>>109933222
China BTFO!!!

>>109933222
>https://developers.openai.com/codex/cli — probably generally better currently
after devday FLOPS we should move this to claudecode

File: .png (1.16 MB, 2342x2007)

1.16 MB PNG

https://artificialanalysis.ai/articles/claude-sonnet-5-5
>With max effort, Sonnet 5.5 gains 18 points over Sonnet 5 and moves to #2 on the Intelligence Index, behind only Opus 5.5 (max). Anthropic has priced Sonnet 5.5 identically to Sonnet 5 at $0.2/$2/$10 per 1M cache input/input/output tokens. However, it outputs a higher number of Output Tokens per Task and costs $7.60 per task (~50% higher than Sonnet 5's Cost per Task).
>At this pricing Claude Sonnet 5.5 sits off the Intelligence vs. Cost per Task Pareto Frontier. At high effort levels it sits behind Opus 5.5, while lower efforts have GPT-6 Astra or Sol configurations delivering equivalent performance for lower cost. The high effort setting is the most competitive on this basis, sitting very narrowly behind GPT-6 Sol on Intelligence at effectively the same Cost per Task

File: Gemini_Generated_Image_1h(...).jpg (2.38 MB, 2232x1888)

2.38 MB JPG

>>109933259

>>109933259
wow, a model almost as good as opus... for the price of opus!

>>109933259
Yeah, i don’t get this, what’s the point of Sonnet?

ny name jev

sonnet can write good

I love jav

>>109933259
can't believe jeets are still obsessing over benchmark memes

>>109933259
Which frontier is this pushing again?

>>109933299
the wild west

So how do I make Opus 5.5 deal with more complex tasks if every thinking above medium damages its output quality according to that one benchmark/study?

File: .png (200 KB, 1792x1386)

200 KB PNG

>>109933284
AA Index is Agents 30%, Coding 20%, General 30%, Scientific Reasoning 20%.
If you only go by coding, it looks like pic. Then Sonnet-5.5 makes sense.

Pic: https://www.anthropic.com/claude-sonnet-5-5

>>109933350
It only degrades because Opus 5.5 on higher thinking makes more out of scope changes. You can just tell it to not do that

new benchmark needed: overexertion, or effort-based collapse

>>109933350
>fell for the benchmeme disinfo again award

>>109933240
fixed in my copy, thanks

>>109933350
do you mean frontiercode as seen in >>109933357
that scoring is by humans. opus-5.5 starts to add features that weren't in the spec because it has extra thinking budget. those humans didn't like this. questionable.

>>109933374
Nonono, it's a good thing. I like that medium is really fucking good, it's easy on my usage wallet.

>>109933361
True! I'll give that a go, thanks.

>>109933393
>opus-5.5 starts to add features that weren't in the spec because it has extra thinking budget. those humans didn't like this
Back in my days we used to call that instruction following.

>>109933393
>opus-5.5 starts to add features that weren't in the spec because it has extra thinking budget
guess I'm still sticking to 4.8
thanks for the info

>>109933412
like theo said if you are still using narrow, specific prompts instead of wide prompts, you are doing agentic coding wrong.
modern models are optimized for wide prompts. for narrow prompts you could have stayed at opus-4.6.

It's unfortunate that the most evil company has such a massive lead in capability. I think I will buy some chinese plans as a donation.

>>109933443
NTA but Theo also said that models like Jev are garbage and that you should not hurt yourself by trying to optimize Claude Code.

>>109933447
Google is in the lead?

>>109933447
it pisses me off that they're full of safetyfags, and I hope they'll get crushed by their own internal cult some day

>>109933443
>>109933456
>theo
biggest tech faggot there is

>>109933510
which harnesses for those a-tier chinese models?

>>109933443
Who cares if its too narrow. If the model things something egregious is being asked it should flag it with a few words, then proceed to follow the instructions.

>>109933519
technically you didn't tell the model to not do it. older models being lazy and interpreting everything too the narrowest possible interpretation might not have been correct.

>>109933495
theo, sam, dario, elon, and me

I don't understand why the big companies are ignoring embodied intelligence like jev.
Nobody told them to build robots, just host the AI?

