OpenAI’s Head of ChatGPT: We’re entering a new era of AI (again) | Tibo Sottiaux
Tibo Sottiaux, head of ChatGPT and Codex at OpenAI, discusses the future of AI agents, the shift from complex workflows to learning systems, and OpenAI's approach to safety. He shares insights on skills trending up in the AI era and the vision for an open AI ecosystem.
Deep Dive Analysis
18 Topic Outline
Anticipating the Future of AI Agents
Evolution of AI Workflows and Agent Teams
The Vision for Always-On AI Assistants (Dots)
OpenAI's Open Ecosystem and Plugin Strategy
The Research and Safety Behind Dots
Human Value in the AI Era
Addressing Agent Fatigue and Loneliness
AI's Role in Reducing Noise and Fostering Creativity
An AI Agent Spotting a Production Outage
Guardrails and Multi-Device AI Control
Trending Skills in the AI Era
Advice for Early Career Professionals
OpenAI's Internal Building and Shipping Culture
Autonomy and Trust at OpenAI
Unpriced Changes and Second-Order Effects of Agents
Tibo's Mindset Shifts on AI Capabilities
OpenAI's Approach to AI Safety and Alignment
The Goal of Simplicity in AI Interfaces
5 Key Concepts
AI Agents
Permanent, active intelligences that work 24/7, understand user goals and preferences, learn from feedback, and are available across various devices and clients without constant configuration.
Dots
OpenAI's new personal AI assistant platform designed to be an incredibly smart, always-on agent that learns user preferences and goals, with plans to integrate its capabilities into ChatGPT.
Open Ecosystem (OpenAI)
A strategy to open up OpenAI's infrastructure and capabilities, including sign-in with ChatGPT, plugin extensions, and revenue sharing for popular plugins, to foster broader community development and distribution.
Pacing the Frontier
OpenAI's approach to AI development that involves investing heavily in alignment, safety, security, and guardrails for models, including significant compute for secondary monitoring, before releasing advanced capabilities.
Model Pickers
The current need for users to select specific AI models (e.g., multi-agent, ultra) for different tasks, which OpenAI aims to eliminate for a simpler, more intuitive user experience.
8 Questions Answered
The majority of actions on the internet will soon be taken by AI agents, requiring products to be built for a high level of scale and seamless agent interaction.
The future involves permanent, active AI intelligence (like Dots) that knows user goals and preferences, is available across any device, and gets out of the way when not needed, freeing users from being tied to specific technology.
Plugins are discovered and recommended based on retention numbers, quality, and how successfully they add utility, not just on initial downloads or keywords, encouraging developers to build genuinely useful tools.
Human brains will remain most valuable in areas like taste, creativity, and the will to build, as AI is designed to be an extension of human capabilities, fostering more builders rather than replacing human ingenuity.
AI aims to reduce configuration fatigue and make interactions more delightful by integrating into physical spaces, allowing agents to observe and collaborate naturally with humans and other agents, reducing screen time and fostering engagement.
Great taste, user-centric thinking, connecting with the audience, and a passion to build things that matter are trending up, while skills like fast typing are becoming less essential.
OpenAI paces the frontier by investing heavily in alignment, safety, security, and guardrails for models, including significant compute for secondary monitoring to prevent high-risk actions or prompt injections.
The complexity of model pickers and the need for users to understand which specific model to use for different tasks is a major annoyance that OpenAI aims to eliminate for greater simplicity.
10 Actionable Insights
1. Build for Agent-First Internet
Design products and services with the expectation that the majority of internet actions will be taken by AI agents, requiring a focus on scale and seamless integration to be successful.
2. Prioritize User Taste and Creativity
Focus on building AI as an extension of human will and taste, empowering users’ creativity rather than replacing it, as this is where human value will remain most useful over time.
3. Embrace Bottoms-Up Innovation
Foster a culture where small teams can quickly hack on exciting ideas, make them accessible internally, and iterate rapidly, allowing unstructured energy to lead to productive product launches.
4. Cultivate Kindness and Collaboration
For early career success, prioritize being incredibly kind, collaborative, and a fast learner, always focusing on solving important problems in a way that doesn’t put personal gain first.
5. Focus on Plugin Retention for Discovery
To get plugins discovered and recommended in an open AI ecosystem, build high-quality plugins that demonstrate strong user retention and actual utility, rather than relying on superficial metrics.
6. Strive for Simplicity in AI
Aim to eliminate complexity like model pickers and multi-agent configurations, moving towards a single, intelligent system that learns user preferences and goals for a seamless experience.
7. Push for 10x Improvement Mindset
When building with AI, imagine the technology being roughly 10 times better than it is today in a year and build in a different way, rather than just incrementally improving current capabilities.
8. Leverage AI for Analysis
Use AI tools like Codex to automate code generation for data analysis, understanding business trends, feature performance, and launch success, freeing up time from manual coding.
9. Disconnect for Enhanced Creativity
Recognize that fully disconnecting from constant digital demands, such as during holidays, can foster more creative and different ways of thinking, and AI should aim to enable this in daily work.
10. Prepare for Agent Traffic
If building a product, anticipate a significant increase in traffic from AI agents and prepare your systems and economics for that scale and interaction, as it is inevitable.
7 Key Quotes
The majority of actions on the Internet will be taken by agents.
Tibo Sottiaux
Over time, you just want a system that learns.
Tibo Sottiaux
It's all about breaking free from the technology and having this sort of like permanent active intelligence that knows, you know, everything it needs to do.
Tibo Sottiaux
We may not have coders anymore, but we have more builders than ever.
Tibo Sottiaux
I think the next Tebow is already at OpenAI. His name is Ahmed Ibrahim.
Tibo Sottiaux
You get a lot of autonomy and then, you know, you have to just own up to it.
Tibo Sottiaux
I just want to get rid of all of that as quickly as we can.
Tibo Sottiaux