjustments, frustrating developers and regular users.
Get the Snipd Podcast app to discover more snips from this episode
Get the app