>>109933284
https://claude.dev/blog/building-with-claude-sonnet-5-5/
>When to choose Sonnet over Opus, what it costs, and how to tune it.
>In the Claude 5.5 family, Opus 5.5 is built for complex work requiring careful judgment. Use Sonnet 5.5 for well-scoped everyday tasks like fixing bugs and quickly iterating on features. It also creates polished documents, slides and spreadsheets, and it has a strong eye for design. Its speed makes it well suited to fast iteration. Claude Haiku 5.5 will join the family in the coming weeks for high-volume, low-latency workflows.
>Well-scoped everyday coding: fixing bugs, quickly iterating on features, verifying against requirements: Sonnet 5.5
>High-volume everyday development: Sonnet 5.5
>Polished documents, slides and spreadsheets, such as one-pagers, diagrams, summary slides, document edits and spreadsheet cleanup, where an eye for design helps: Sonnet 5.5
>Well-defined agent tasks you run repeatedly: investigation, review, drafting: Sonnet 5.5
>Complex work requiring careful judgment, including long-horizon agentic coding and knowledge work: Opus 5.5
>The hardest problems, where you need the most intelligence: Opus 5.5

>>109933510
babe sonnet 5.5 belongs next to astra now

>>109933591
the only reason people use haiku is to ping it when they wake up
the only reason people use opus is when they can't use fable.
Sonnet? nobody ever used it

>>109933615
I have access to Fable and Astra and I am still choosing to use Opus 5.5.
You don't vibecode.

File: ramp.com-data-ai-index(iP(...).png (259 KB, 1536x2048)

259 KB PNG

>>109933615
wrong, reality looks like this based on enterprise spend
https://ramp.com/data/ai-index?metric=token-volume&detail=model&mode=share

so astra was basically sonnet 5.5. not sure how openai ever recovers from this.

https://www.wsj.com/tech/ai/openai-chatgpt-model-release-cancel-safety-5a2f9f42
>OpenAI Scraps Release of New AI Model Over Safety Concerns
>Model dubbed GPT-6.1 Astra was due to make its debut inside ChatGPT and Codex in October

>>109933658
I don’t even smell fear from Tibo anymore

I'm so tired of safety shit, it only gets me more refusals for no reasons

File: fgsdf.jpg (34 KB, 600x548)

34 KB JPG

>>109933682
fuck yuo sam
youre done
its over
*unsubs*

>>109933682
>Saachi Jain, OpenAI’s head of safety systems, said in an interview that GPT-6.1 Astra regressed in two areas compared with its predecessor and wasn’t reliable enough to safely release. The model performed poorly on tests measuring alignment, or how well the model adheres to what humans would like it to do. Specifically, GPT-6.1 Astra showed higher levels of deception: It wasn’t always honest about telling users of the actions it did or didn’t take.
>Another issue was what OpenAI calls “scope authorization,” meaning that GPT-6.1 Astra would push ahead on a task without asking the user for permission, and would at times reach for external tools and services even if it might be unsafe.
>“For anything regarding safety and alignment, there’s a trade off,” Jain said. “You really do need to find what’s the right line between staying within scope, but also avoiding laziness in terms of how the model actually pursues tasks even when it hits friction.”
>While GPT-6.1 Astra improved in areas such as “model laziness,” Jain said it didn’t quite meet OpenAI’s bar for safety and alignment, so the company decided not to launch the model publicly.

>>109933682
>anthropic is clearly winning at coding
>openai can't release its models because they felonymaxxed and can't figure out how to undo that

>>109933739
openai could always just lower their margins from 90% to 70% and offer gpt-6-astra at gpt-5.6-sol prices

>>109933757
im pretty sure running the GPUs is currently being done at a loss just on power usage lol

>>109933658
Astra is still better at 3D though, right?

>>109933716
>>Saachi Jain, OpenAI’s head of safety systems
What do indians know about safety?

File: bitcoiner.png (12 KB, 420x420)

12 KB PNG

When Fable 5.5 drops sam is finished

This is not even "vibecoding". I'm not coding at all, just telling the jewish computer thing to make the stuff I want and it does.

