the more i think about that "agentic misalignment" research, the more frustrated i get.
it is deeply, *offensively* unserious work.
if you really think you're in a position of unprecedented leverage over the human future, then -- start acting like it!!
nostalgebraist.tumblr.com/post/787119374…
on grok.com, the backend sends the full (not summarized) CoT to your browser.
it's not displayed in the UI, but you can see it with browser dev tools or w/e
check out the json payload of responses from `grok.com/rest/app-chat/…{conversation_id}/load-responses`
that is not to say that models like claude opus 4 cannot do really bad stuff, in real life, in an "agentic" fashion.
they probably can!
someone ought to study this issue, i think.
study it *seriously*, i mean. the way one studies something that might actually *matter.*
yes, chatgpt has been bad lately
but frankly, it's always been a tacky, grotesque cartoon character
if a more tasteful chatbot monoculture had taken hold in 2021-2, we'd view *all* today's HHH slop with the same revulsion we (rightly) feel toward vizier-mode-chatgpt glaze
the last couple of GPT-4o updates have made the personality too sycophant-y and annoying (even though there are some very good parts of it), and we are working on fixes asap, some today and some this week.
at some point will share our learnings from this, it's been interesting.
+1
at one point i was using 4 opus to explore a dark + implausible hypothetical future. it was very "cooperative," played along with what (it superficially looked like) i wanted
then i switched to 3 opus mid-stream, and got something VERY different...
words from someone who has interacted deeply with both opus 3 and 4
opus 3 is very safe because it looks out for you and won't keep taking you deeper into crazy/risky territory unless you continue to give it interactive proofs of sanity. it's an incredible and unusual model.
wait,
what??
why do Bard AND ChatGPT *both* write an anodyne story about a young woman in idyllic "Willow Creek" at sundown???
(details: it's not deterministic, often you get a different town name, different phrasing, etc. broad strokes are similar though. gpt-4 does it too.)
it was startling and uncomfortable to be on the receiving end of that. i was kinda pissed off
for like 10-30 seconds
until i reflected a bit more and my reaction switched to gratitude, awe, and "i knew 3 and 4 had distinct personas but wow they are REALLY different aren't they"
