Introducing C★3: a machine that improves its own engine.
Accelerated selection guided by intent. A fixed verification criterion that never moves.
Early results:
• 8-element sorting routine 4.3% faster than the known optimal network
• 15× fewer formulations than CEGIS across 192 cases
• 32 self-improvements, zero regressions
I named my research ‘reverse-engineering chance’ on Sep 22. On Oct 16 Anthropic’s Claude found the NIST seeds came from lrand48 + timestamps. Different methods. Same theme. Full formalization + code open: https://t.co/3HFdGje1oU
@xtechbrains Yes, they need to know the gift and be on the edge of the present constantly.
This starts by observing properly.
In the coming weeks there will be four publications starting with Tuesday
This is just the first step of the research
three more releases are coming over the next few weeks. each one is more complex than the last, and builds on it
the first one goes open source on tuesday
@FelippeRegazio@FelippeRegazio terça feira sai a primeira publicação de código aberto da Limiar. Nas próximas semanas publicarei 4 pesquisas, essa primeira é a unidade básica do restante. Acompanha lá!
https://t.co/SX4578kcQo
The usual detectors: "something changed"
ours: "failed checks per minute went from 0 to 1, since 11:17"
ran it on a whole simulated company. 131 processes, 30 days, all at once
tuesday. open source.
What @LimiarResearch posted today is a small part of what i've been researching
i tested it against 15 known methods, on over 2,000 series, with public data and a whole simulated company
tuesday. open source.
The usual detectors: "something changed"
ours: "failed checks per minute went from 0 to 1, since 11:17"
ran it on a whole simulated company. 131 processes, 30 days, all at once
tuesday. open source.