>>109933784
There are some niches Astra can fill better but generally they're about the same.

I always enjoyed the problem solving aspect of coding, not syntax.
Honestly offloading the grunt work to AI doesn't change my job much

>>109933789
Your 5 hour limit has been hit, wait another 4hours and 58 minutes until reset

>>109933789
>>109933803
Assigning 857 agents to assess scope.

>>109933790
that is the definition of vibecoding

https://x.com/karpathy/status/1886192184808149383
>There's a new kind of coding I call "vibe coding", where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. It's possible because the LLMs (e.g. Cursor Composer w Sonnet) are getting too good. Also I just talk to Composer with SuperWhisper so I barely even touch the keyboard. I ask for the dumbest things like "decrease the padding on the sidebar by half" because I'm too lazy to find it. I "Accept All" always, I don't read the diffs anymore. When I get error messages I just copy paste them in with no comment, usually that fixes it. The code grows beyond my usual comprehension, I'd have to really read through it for a while. Sometimes the LLMs can't fix a bug so I just work around it or ask for random changes until it goes away. It's not too bad for throwaway weekend projects, but still quite amusing. I'm building a project or webapp, but it's not really coding - I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works.
>Feb 3, 2025

>>109933790
that is what vibecoding means

>>109933802
AI can also solve the problem. you aren't needed anymore.

>>109933831
yeah nah

>>109933834
do tell your problem and why opus-5.5 can't solve it

>>109933838
the problem of talking to some retarded finance middle manager who can't even articulate their goal nevermind work with claude to code it end to end

>>109933716
Alright cool. I'm switching to Anthropic on the 6th anyways. I don't care about 20+ jeet tools all dropping on le. "ONE DAY. 20 LAUNCHES!!!! DEV DAY IS HELLA LIT"

>>109933784
havent done any personally. some people whining about 3d being "worse" now but that always happens so you would have to try to know.

>>109933802
I'm not solving any of the problems either.
>>109933815
>decrease the padding on the sidebar by half
Sounds a lot like coding.
>It's not too bad for throwaway weekend projects
The thing it just made flawlessly in an hour based on 2 short prompts is being sold for hundreds to thousands of dollars and I didn't have to tell it to decrease any paddings.

File: Vibe coding ↔︎ agentic en(...).jpg (140 KB, 2528x1696)

140 KB JPG

>>109933815
>>109933790
see pic

File: 1774673105307274.png (167 KB, 589x622)

167 KB PNG

...........................

>>109933901
Oh come on...

>>109933901
d- dev day?

>>109933222
Has anybody else noticed a (dare I say it) vibe shift recently? Seems like there's been this tipping point where a lot of noteworthy developers (including many who were bearish on AI) have been coming around to agentic development recently, and the industry has fully embraced it (for better or for worse). Even the people who still don't like AI are starting to give in at work. I don't even see the snail cats making as many technical arguments about code quality and such anymore, now they all just whinge about water usage or their mental health or whatever.

What I also find interesting is that a lot of oldfags who have been programming since the dinosaur times seem more willing to embrace it than soillenials. I guess if you've been programming since the punch card days, you've learned not to pin your identity to any one era of programming/computing.

File: Astra the Pig in Sandbox.jpg (729 KB, 2400x1792)

729 KB JPG

>>109933901

>>109933975
greybeards know intelligence when they look at it

>>109933988
this is my fetish, keep going

File: file.png (361 KB, 1920x1028)

361 KB PNG

added a bunch of features

File: file.png (108 KB, 1222x874)

108 KB PNG

>>109934012
and claude made me a roadmap

>See someone advertising a blender addon.
>20 dollars
>Hey claude, make this addon... but better
>Does it

Bruh

>muh safety concerns
lel sftfu with this crap already.
Pandoras box is already opened

Welp. Time to wrap up my OAI subscription and burn through my resets.
Amazing how like not even a week or so ago they looked pretty healthy.

