Recursive self-improvement (RSI), using models to speed up the development of the next models, became one of the most-used terms in AI circles on X this past week after several big-tech CEOs talked about it in public. Google CEO Sundar Pichai framed it differently on two podcasts. On Alex Kantrowitz's Big Technology Podcast he said "We are pretty far from the end state of RSI." On Alex Heath's Sources podcast he said researchers are co-working with models "in a deep way" and "I would expect the next 12 months to see continued acceleration because of RSI." The widely shared 12-month line comes from the Sources episode, not Big Technology. OpenAI CEO Sam Altman told Stanford GSB that besides AI recursive self-improvement there is another loop, robots that build robots and data centers that build data centers, that gets "much less talk". Anthropic CEO Dario Amodei, replying to a critic on X, wrote about the point where AI-driven acceleration gets fast enough that safety struggles to keep up: "We are nearing that point." This is an X discussion of public remarks by CEOs, not an announcement of any new research program.
Key Takeaways
- ✓On the Big Technology Podcast (released Oct 20), Pichai said "We are pretty far from the end state of RSI", that today there are "RSI loops along with AI researchers", and that he expects that continuum to keep progressing into next year (Big Technology recap; Kantrowitz post, 68 likes and 9,643 views when we pulled it)
- ✓On the Sources podcast (released Oct 21, from about 11:52), Pichai said researchers co-work with models "in a deep way", with models suggesting recipes and automatically evaluating next checkpoints, and added "I would expect the next 12 months to see continued acceleration because of RSI" (Alex Heath episode post, 253 likes and 388K views when we pulled it)
- ✓Reposts that place the 12-month line in the Big Technology interview are wrong; after checking transcripts of both shows we found it in the Sources episode. The early repost from @ONagel33303 had only 4 likes, and its "CONFIRMED: Google plans to accelerate" framing is the reposter's own reading
- ✓At Stanford GSB's View From The Top (video released Oct 21, at about 33:18), Altman described "robots that can build more robots and then data centers that can effectively build more data centers" as a loop besides AI recursive self-improvement that gets "much less talk"
- ✓Anthropic CEO Dario Amodei, replying to Encode AI's Nathan Calvin, wrote about the point where AI-driven acceleration gets fast enough that it is hard for safety to keep up: "We are nearing that point" (post, 4,103 likes and 1.59M views when we pulled it); Calvin and journalist Garrison Lovely then criticized Anthropic for warning about RSI while backing the AEGIS Act, which they say makes it harder for governments to stop it
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background
RSI (recursive self-improvement) means using AI models to speed up the development of the next models. It was discussed heavily on X this past week after several big-tech CEOs talked about it in public, and they did not all say the same thing.
What was said
On the Big Technology Podcast, Google CEO Sundar Pichai said "We are pretty far from the end state of RSI", describing today as "RSI loops along with AI researchers", with agents helping researchers build models in some automated way. On the Sources podcast, released a day later, he said models suggest training recipes and automatically evaluate next checkpoints, and "I would expect the next 12 months to see continued acceleration because of RSI." OpenAI CEO Sam Altman told Stanford GSB (at about 33:18) that besides AI recursive self-improvement there is another loop, robots that build robots and data centers that build data centers, that gets much less talk.
Benchmarks
This is not a benchmark; none of the companies has published a measurable RSI metric. Engagement when we pulled it: Dario Amodei's long reply to Nathan Calvin had about 4,103 likes and 1.59M views and says of the point where AI-driven acceleration outpaces safety, "We are nearing that point"; Alex Heath's episode post had about 388K views; Kantrowitz's "pretty far from the end state" clip had 68 likes. Note that reposts placing the 12-month line in the Big Technology interview are wrong.
What it means for developers
The discussion splits two ways: some read Pichai as "Google will speed up with RSI", others stress that he also said the end state is far off and humans remain in the loop. Critic Garrison Lovely questions Anthropic for warning about RSI while backing the AEGIS Act, which he says weakens government intervention. To judge for yourself, listen to the relevant parts of both episodes rather than relying on clipped screenshots.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.