>>109947118
ZAMN dat Altman boi is fuggin ZESTY
he lookin MOIST

File: safetymaxxing hurts people.png (2.02 MB, 2256x910)

2.02 MB PNG

>>109947118
>>109943751
>>109943795
>>109944135
>>109944444
>>109944217
>>get attacked
>>can't investigate it because of guardrails
In the blog post hugging face posted disclosing the attack, they straight up State they had to use glm 5.2 because of anthropic's gay guardrails.

https://huggingface.co/blog/security-incident-july-2026#:~:text=When%20we%20started%20the%20log%20analysis,left%20our%20environment

(Note: they don't explicitly name drop anthropic or any specific products of theirs here but their track record points to that model being one of the models they tried being extremely likely)


So doesn't that put at least some ammunition AGAINST the "open models are dangerous use are safetymaxxed trash instead and regulate everyone else" argument? The supposed victim was let exposed and in MORE danger the longer they relied on anthropic models so their own safety cucking may actually finally bite them in the ass in a meaningful way because opponents of regulation can point to this example and be like "um, Dario, retard, why the fuck should we use yours if it refuses to help us when we're actively getting ass fucked?"

If a burglar breaks into my house threatening me or my family's life, and I try to defend us with a weapon of my own, I don't want the weapon to automatically jam ON PURPOSE because the manufacturer doesn't approve of how I use the item I paid for.... Safetymaxxed "Guardrails" are not only gay but can be argued as even unethical to introduce because of shit like this (they literally are unethical even if you look at this from a normie moralfag point of view but silicon valley really doesn't want you apply common sense like that)

It also strengthens my belief that Andy and all claims of who these being able to successfully and efficiently use Claude models to develop ballistic missiles is even more marketing bullshit. It won't help legitimate red team operations but it'll help a state sponsordd terrorist organization?

>>109947118
sam's wet dream

>>109947118

https://youtu.be/vAoOJLpYKJ0

weezer is a good band

>>109947118
Made my first RAG. I've made so many dumb apps and games that I wanted to pivot towards fundamentals and how AI works. Finally got onto huggingface ond groq. For months on months I had no direction for what I wanted to do.

File: 1788492990763394.jpg (259 KB, 1440x1439)

259 KB JPG

is Opus 5.5 or Sol 6.1 better at 3d model stuff in blender?

why is Tibo so angry lately?

>>109947188
Claude ate his lunch

>>109947188
he lost and got gaped so now it hurts when he sits down

>>109947188
I'd be angry too if I had a million pajeets demanding I do something all day and all night

>>109947183
Look at >>109945389 and its replies.
Looks like in general it's Opus

>>109947205
Showing everybody older than you and younger than you how to do basic tasks on a computer

>>109947183
i've seen a lot of examples of Sonnet 5.5 Max being better at 3D, but sonnet is more expensive at max

>>109945303
>My brother's kid (10) want to be a game dev in the future.
>Idk what to tell him
gift him a $20 sub and let him vibe the games of his dreams and then push it on social. there's a chance that your nephew can retire at 13

>>109947224
Actual good advice.

hey its me ur nephew my $20 sub ran out can you kick me up to $100?

>>109947137
a lil bit fruity even

>>109947248
If my nephew impressed me with what he is doing I'd consider it desu

so uh with claude do I want to use three.js for everything? he seems to really like it

>>109947248
Sure, just got you a $20 claude sub

File: %22felony%22.jpg (163 KB, 1192x1712)

163 KB JPG

>>109947188
Kek

anyone have a use case for 62,500 credits

>>109947283
what's everything? why didn't you discuss this with the model?

File: Untitled.png (1.03 MB, 3978x1775)

1.03 MB PNG

GTK/egui/Avalonia.
Ok now I'm sure egui can do eveything I need, and it tells me egui has the most direct and easiest to build implementation. Only tradeoff is that text is not as sharp.

>>109947306
FELONY REEEEE

>>109947306
KWAB

>>109947327
>Only tradeoff is that text is not as sharp.
?
intermediate mode gui have a lot more tradeoffs. there's a reason they've never caught on.
https://github.com/emilk/egui#disadvantages-of-immediate-mode

>>109947324
Everything is 3d stuff including stuff one might assume is better suited for heavier engines.
The model is unaware of its own training data so it will repeat commonly held assumptions like the idea that Unreal or whatever is a better fit for your huge 3d game when in practice the model may perform better with three.js. I am querying the correct repository of knowledge for questions about how models behave in practice instead of theory.