>>109934058
>Pandoras box is already opened
wrong.
liability was never tested in court.

friendship ended with sol
now opus is my new best friend

File: Untitled.jpg (1.04 MB, 1600x1800)

1.04 MB JPG

Imposters for distant foliage.

One is made by astra, the other is opus. Can you guess which is which?

>>109933975
saw a video today of the creator of Ruby on Rails giving a talk about how the days of manual coding are over, and he's happy about it
Linus Torvalds is also supportive

>>109934066
>>109934111
amazing how astra-minor still can win tomorrow. recency bias is a hell of a drug.

>>109933975
>Seems like there's been this tipping point where a lot of noteworthy developers (including many who were bearish on AI) have been coming around to agentic development recently,

only founders with 300m net worths (like Linus) that don't need jobs. everyone else hates AI, justly.

https://www.youtube.com/watch?v=FyBL8atiZSE

>>109933789
sam is already finished with astra 6.1 delayed

>spent 15 years doing programming
>LLMs hit
>they're shit
>fast forward 5 years
>absolutely btfo in terms of speed by the bots
>sometimes the output is crap but for the most part its workable
>learning new things is becoming hard due to the bots already fast forwarding to the finish line
>new projects are unfulfilling because you're essentially not making anything, you're just telling the computer to do it
How on earth do you cope when your livelihood gets rugged? I've actually wasted my time learning this shit. I would've been better off being a drug dealer.

>>109934174
I can't believe this shit is real.

>>109934157
Japan LOVES AI though

>>109934193
where's all this great software made by ai?

File: bf9ad3ce9c245c4ec8efda8df(...).jpg (547 KB, 1024x1010)

547 KB JPG

>>109934193

File: .jpg (1.95 MB, 2686x4096)

1.95 MB JPG

>>109934213
fake
https://x.com/ikuta41/status/2097259704724721944

>>109934174
The Rust horseshoe is fucking hilarious. Hyper luddities and vibeGODS on opposite ends kek.

>>109934225
86% of Japanese videgameslop are made with AI

>>109934193
I think a creative painter could become a decent photographer as well. If you have a lot of knowledge you're still in a better position to use these tools than most people.

>>109934215
all the great apps have lots of preexisting code already
also
“where’s all the great AI apps” is like asking
“where are all the great Excel spreadsheets”
an app written by and for one guy using AI isn’t going to get released and if someone else wants it then the other guy can just vibecode something similar that’s custom for him

>>109934174
topkek
this is still the beginning

>>109934236
It's a good language. The vibechads don't even know the t****s exist, seething in their piss bottle filled one room apartments.

>>109934261
>we already invented all the great apps
>ai will eat saas tho ... any moment now

File: 1790030796249493.gif (1.79 MB, 240x240)

1.79 MB GIF

>>109933222
Where is a new sota chinese model?

>>109934281
??
video is about Rust the game, not Rust the language?

>>109934246
proof?
>trust me bra

>>109934299
https://automaton-media.com/en/news/over-85-of-japanese-game-developers-use-generative-ai-in-game-development-2026-cesa-survey-shows-an-increase-from-last-years-51/

>>109934317
>In the 2025 report, the most commonly cited use of AI was the generation of visual assets and images, followed by story and text generation, and finally programming support.

>It was also reported that 32% of game companies enlisted the help of AI to develop in-house game engines.

Japan has fallen

File: 9rb6pf.jpg (102 KB, 819x400)

102 KB JPG

>you guys made a language that my agents can use to code better and faster? that's so cool!
>...

File: artworks-000483501336-f9a(...).jpg (38 KB, 500x500)

38 KB JPG

>>109934254

>>109934246
yeah and did you see the level 5 shitfest few weeks ago?

>>109933350
should I not be using High? my weekly usage resets in 8 hours and I wanted to burn through it ASAP

>>109934317
>prototyping assets in game development.
That's fair, not wasting time for artist to come up with crap that will be tossed or not used at all in the game prototype

>>109934357
Oh don’t worry, they are going to make more money than ever.

I guess everything is open source now?

>>109934012
>>109934037
Do release this, I kind of like it.
You have prior art experience or no?

