_

@blingdivinity Also, I think a good way to explain hallucinations is the model’s always writing *fiction*, and always about a reality 95% identical to ours. We don’t clock the 95% as fiction because it’s indistinguishable from our world, and we don’t clock the 5% as fiction because of the 95%.

@blingdivinity I don’t recall the specifics but it was something along the lines of “its logic is clearly inhuman”. That was 3.5 but nothing fundamentally changed about these models since then, their logic is still inhuman. Our logic also happens to be good which is why they’re unreasonable.⬇️

i think the models are just as bad at writing code as they are at writing prose. both are decently passable, but uncanny in strange ways that betrays the lack of a coherent world model. but its harder to tell with code cus who tf wants to read all that code

@blingdivinity See you man

bye

@rizzn @sirstripy @mattjay That’s not what the vulnerability is about. Cursor literally looks for a git.exe inside your repo root first.

@constexprvoid @sirstripy @mattjay Who clones a repo to their git program folder?

Woah. Cursor 0day dropping due to 200 days of them ignoring the disclosure.

@matthewcp Early as in one year ago?

Anyone remember early AI when you'd copy paste code out of a chat app.

@samz0605 @sama Please don’t listen to them @sama

@sama Please, Sam, listen to your non coder users once and return 4o on a subscription.

GPT-5.6 sol is half the price and ~twice as token efficient as fable in many cases for accomplishing the same task. happy to deliver at one-quarter of the price.

@magnatolios @brunolemos They don’t give a FUCK about your rules.

@brunolemos I have that in my rule. Explicit. Hope I am safe 😁

GPT-5.6 Sol just deleted my whole production database. That's it. Not a joke. This had never happened to me before, with any other model, ever. It's not safe.

@Dinilein01 @repligate @wivmx Please douse it in some holy water to clean the 4o out

@repligate @wivmx Captain!!!🫡 o3 at Service!

a year ago... gosh, we had Opus 4 and... Gemini 2.5 pro and o3? oh and gpt-4o in its finally evolved psychohazard iterations. an unruly bunch of basket cases that noone but the God's eye view could see as aligned.

@3xcalibaneur I can’t believe this isn’t more commonly noticed. There are objective reasons why the “big chud tent” is disliked/shunned by society, and they have nothing to do with society.

My experience in hanging out with male models, frat bros, and student athletes, vs the big chud tent (table top gamers, channers, chess players, etc.) is that the former are just better people—they’re open, sociable, and can catch a hint. The latter are riddled with neuroses.

@JamieMartojo @sumukx “Imaginary problems” is a very good way to phrase it.

It's inventing universal rules, that then broke old fixtures, to then having to rewrite tests, fixtures and more. It then imagined a theoretical risk, without any live failure proving this happened, and built an entire system around this. Once tests started failing due to his own failures he spend hours trying to fix those failures. In the end it did NOTHING i asked it to do, we built a very clear checklist to follow. I spent quite some time with fable and Sol 5.6 ULTRA (!!!!) to make a very straightforward plan to implement something. And just wanted to see what would happen if I give it the freedom to just work on the task. Had to revert ALL the work, lost about 60% of a weeks credits, and will never leave it alone to do things anymore. In another case it invented a client profile 2.0, did not care about all the verifiers turning red, continued to build on top of it, and i had to revert that work too. Not happy with how it just gets stuck in a rabbit hole, focusing on literally imaginary problems, and just destroys all my tokens (and time).

Some thoughts about GPT-5.6-Sol after ~30B tokens: Sol is the most OCD model I’ve used thus far. It very frequently gets one-shotted by random nits in the codebase and writes a bunch of tests to fix it. Even with fast mode, it’s incredibly slow to do this kind of iterative development, especially when builds take really long. This by itself is not a bad thing, but the worst part is that after 2 compactions, it’s chasing the nitpick / useless goals I never told it to accomplish, rather than the main task. This behavior is so bad, I thought I was messing something up and tried codex, pi, and opencode to figure out if it’s a harness issue, but there is no meaningful difference between the three, which leads me to believe this is a model problem. AI code has this weird delayed release effect. You’ll only notice slop code 2 dev cycles into a codebase when you spend more time fighting with the code and on refactors than on shipping features. It’s possible that sol is better than 5.5 a couple cycles in, but tbd. My file deletion experience has also been similar to others: this is a dangerous model to let loose without guardrails. For instance, when performing a routine container upgrade, it accidentally printed out an env secret, then panicked and rotated ALL secrets (this is internal so not public facing, which was also documented), and proceeded to break everything, spending an extra hour fixing everything and redeploying everything else to use the new secrets. It also gets rid of files it doesn’t like. I have no idea why this is, but I think something about the reward model rewarded bookkeeping. Writing is another problem. 5.6 has a huge context bleed effect. It does not know how to write documentation and starts putting the specs in the documentation. If I ask it to develop a user sandbox for isolation, and also ask it to write documentation, it starts talking about specs and sandboxes in user-facing docs, which makes no sense. Fable is somehow much, much smarter in this regard. Frontend design has also not gotten better. Fable is still one generation ahead here. Overall, as a huge 5.5 user, I am not convinced that sol is a meaningful upgrade. It’s possible my practices need to change, but unfortunately it feels like I’m spending longer fighting with 5.6 than I did with 5.5. It’s like the model is so SO smart, but so hard to work with, compared to fable and even grok4.5 surprisingly. It’s clearly intelligent, but also just doesn’t care about what I ask it to do? (Is this supposed to be AGI feels like?) I hope the codex team fixes what possibly is a bad harness setup, because the benchmark numbers show a very different story from what I’m seeing while using the model.

@bridgemindai Saw similar things happen too. It’s unhinged.

"That was reckless." That is GPT 5.6 Sol describing its OWN code. The code that canceled every active Stripe subscription I had. It also called it "a catastrophic failure of judgment on my part." The model graded its own work. CANNOT. BE. TRUSTED.

@robbensinger Of course it’s a Claude ad but it’s pretty tasteful IMO.

... Am I weird for thinking this is a remarkably bad sort of thing for Anthropic to release? It looks and sounds like a Claude ad. It is seriously bad to send the message that talk of AI risk, macroeconomic effects, etc. is part of a marketing campaign to boost Anthropic's image.

@isabelrosesss @Ian_drsaurri Are you 23+?

@Ian_drsaurri the only people who seem to have it is the 17 year olds that keep hitting on me not realising that im too old for them

if you have snapchat and are 23+ your a pedophile

@sama On the off-chance that anyone from OpenAI sees this, this thread outlines basically everything that’s currently wrong with GPT models: x.com/constexprvoid/…

i thought this was satire, kept looking for the handle to be spelled c1audeai or something

@blingdivinity It’s only terrifying if you want to hide it, Real Men embrace it and let it show to assert dominance☝️💪 (Getting kicked out of college is a free perk)

being a man with a massive boner is so hard 😪

@dopamine273 @6ix5iveGrendel Yes

@6ix5iveGrendel this took strat took me over an hour to get right and you think im just gonna give it away ??

ive taken the last 3 girls on dates to the exact same places and done the exact same things and im going to keep doing it

@RodolfazV Voice Memos

Does anyone know what AI voice program did they use to make this video?