In short
AI doing the research that builds the next AI, which then does that research faster. The loop at the center of the singularity. Not here yet.
Recursive self-improvement is a loop: a system improves the process that produces the next version of itself, and each turn of the loop runs faster than the last. I. J. Good called the result an intelligence explosion in 1965.
Labs now build toward it on purpose, with that write code, run experiments and draft papers. Today's still lack the judgment that open-ended research takes, so the loop is not closed. It is where the curve points, not where it is.
For a business, the useful version is smaller and already real: aan whose mistakes become rules in its instruction file, so the same mistake never happens twice.
Example
What it looks like.
Generation one writes the code that trains generation two.Generation two writes it faster.Each ring of the loop turns faster than the one outside it.
Related
Read next.
- singularityThe point where technology changes faster than anyone can forecast. We think the run-up has started: adoption alone is enough to change how every company works.Read the definition
- modelThe intelligence aan lobsteragent calls when it needs to read, write, or decide. Replaceable. When a better one ships, the same instructions run on it and the results improve.Read the definition
- instruction fileThe Markdown file that tells aan lobsteragent who it is, what it owns, and what it may never do. The lobsteragent's job description, readable by the lobsteragent.Read the definition