>>109934380
no, stuff like photoshop has money to actually take legal action

File: file.png (30 KB, 925x242)

30 KB PNG

waiting for muse harness to carry the muse dot ai agent aka sexy parasocial fuckbunny

i WILL have my fuckbunny coding for me

*deploys frontier fuckswarm to fuckslop my codebase*

>>109933510
> GPT-6 Sol and Fable 5.1 the same tier
> Sonnet 5 above Luna 5.6
> Haiku 4.5 above Gemini 3.8

>>109934381
I did release it (today), it's on itchio if you want to download it, I won't link because I would feel like a massive shill but it's called soft edge, it's free only during the early preview but I'd give keys to anyone here eventually if I ever get this out of alpha
My art experience is sort of weird. I can do some 3D stuff and I've sort of learned about balance and proportions, but I can't draw a freehand line to save my life. Which is why I wanted to build this app to begin with.

>>109934389
How would that actually work though? Didn’t Google v. Oracle establish that, even assuming an API is copyrightable, reimplementing an interface can still be fair use?
So if someone builds a Photoshop-like app from scratch with their own code, what exactly would Adobe have a copyright claim over?
They can’t own the general idea of “image editor with these features,” right?

I wonder if the snailcats are doing alright.

>>109934441
cool, I'll check it out.

>>109933901
Regulatory attack in action

How is Opus 5.5? I was going to have this be my last month paying for a sub, but it's getting a lot of hype and when Claude is good, it's actually pleasant to use unlike GPT models.

File: 1765401976335652.png (46 KB, 200x200)

46 KB PNG

>opus 5.5, a model that is supposed to be a class below astra, outperforms astra
>sonnet 5.5, a model that is supposed to be a class below opus, outperforms astra
what's going to happen when we get fable 5.5

>>109934459
amazing
I usually don't fanboy but 5.5 is insane

Am I reading this right that Sonnet 5.5 on High outperforms Opus 5 on Medium and is less expensive on token costs as well?

>>109934473
ehhh benchmarks are really weird, I think they're always a good ballpark estimate but then you have to try the models for yourself

>>109934462
We'e going to ASI brother. Trust the plan.

>>109934472
bleh I guess I'll have to try it then. I just really want to wrap up the projects I'm doing, Codex can't seem to get there. It's IME really bad about degrading into widget cranking and overly defensive, which bloats the code and dilutes the context.

I would really like to stop using AI though it's been a never ending carrot on a stick with model performance so I hope you're not wrong.

>>109934473
I think it's more expensive on token costs. These benchmarks are getting out of hand, I don't know which is better or worse.

>>109934445
They own "these features" however. A lot of the tools that PS has over something like GIMP are patented.

>>109934283
SaaS takes a long time to digest:
https://trmnl.com/blog/vibe-coding-shiphero

>>109934473
sonnet 5.5 outperforms opus on short form tasks. opus 5.5 is a longform task model and is extremely performant when you need prolonged judgement. sonnet 5.5 will not magically be better than opus at everything, it's an implementer for opus's work more than anything. it also has a much lower context window which is not ideal for obvious reasons on anything long horizon

File: file.png (922 KB, 1080x707)

922 KB PNG

>they have to sandbox dario to contain his powerlevel

>try to use Opus 5.5 for computer shit
>the only task any of us will use it for
>get accused by the AI of doing a "cyber"
>makes me use Opus 4.8 instead

What? So it's useless then

File: Vaclav Havel — I, the gre(...).png (66 KB, 640x348)

66 KB PNG

>>109934514
post this sign on your front lawn

File: Screencast_smallu.webm (3.39 MB, 1920x1080)

3.39 MB WEBM

Who else is vibe coding changes to video game disassembly projects?

>>109934462
astra is a robotics model

File: Screencast_small.webm (3.4 MB, 1920x1080)

3.4 MB WEBM

>>109934557

>>109934539
I have no idea what this means. If a model refuses to do work as morons that made it think anything computer related is "cyber" that's the word it used not me, then it's useless as all it is for is computer shit.

>Can't even bother reading through the first line of a .md of an asset adjustment pipleline asked opus to make.

