All-In E178: In conversation with Sam Altman
Episode 178 • 5/10/2024
Key Discussion Points:
GPT-5 and Model Development
- Sam indicated GPT-5 might not be released as a numbered successor, suggesting a shift toward continuous model improvements rather than discrete versions
- Discussed how GPT-4 has improved significantly since release
- Focus is moving toward making advanced AI technology available to free users, though cost remains a major barrier
AI Infrastructure and Computing
- Addressed challenges around compute costs and latency
- Discussed the need for massive infrastructure investment for AI development
- Sam mentioned algorithmic improvements could effectively double compute capacity
Open Source vs Closed Source
- Sam explained OpenAI's stance on keeping models closed source while supporting some open source initiatives
- Expressed interest in open source models that can run effectively on phones
- Acknowledged the value of both open and closed source approaches in the ecosystem
Future of AI Interfaces
- Discussed vision for AI as an "always-on super low friction" assistant
- Emphasized importance of AI systems that can push back and reason, rather than just follow commands
- Explored potential for new device form factors beyond phones
November 2023 OpenAI Drama
- Sam shared details about his firing via text while in Las Vegas
- Described the chaos of the five-day period and eventual resolution
- Expressed respect for board members' commitment to AI safety while disagreeing with their decisions
AI Regulation
- Advocated for international oversight of frontier AI systems
- Supported focus on safety testing rather than source code review
- Expressed concern about both over-regulation and under-regulation
Notable Quotes:
On evolving AI models: "What I will say is a lot of people have noticed how much better GPT4 has gotten since we've released it... I think that's a better hint of what the world looks like."
On AI regulation: "I think there will come a time in the not super distant future where the frontier AI systems are capable of causing significant global harm."