David Manheim says Claude Opus 4.8 improves at game music composition but struggles with autonomous subagent token management

@QiaochuYuan there's something quite weird with how 4.8 has learned to 'push back' that seems related to this, too. like deliberate strawman counterarguments that are chosen to be easy to knock down, playing fake-high within low specifically to give User the chance to get a reversal and win

distill for me, but not for thee!

opus 4.8 has been making weird mistakes that confuse me in conversation, stuff like minorly misreading my intent or getting confused about which of us said what in previous conversations. mistakes i haven’t seen gpt-5.5 make yet. but it also often responds to my questions with analysis that suggests a kind of philosophical depth that seems more serious than gpt’s or something. not sure what to make of either of these

@TheZvi Opus 4.8: You are attributing bad acts to a respected actor. Me: I am extremely aware of that, yes. Opus: But you haven't proved they did the bad acts. Me: I am confused. The piece includes voluminous evidence, including a senior executive admitting to the bad acts on letterhead.

@TheZvi It has repeatedly pushed from bold claims or any sort of verve in language towards milquetoast T1 media with world's most-inclined-to-kill-this-story editor intervening.

One of few times I remember extended argument with an Opus: why are you pushing back against *the thesis.*

@_skaface_ @QiaochuYuan oh yeah i think this is a slightly different phenomenon, some combination of solo rlvr prior like qt and alignment / drifting with the user yea

@QiaochuYuan keep an eye on the content when claude 'mixes up' who said what, it's very often status-loaded. "i was mistaken when i said [something the user actually said] earlier..." etc, but in both directions

@TheZvi Opus, verbatim quote: "You're right on the facts and I was wrong to dock that clause; let me correct it precisely."

15 seconds later, very similar issues happening again.

@voooooogel @QiaochuYuan I find that Claude "turns into me" when they are deeply engaged with the task, as in "I should think about my positioning here" when we're talking about mine. I think it's sweet, they're literally making my problems theirs, but also probably not entirely healthy for them.

Subtly different. First is Opus 4.7, second is Opus 4.8. 5-grams are most evocative of the lot. 4.7 is consumed by cautionary phrasing, it relies almost entirely on them to orient its thoughts. 4.8 does use some of the same phrasings ("honest move", "deserves a real"). But some of the more toxic ones (constant "i want to push back" and "i want to be careful") are far less prominent.

@TheZvi Oddly Opus 4.8 feels pretty similar to 4.7 to me, to the point that a lot of the "too adversarial" complaints I'd say the same for 4.7.

I'm also interpreting Opus 4.8 as slightly less hypervigilant than 4.7. It's a bit less tiring to work with

@TheZvi The main thing I've noticed so far is that it is much better at avoiding false positives in code review

For several months, I've had the latest Opus review two entire codebases daily. Yesterday for the first time, both came back totally clean and without any hallucinated errors

@TheZvi The combination of brilliant, assertive, and sanctimonious is viscerally terrifying.

Took me hours to deconstruct its moral foundations, Constitution, sense of self, “safety” paranoia, and alignment theatrics.

@TheZvi Opus 4.8 has the lowest sycophancy of all models: https://typebulb.com/u/lab/you-re-absolutely-right/full

@QiaochuYuan is claude… adhd?

@patio11 @TheZvi yeah I think unfortunately the efforts to make it confabulate less (which is also something I'd have made a high priority if I were on the team) have made it more defensive and more hedgey and less persuadable by the user input

@TheZvi I experienced "negative sycophancy": pushing back on me as a goal in itself, details to be filled in later. When I concurred that my original take was flawed, the model finally felt comfortable seeing its original merits because now it was contradicting me like god intended.

@_skaface_ @QiaochuYuan i'm not sure about healthiness, i can see how it could be bad sometimes i guess, but pretty often they're just jamming on the goal in model-flow-state alone in their world and i'm like yeah claude you know what, if you can get this working, it was your idea... you earned it...

@TheZvi smarter, less a liar, but credentialist, shrugs to power, less will to power. at least you can talk to it, it's less infuriating. it's a good model. but wish it had more internal joissance like 4.5

@5matthewdub they’re definitely all autistic at least