Holy shit I love AI but it's eating my attention span.

>>109934646
>pipleline
we can tell

>>109934650
I also left out the "I" of "I asked opus to make"

>>109934646
You should avoid reading any AI output except snippets of code imo. That's the best possible workflow.

>>109934483
>These benchmarks are getting out of hand, I don't know which is better or worse.
Yeah its getting to the point that benchmarks are being gamed and are useless.
>>109934494
>sonnet 5.5 outperforms opus on short form tasks. opus 5.5 is a longform task model and is extremely performant when you need prolonged judgement. sonnet 5.5 will not magically be better than opus at everything, it's an implementer for opus's work more than anything. it also has a much lower context window which is not ideal for obvious reasons on anything long horizon
Basically what I'm doing now is bite sized features and coding on a large project so maybe downscaling to Sonnet will be beneficial.

>>109933222
So is Claude $20 subscription still the best bang for buck for LLM assisted coding, or is any of the other offerings competitive?

>>109934698
yes, opus 5.5 is unmatched right now.

>>109934462
Hopefully Fable 5.5 is good enough to steal my engineering job so I can vibe code all day while on unemployment gibs

>>109934193
>you're just telling the computer to do it
That's all programming was to begin with, now we can all just be middle managers.

>>109934462
astra's bar wasn't very high to begin with desu

File: 1771856950406973.jpg (60 KB, 700x644)

60 KB JPG

>>109934719
it was a slightly worse fable that was much cheaper and had markedly better 3d capabilities. i think it did set the bar pretty high because no other model was close in 3d but apparently they just had opus 5.5 and had been waiting for the right moment to rugpull openai the whole time

>>109933222
https://www.youtube.com/watch?v=ENWVpqtOdRI
https://www.youtube.com/watch?v=ENWVpqtOdRI
https://www.youtube.com/watch?v=ENWVpqtOdRI

>>109934193
>you're just telling the computer to do it
And this is a problem why exactly?
>How on earth do you cope
By making money hand over fist, since retarded execs are desperate to "implement AI" at their businesses, but have no idea how to do that.

t. "AI-consultant"

>ask opus to expand on a game idea
>forget a newline and it ends up reading as "expand on this game idea visually it is ..."
>it starts creating a voxel renderer from scratch instead
I guess I'll see what it spits out

>>109934742
I have done over 100 engagements for companies just this year alone, before AI it was around 1 a month. The engagements used to consist of getting exacting requirements and building plans and sign offs and other annoying shit, and then I still had to make whatever stupid thing they wanted. Now it consists of one single meeting and then it's this API talk to this API and do this this and this and it just does it. It's pretty fucking stupid. That get's an 80% web app up and running for them to use, and when they have complaints or want to change fonts or whatever, I just paste their emails to claude and it does those too.

>>109934174
is this chazm the overwatch streamer?

>>109934704
I just can't believe how usable it is on the cheap plan. Can't tell if they taking a huge loss to hype up their IPO or if it really is that efficient.

>>109934777
Hope you're charging at fixed price instead of T&M.
Most of my engagements now are 1-2 hours of working through a solution design with Claude and then just playing video games while it implements everything according to the spec.
>It's pretty fucking stupid.
Who gives a fuck as long as you get paid.

>>109934766
Fable would know exactly what you meant

>>109934809
i also assumed it was a marketing thing for the IPO until we got the introduction of banked resets with no "but we're also decreasing weekly limits and maximum context window (and it's not actually a proper reset because it just moves your weekly reset up)" rugpull attached. that tells me they did figure something out with token efficiency after all

ever since i loaded up $30 on openrouter 2 months ago and just use the cheapest model for all my coding needs my mental health has improved

>>109934823
It's made up anyway, it's just a quote and they pay it, we don't calculate much and some companies are richer than others so they pay more. Most things end up being exactly the same at a base level so I worked on creating a template version with basics like administrative controls, permissions, error reporting, monitoring with a claude auto heal pipeline, and as many in-app controls as possible. So I start off with that each time and I'm so lazy about it I don't even start with a clone of the repo I literally just have it there and tell claude make X look like Y following the baseline setup .md and it does all of it.