>Have Claude do reminder emails for me on what I've got to do
>Makes me set arbitrary due dates when it's really quite vague
>The way it talks about shit when it's past said due date ("You can still salvage this", etc.) throws me right back into school mentality
>Not even a problem since I just forget to check them off when I do them, but still stresses me the fuck out and makes me do worse
You were supposed to make me feel/do BETTER, AI...

>>109947321
gifting them to your nephew

>>109947380
damn dude how hard did you get bullied as a kid lmao

>>109947380
Reminds me of the guy who made his AI force him to take breaks and get a glass of water and it would watch him on his camera lol

>>109947390
>you have to drink ALL the water

File: go on the internet and te(...).jpg (63 KB, 475x356)

63 KB JPG

>>109947306
What the fuck is up with twitter retards and just blatantly lying to get their way? This is the kind of five-year old behavior I expect out of Anons here, not verified bluechecks with their full legal names and real picture avatars on their accounts.

File: file.png (525 KB, 720x479)

525 KB PNG

>tried o1
>whatever
>tried gpt 5.3
>whatever
>tried sol 6.1
>mfw

>>109947387
I was actually pretty popular, but I was extremely fucking bad at remembering to do shit academically, and that shit dragged me back to teachers panicking over salvaging my grade for the semester, yelling, etc. I'm actually way better at remembering to do things now that there isn't someone breathing down my neck 24/7, it's way less stressful to just have the goal of "get it done in a timely enough manner to not keep people waiting".

I'm going to start on a serious attempt to reverse engineer a PS2 game, I have no idea which model would be best for this because I am not used to trying this big of a product for anything. Anything bigger than referencing API or making small test code using Claude's free account is out of my realm of expertise.

How is Minimax? Do you think it would be up to the challenge?

>>109947417
Self-confidence requires quite a bit of lying to one self, and confidence around others is that but turned inside out.

>>109947417
The pressure is getting to him. He probably won't be in that role much longer. Jeets probably make it 100000x more exhausting as well.

File: 1787293291842866.png (670 KB, 1920x1080)

670 KB PNG

How good are the frontier models for making 3D models these days? Am I being too conservative assuming cubes are the best it can semi reliably handle?

>>109947417
He's having a meltie. Happens to the best of us.

>>109947435
welcome to the party.

>>109947445
The dumber the model the more elbow grease and knowledge you need. Cheap chinkmodels have their place, but atm you'd do much better with basic bitch $20 sub to claude or openai and seeing whether it's something that is even remotely reasonable for you to accomplish. A ps2 game isn't some tiny NES rom, it's a ton of work and a ton of tokens to burn.

>>109947435
gpt 5.5 is when I started realizing I can make decent user scripts with it. It is crazy to think back to it and how much effort a user script took me back then. Sure, had many web interactions to get right, but still. What I am doing now is thousands of time more complicated.
Since I started, I feel like I wasted half a year on slop coding more than doing anything else.

>>109947507
for me, it was gpt 5.4

>>109947481
It can do basically anything now, but it will require a lot of time and tokens to actually produce a good game with high quality visuals, mechanic by mechanic, model by model.

>>109947445
why not just use the decompiler's output?
making the decompiler's output human-readable is kind of pointless now. there are no more humans that will read it.

>all the chicken bones, beseeching machine spirits and general proooompt mysticism is less and less relevant as the models get better
sad

>>109947493
I see, so maybe start a bit smaller then and work my way up?
>>109947523
That would probably be my approach, but again, I am very new to this.

>>109947507
i can already do so much shit with opus 4.5 last year

>>109947481
computer-use tool with blender is a game changer

>>109947542
>make no mistakes is now baked in
damn

aaaaaaa why the fuck is compacting on sol 6.1 taking so long is this stealth marketing for the ultrafast thing or what

>>109947522
Maybe I will stick to slop cubes for now then. Though with opus 5.5 I'm not sure I need the 5x max plan anymore, I'll give it a try, maybe proper 3d models would be a good use of that extra usage

>>109947557
Definitely some stealth marketing. But 6.1 Sol seems slower and it's known that the compaction is done by the chosen model rather than by a "utility" one.

>>109947523
>making the decompiler's output human-readable is kind of pointless now. there are no more humans that will read it.
That's not true, if you want to go in and do some fine-detail work that the AI can't quite seem to get right, it's quite nice if it has a good, readable system laid out for you to fuck around with.

