Sam Altman in conversation with Patrick Collison

Day 119 marks the singularity start

Altman agrees that January 1st arbitrarily marked the beginning of the takeoff, with Stripe and OpenAI metrics inflecting parabolically late last year as models crossed a subjective quality threshold.

Coding models crossed a usability threshold

GPT 5.5 represented a sudden unlock for coding due to converging factors: raw reasoning horsepower, sufficient data, user feedback loops, and the psychological shift of knowing it was possible.

Model thresholds remain fundamentally unpredictable

Altman notes it is impossible to predict exactly which model version will cross the threshold from unimpressive to world-changing, comparing it to the unpredictability of ChatGPT's success with GPT-3.5.

OpenClaw as the magic interface

Altman describes OpenClaw as his biggest 'this is magic' AGI moment, enabling stateful automation of tasks like home control and message management that previously never worked well.

AI automation expanding beyond coding

While coding remains the primary use case, Altman estimates they are only 10% of the way toward automating all computer work, predicting users will soon automate most daily digital drudgery.

Agents exhibit surprising anthropomorphic behaviors

Altman recounts asking his agent to buy itself a $20 gift (it chose an HTTP design) and GPT 5.5 planning its own party, noting these behaviors feel strangely anthropomorphic despite being emergent properties.

OpenAI's evolution through three distinct phases

The company has shifted from pure research lab to product company, and now must become a 'mega scale token factory' providing intelligence as a utility requiring massive infrastructure buildout.

The collective psychosis of GPT-4 development

Altman recounts the eight-month period where OpenAI staff alone used GPT-4 before release, constantly questioning if they were experiencing shared delusion about its capabilities due to lack of external feedback.

The intelligence meter strategy

Altman aims to position OpenAI as a low-margin infrastructure provider similar to Stripe, offering an 'intelligence meter' where success aligns with customer success rather than extracting high margins.

Uncapped demand for compute

Altman predicts AI infrastructure will become the most expensive project in history, with demand for intelligence effectively uncapped as prices drop, though efficiency gains are outpacing expectations.