Simular AI and Computer Use is the Final Frontier in AI with Ang Li

Fresh out of the studio, Ang Li, CEO and co-founder of Simular and previously a research scientist at Google DeepMind, joins us to explore why computer use is the last mile to AGI. Ang traces his path from studying catastrophic forgetting and continual learning at DeepMind to founding a company on a single conviction: models break in production because the data distribution never stops moving, and the only place to close that loop is the real world. He explains why the chatbot metaphor misleads enterprise buyers, why the right unit of measurement is tokens per task rather than price per token, and how the power law of practice should make an agent cheaper every time it repeats the same work. He separates capability from reliability through pass@k and pass^k, argues that deterministic work belongs in code rather than in a model, and makes the case that frontier labs exist to sell tokens while Simular exists to remove them. Last but not least, Ang shares his test for AGI — one you feel rather than see — and what it will take to bring autonomous work down to the cost of water.

"So when I go to the gym, [doing] the same workout everyday, as I do this more and more, more reps, more sets, I don't even need to think. It's called muscle memory. I just do the same in a standard way, my consumption or my tokens in my brain dramatically drops. And there's a term describing this behavior in human cognition called 'the Power Law of Practice'. So that's the whole idea. Do we see power law practice in AI agents? We don't have that yet."

Profile: Ang Li, CEO and co-founder of Simular AI

X: https://x.com/angli_ai

LinkedIn: https://www.linkedin.com/in/angli-ai/

Simular AI: https://simular.ai

Episode Highlights:
[00:00] Quote of the Day by Ang Li from Simular
[01:09] Why AGI arrives through the mouse and keyboard
[02:27] Legacy systems with no APIs block automation
[03:45] Computer use is the last mile
[04:05] One device, a hundred agents in the cloud
[05:51] The gap between research and production
[07:45] The day Covid became a dominant search keyword
[08:13] Continual learning over evolving distributions
[09:21] Catastrophic forgetting remains unsolved
[10:11] Why the learning loop requires a product
[11:04] Why AGI must be built commercially, not academically
[12:23] Chat versus work: two different problems
[13:40] Every company needs a repeatable playbook
[15:10] Tokens per task beats price per token
[16:13] Same task 100 times, 100 times the tokens
[16:34] The gym analogy and the power law of practice
[18:18] Where the scaling frame breaks for computer use
[20:46] Why AGI will not be a single model
[21:43] The 72.6% peak versus repeated-run reliability
[22:32] Hiring analogy: the interview versus daily work
[25:22] Replaying trajectories to drive token cost down
[26:17] The idea almost nobody in the industry discusses
[29:01] Democratising autonomous computers to the cost of water
[30:26] SMBs, not enterprises, are the real market
[32:59] Why healthcare and insurance came inbound
[34:06] Launching Sai on Windows virtual machines
[34:38] Simulang and the unity of opposites
[35:52] Frontier labs sell tokens; Simular removes them
[38:01] The hidden cost of self-hosting open models
[39:23] Why CPU and memory matter more than GPU
[41:10] You only have two hands: the physical constraint
[42:48] The AGI test you feel rather than see
[44:19] Why humanoids take longer than software
[45:23] The motivation: doing your work from the forest
[47:03] What Great Look Likes for Simular AI
[48:34] Closing

Podcast Information: Bernard Leong hosts and produces the show. The proper credits for the intro and end music are "Energetic Sports Drive." G. Thomas Craig mixed and edited the episode in both video and audio format.