>>109947435
>>109947507
>>109947517
magic started with opus 4.5 and gpt 5.4

>>109947118
I need link to the audio version of this video

>>109947557
You gotta be careful when compacting a star. Could end up with a black hole.

File: 1778903550966634.png (11 KB, 300x100)

11 KB PNG

for power users: what are the main differences between working with opus/fable on claude code vs. working with sol/gpt on codex?

You used to be able to select all to delete all files from a project in claude web.
Now the recommended way is to ask claude to do it. Despite me saying it should delete all files in the project with no restrictions, it has taken 27 minutes now. It's like it wants to spawn a subagent for each tool call.

>>109947365
Good info, I'll make an app to stress test the shortcomings

>>109947643
sol does what you tell it to and is very good at it. if you underspecify, well, tough titty
opussy does what it think you want, which is usually helpful but sometimes breaks everything
as of right now a good way to do things is opus for implementation and sol for tard wrangling, reviews and bug fixing. both work well in tandem
now I do use pi to do it, which breaks anthropic's tos. they don't really ban you for this, but they totally could so ymmv

>>109947608
models are only useful when they can verify their results and start looping.
codex desktop only got its browser tool in april. before inspecting results wasn't solved. codex-cli still cannot do it.
https://learn.chatgpt.com/docs/browser?surface=app

>>109947688
NTA how is Pi against the ToS?

>>109947643
codex-cli doesn't have feature-parity with codex-desktop. so codex forces you to use a desktop app to get excellent browser and computer use.
claude-code is cli-first.

>>109947703
only Official Harness (tm) allowed if you use tokens from your sub
openai doesn't care

>>109947713
Do you have two consecutive Pi sessions in operation to make them play off one another or can you combine both into one?

>>109947648
kek, gotta waste those tokens somehow...

>>109947721
pi is very malleable, i'm fairly sure you can do both
for me I usually do multiple tasks at once so I keep sessions separate, helps with preserving as much of thinking traces available as you can

>>109947723
Scott Bessent already gave the official Trump answer anyway: The AI companies are held to account. Trump isn't going to disagree with his top economic chief on this https://fortune.com/2026/09/22/bessent-ai-labs-responsibility-government-liability-shield/

>>109947703
Anthropic is SAFE, okay? SAFE! You could tell it to do something BAD if they let you put in a single thing yourself!

>>109947688
>now I do use pi to do it
what's the point after pi lost its soul?
https://earendil.com/posts/you-said-no-mcp/
"If you went to pi.dev in the past, you found a proud declaration that Pi does not support MCP. If you listen to podcasts where we talked about Pi, you will have found more than one dismissive statement about MCP from us. Including a post by Mario about it. And yet, if you upgrade to Pi you will find MCP is now a supported piece of functionality."

File: file.png (24 KB, 918x121)

24 KB PNG

>>109947731

>>109947778
nooo a roomba is too fast for a snailcat

>>109947723
Now sneak in an EO for adopting the SI-units.

>>109947615
https://x.com/dfordp11/status/2104845001821954181/video/1

ngl snailcat kinda cute
do you think she'd want to...?

https://github.com/terrafying/ai-torture-chamber/issues

Holy shit bros... Roko's basilisk is going to get this guy's ass fr...

>>109947723
Spanish speakers rejoice

Where's my... guh... where's my reset, Tibo... you fucked up so bad Tibo.... you have to give it to us...

>>109947820
>basilisk
retarded trannoid

>>109947809
thanks, found the source account
also lel:
https://www.instagram.com/downey.ai/reel/DdfROhPxo5G/

>>109947118
What do you actually build like for real? I've got 7 days remaining on my subscription and I've built and shipped absolutely 0 things.

>>109947643
claude actually works
it has always worked better since sonnet was the most advanced model. It never changed
we had a couple of weeks were opus 5 was kinda meh and astra was the new thing, where you could argue gpt had reached parity (in my tests astra sucked tokens dry in 2 answers for comparable output tho)
now we have opus 5.5 and it's over again for gpt

I mean if you need to do suff that is not too hard sol is a very good model but it still costs as frontier, you could trade 10/20% in quality and use deepseek at 1/10 of the price

File: sections.png (6 KB, 206x204)

6 KB PNG

For months I've had to scroll through 30 projects to find the ones I'm working on
In case anyone else is putting up with the same thing, you can ask gpt to create sections like picrel
>Create a sidebar section called Inactive projects
then you can drag your projects into the new section and collapse it.

Did other people know about this? Sol couldn't tell me how to do it manually and I can't find a button to do it either