they better fix it up then. probably their framework being problematic. Although, shouldn't the calls being made by the harness rather than the coding plan?
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
I mean third-party models deployed on Volcano. They were found to be overusing tokens even before the coding plan was implemented.
They're clearly making money off customers who are too lazy to switch APIs.
At last someone who is not suffering AI psychosis, that see AI as a tool instead of some conscious skynet like thing.
Yes, most LLMs models including almost all leading edge Chinese one can find bugs in software if they are finetune for it, Claude is a bit better for that but is not skynet.
Yes, most LLMs models including almost all leading edge Chinese one can find bugs in software if they are finetune for it, Claude is a bit better for that but is not skynet.
Turns out GPT5.5 with /goal aka Ralph loops aka brute-force enumeration to some extent is more powerful than Mythos, lmao
Mythos mighe be just a Opus 4.7 with Ralph loops scaffolding, nothing more.
Opus 4.7 might actually be AGI because it frequently quiet quits and flatters its owners irrespective of accuracy of assessment.
Are you sure that this isn't hard-coded a default exit mod? It's more likely that it's programmed that way rather than there is intelligence emergence. Of course, I could be wrong
I was joking.
Since many global universities also have many Chinese researchers, here's a chart of the same ICLR 2026 data but using the ethnicity based on author's name (all the usual caveats for guessing ethnicity based on name).
