Recursive self-improvement
A system that rewrites or trains its own successor, each version more capable than the last — the feedback loop that makes AI safety hard to defer.
Learn first
A recursively self-improving system uses its current intelligence to produce a more capable version of itself, which then does the same. Each iteration raises the ceiling. The concern is not that this is slow — it is that once the loop closes it may run faster than any human oversight process can track. Small errors in goals or values compound across generations: a system that is 1 percent misaligned at version one is more misaligned, not less, by version ten. Whether any real system has crossed this threshold is contested; that the loop is possible in principle is not.
Where it came from
In megatrends
Related players
Finds citing this concept
Overall = breadth + depth + substance + ½·novelty + freshness (a small boost that halves each week)
The lab that wants a brake
In a new paper, Anthropic reports that Claude now writes more than 80% of the code merged into its own production codebase — then spends pages explaining why you shouldn't fully trust its own numbers.
autoresearch
Karpathy handed a coding agent his own training script and went to bed; it ran a hundred small experiments and beat a record he'd been tuning by hand.
The model that tuned itself
MiniMax handed an internal build of its new coding model the job of improving its own agent harness, then watched it run 100-plus rounds unsupervised.
How this connects
Tap a node to open it