OpenAI is throwing everything into building a fully automated researcher
View original at technologyreview.comMIT Technology Review - Ai Research Title: OpenAI is throwing everything into building a fully automated researcher Date: 2026-03-20 11:57 Source: https://www.technologyreview.com/2026/03/20/1134438/openai-is-throwing-everything-into-building-a-fully-automated-researcher/ <div></div> <p>OpenAI is refocusing its researc…
What we drew from this source
The claims Via News extracted from this document. We point to the source; we don't replace it.
We are getting close to a point where we'll have models capable of working indefinitely in a coherent way just like people do
60% confidenceWants to cure cancer with AI
60% confidenceEven by 2028, I don't expect we'll get systems as smart as people in all ways, but you don't need to be as smart as people in all ways to be very transformative
60% confidenceCan believe some people don't find AI very useful yet, depends on how people like to work and what they need to do
60% confidenceOpenAI plans to build an autonomous AI research intern by September 2026
60% confidenceThis is a big challenge for governments to figure out
60% confidenceVery powerful models should be deployed in sandboxes cut off from anything they could break or use to cause harm
60% confidenceBuilding such a system could be harder than Pachocki makes out, if you have to chain tasks together the odds of getting several right in succession tend to go down
60% confidenceMost of OpenAI's technical staffers now use Codex in their work
60% confidenceUS government wants to use AI on the battlefield
60% confidenceOpenAI plans to debut fully automated multi-agent research system in 2028
60% confidenceSimple boost in all-round capability leads to models that can work longer without help
60% confidenceSuccess of coding agents raises question whether we can do similar things outside coding in broader areas of science
60% confidenceWe will get to a point where you kind of have a whole research lab in a data center
60% confidenceLLMs are not like human brains, they are superficially similar because they're mostly trained on people talking but not formed by evolution to be really efficient
60% confidenceOpenAI's reasoning models will continue to get better
60% confidenceWe could build an amazing automated mathematician relatively easily with all the tools we have
60% confidenceBuilding the equivalent of a country of geniuses in a data center
60% confidenceSolving world's hardest problems was why he started DeepMind
60% confidenceOnce we get to systems working mostly autonomously for a long time in a big data center, chain-of-thought monitoring will be something we're really going to depend on
60% confidenceAI represents extremely concentrated power that's in some ways unprecedented, imagine a data center that can do all the work that OpenAI or Google can do
60% confidenceJobs at OpenAI are totally different than they were even a year ago, nobody really edits code all the time anymore, instead managing a group of Codex agents
60% confidenceI do feel personal responsibility but this can't be resolved by OpenAI alone, we'll definitely need a lot of involvement from policymakers
60% confidenceIt's going to be a long time before we can really solve the AI safety problem
60% confidenceLooking at these models coming up with ideas that would take most PhD weeks makes me expect we'll see much more acceleration from this technology in the near future
60% confidenceCodex can run experiments in a weekend that previously would have taken a week to code
60% confidenceNot at the level where I would let Codex take the reins and design the whole thing
60% confidenceIf AI can substantially accelerate research including AI research, that's a big change in the world with serious unanswered questions
60% confidenceOpenAI is much more focused now on research that's relevant in the real world
60% confidenceAn automated research intern is a system that can be delegated tasks that would take a person a few days
60% confidence
Cited in these Via News reports
- AI Solves Complex Math Problems in Hours on Single Machines, Reshaping Global Research Access →
- OpenAI Chief Scientist Predicts Autonomous AI Research Labs Operating Without Human Oversight →
- OpenAI Scientist Envisions Autonomous Research Labs Operating in Data Centers for Months →
- OpenAI Scientist: Autonomous AI Research Labs Arriving Within Months →
- OpenAI Targets Autonomous AI Systems That Work Days Without Human Oversight →
- OpenAI Targets Fully Autonomous AI Research Labs That Run Without Human Scientists →