OpenAI's o1, aka. Strawberry

OpenAI's new model is out, and we are going to have to rake through a lot of muck to get the value out of this one!

⚠ Opt out of LinkedIn's GenAI scraping ➡️ https://lnkd.in/epziUeTi

• (00:00) - Intro
• (00:25) - Other recent news
• (02:57) - Hot off the press
• (03:58) - Why might someone care?
• (04:52) - What is it?
• (06:49) - How is it being sold?
• (10:45) - How do they explain it, technically?
• (27:09) - Reflection AI Drama
• (40:19) - Why do we care?
• (46:39) - Scraping away the muck

Note: at around 32 minutes, Igor says the incorrect Llama model version for the story he is telling. Jacob dubbed over those mistakes with the correct versioning.

Links relating to o1

• OpenAI blogpost
• System card webpage
• GitHub collection of o1 related media
• AMA Twitter thread
• Francois Chollet Tweet on reasoning and o1
• The academic paper doing something very similar to o1

Other stuff we mention

• OpenAI's huge valuation hinges on upending corporate structure
• Meta acknowledges it’s scraping all public posts for AI training
• White House announces new private sector voluntary commitments to combat image-based sexual abuse
• Sam Altman wants you to be grateful
• The Zuck is done apologizing
• IAPS report on technical safety research at AI companies
• Llama2 70B is "about as good" as GPT-4 at summarization tasks