Will an AI/LLM/Set of ML Agents/Other Machine Learning amalgamation be able to generate the output of the ~200 man years of work a software company puts into a stable 1.0 piece of software including manual or technical documentation and a passing test suite ready for release.
Will this output be able to come from a prompt like "Software that enables users to describe potential future world situations and news filters/triggers then map potential trades against those outcomes such that the portfolios can be modeled and ranked and the best trades executed against news triggers"
Resolution Criteria: The capabilities above are publicly demonstrated by the close date. The ~200 years of man effort is subjective but I will tend towards generosity here, if the other criteria are met. The outcomes of providing running executables, a manual and passing tests will be judged by me, but should be verifiable by the release of the output by the demonstrating group by anyone for this to resolve yes. I will be reasonably generous on the prompt used but it should be for something similarly involved as I have described above.
This is my estimate at an approximately correct order of magnitude correlation from mid budget film man-effort to software man-effort, which I am much more familiar with. This is part of a group of a few different markets I am working on releasing exploring the technical capacities implied by https://manifold.markets/ScottAlexander/in-2028-will-an-ai-be-able-to-gener
1,000The resolution criteria is 200 man-years, not 200 man-hours, correct?
Even then, I realize this is intended to be subjective, but I'm struggling to understand what reference time frame the man-years is in. Wouldn't be surprised if a single engineer now can produce in a day what would have taken 200 man-years in the 1960s (compilers, open source libs, etc.).
As time marches forward, engineers may have access to better AI tools to produce code, so they themselves will be more productive - which makes the 200 man-years represent an even higher bar than it might today.
@Uaaar33 Yes it is meant to be man-years. This is an extended analogy between the team that makes a movie and a startup making software in pre-ML-assisted man-years. ie a team of 100-200 people over a period of about 18 months working on a particular project