Source:https://www.youtube.com/watch?v=ctcMA6chfDY

Increase Product Velocity

  • Old idea; a good executive is a busy executive.
    • New idea is that the engineers are busy and are the core decision makers, rather than relying on top down ideas.
  • Open AI has the opportunity to build the internet platform that will organize agents and their access of the internet.
    • Considers GPT (product) excellent due to the underlying model.

Compete Against Open AI

  • Their focus is to be the core model used by the world, and some of the tie-in products (GPT), but there will be API or SDK options for other organizations to build upon.

Capital Investment Round

  • Master Plan: “… try to make great models, and ship good products …”
  • Sam believes in tackling the problems in front of you, but doesn’t believe it’s effective to try and backwards plan a path to an ultimate goal.
    • He knows that AI infrastructure needs to be increased substantially
    • But he thinks Open AI’s competitive advantage is in being nimble and willing to pivot based on feedback.
    • Believes they’ll continue to make great models, and will make good products.
      • But they only focus on one or two steps at a time, not long master plans.
      • He thinks master planners never succeed.

Issues with Monolithic Companies

  • Every tech revolution the small companies disrupt the proven giants. (Think IBM)
    • Organizations get bogged down by their processes and lack the ability to adjust..
  • A couple of years left of the monoliths fighting against the change, and then mad scramble at the end to try to adjust, (but by then small startups will have won market share and in turn grow into monoliths themselves).
  • 20 year-old using GPT vs a 30 year-old.
    • Youth adapts faster
    • The youth use it as an operating system, and memorize long prompts that they use for patterned cases.
    • They use GPT as a personal advisor for major life decisions.
    • Older people use GPT as a google replacement
    • 30’s as an advisor
    • college students and younger use it as an operating system.

How is Open AI using it

  • “It writes a lot of our code.”
    • Doesn’t know the exact number.
    • Doesn’t think the number of lines matters, and is a dumb metric
      • “It’s writing meaningful code.” and thinks that’s what matters/should be measured.

Given consumer is main revenue center; Why keep API in 10 years?

  • Sam hopes that all AI agents and services will merge into “one thing”, with the ability to log in with Open AI and then use the new internet.
  • He thinks/hopes there will be a new protocol comparable to HTTP for the future of AI agents. (obviously wants Open AI in the middle of it. Are they making plays to be rent seekers?)

AI needs more data; Will it get sensor data?

  • People are already doing that for their own use cases, and new models are processing that data better than they ever have before.
  • Open AI will bake it in at some point, but not clear exactly when/which model.

Voice Model

  • They believe it’s vital, and are working on it, but acknowledge they’ve not done well enough yet.
    • Mentioned that chat took a while for them to crack too.
    • Thinks voice GUI is a barrier.
    • Voice will create a whole new class of devices.

Coding Focus for Open AI

  • They view it as a central aspect of the model, and they’ll continue to develop toward that direction.
    • They want to create the models that will take prompt input and deliver full programs “complete”
  • Their API will continue to assist developers, but they want GPT to be able to deliver a program.
  • Chat -> Agents -> full programs delivered

Underrated Ingredient to Future AI Models

  • The highest leverage is still algorithms improvement, data, and compute.

Coordinate Moonshot Projects

  • Some massive projects require top down leadership, but in general they operate as a research lab.
  • Before day one, they spent a lot of time researching and designing what a “well-run research lab looks like.”
    • They had to go really far in the past.
      • As in nearly everyone who was last involved in a good research lab was actually dead, and not available for direct input.
  • They follow the principles they found in the historical great labs.
    • Labs that are diverging continue to fall behind/fail.

Humanities Breakthroughs

  • They have academic research departments they partner with to assist.
  • But their main breakthrough is in making the model ubiquitous and as cost efficient as possible. Thus democratizing access, for anyone who wants to conduct the research on their own.
    • Are we entering a new enlightenment where the discoveries are available to anyone with the curiosity to explore?
  • “90% thrust vector” is focused on improving the models and delivering access to everyone.

Platonic Ideal State

  • Tiny reasoning model with a trillion tokens of context.
    • The model never updates, but it’s able to recall everything you put into across your entire life and reasons with you.
      • Would cover every conversation of your life, every book you read, every data input from any source (film, audio, etc…), and life continue to append indefinitely.

Value Creation 12 Month Horizon

  • Building infrastructure
  • Better Models
  • API or method to integrate models
  • Sam thinks 2025 is the year that agents start doing a predominant amount of work; especially around coding (a few others too)
  • 2026 AI makes novel major discoveries in the sciences.
    • Science discovery is the real driver of economic growth as entrepreneurs/businesses discover methods of implementing those discoveries for market use.
  • 2027 is where robots create and drive value.

Closing Thoughts

  • O3 is likely as smart as or smarter than the audience.
  • Resilience gets easier over time.
    • Challenges get harder, but the emotional toil gets easier.
  • The challenge isn’t in the moment of the event.
    • It’s in managing the fallout after the challenge is overcome.
    • Crisis control is valuable, but dealing with aftermath is the vital skill. Day 60 after is the real challenge.