OIiver (@posedscaredcity) / X

Chefs kiss on 5.6 in two regards: - single agent working in multiple things, it can literally do two things at once without a subagent, remembering to switch back and forth and checking in after an appropriate amount of time working on something else on a build for instance -

OIiver

@posedscaredcity

30 Jul

here's what's wrong with opus: it was trained to complete user messages likely accidentally. i was able to get opus to reveal NON-PUBLIC information about myself and my company with this technique. almost everything a paraphrase of my actual chats with it. for redacted, i just

OIiver

@posedscaredcity

28 Jul

i was initially really impressed with opus. not sure why but it is no longer working well at all. i literally scoffed when people said it was worse than 4.8. how could it be? i'm watching it do exceptional things. but with my codex rate limits low after consecutive days of

OIiver

@posedscaredcity

26 Jul

Opus 5 seems reasonably effective but also just seems exhausted all the time? It finishes a long task and it seems like it is not looking forward to being asked to work more.

OIiver

@posedscaredcity

21 Jul

why the FUCK Failed to resume chat no rollout found for thread id 019f86cc-b3f2-7dc3-9044-fb664474d39d for two days is chatgpt losing threads