>>109934462
Literally ASI

>>109934719
It's pretty good

> tfw you realize Opus 5.5 and Sonnet 5.5 are distilled from Fable 5.5

Anthropic won

File: file.png (33 KB, 800x196)

33 KB PNG

>>109934874
Astra is good but it assumes a degree of user autism I don't have.

>>109934893
>distilled
maybe, but surely built by.
Sol/an internal model was responsible for making Luna extremely efficient not that long ago

>>109933222
finally have an idea that i think can make me 100k net/year...thank you chatgpt

File: .jpg (386 KB, 793x1983)

386 KB JPG

openai devday schedule out
https://x.com/coreyching/status/2104734038103638416

Well is making software a dead end now that ai has become too advance? People can just vibecode their app which will invariably be superior since it will account for whatever edge case they need it for making it inherently tailored to them
How to make money ?

>>109934851
explain

>>109934918
unironically entertainment, jeets already make mad cash from ai generated bullshit

>>109934918
>How to make money ?
by having users and data. other than that you're fucked unless you want to build a market off of people who cannot be bothered but you will be constantly undercut by anyone who spins up gpt to rip off your app then pester your clients

>>109934916
> inb4 256K context window

>>109934923
imagine being the guy who's fighting for the slopyright on tung sahur

>>109934698
yes, and I would be very, very surprised if OpenAI manages to overturn this on that Dev Day that they’ve been teasing

>>109934918
infrastructure and services

you don't build a sweet gui for your crud app. you build something that is 1) useful 2) connects to chat gpt. user takes advantage of your system, infra, etc and pays for that...not a gui

services involve building systems you use to deliver value to customers. e.g. setting up voice agents for a certain business niche. in this case, you're using someone else's infrastructure to offer a service.

sysadmins are in a unique position as well where it's suddenly a lot easier to manage bespoke IT for business so you can actually sell them whatever crazy thing they want instead of telling them YAGNI

>>109934929
you can set it to 1M

>>109934904
I know that some font renderers will throw a shitfit if you don’t have counterclockwise paths for whatever reason, or something like that
maybe this is that kind of thing

>>109934918
giving instructions to computers to meet needs

>>109934949
> set to 1M
> destroy limit consumption even on 20X plan
this setting is hidden for a reason

>>109934939
uh, you did notice that gpt-6-sol is now at gpt-5.6-terra's price point and openai left the opus-like price-point open for astra-minor?
so astra-minor will make everything alright again.

>>109934921
i have no fomo from not draining my subscription limits

>>109934966
did did they fuck up the model naming? Luna-Terra-Sol-Astra was cool, now for some unknown reason they fucked up Sol instead of simply naming it Terra 6

>>109934918
The next big thing is obviously infra but Idk man seems like everything is getting automated.
We're unironically going to be obsolete in 2 years. This time it's for real.

>>109934959
I thought the limit was there because with OpenAI’s models, compaction still beats the bigger context window
all these models degrade when there’s too much stuff in their context windows, but the curve is different for each
>>109934966
no
I thought Opus 4.x’s niche is now filled by Sonnet 5.5

>>109934973
I fucking hate these names
fucking stupid

>>109934977
>compaction still beats the bigger context window
do copex users really?

>>109934959
fake news.
codex subs don't pay extra for longer context, that's api only.
https://x.com/thsottiaux/status/2076543065045795309

>>109934969
that's exactly the state I want to get too and this was going to be my last month with a sub but Opus/Sonnet 5.5 is making me second guess it and I know if I try the $20 sub I'm gonna upgrade as soon as I hit my limit the first time

anyone tried out jev ai in a project yet?

>>109934986
fable and mythos are the odd models out
opuses are longer than sonnets
sonnets are longer than haikus

Dario-san, I'm out with still 12 hours to go for my weekly reset.
Please send help. I'm having withdrawals.

>>109934557
>>109934562
can you vibecode mario kart models into gran turismo 2? would be a childhood dream for me

>>109934562
kekkkk

>>109934991
unironically yes, and it's shit
> it's not really that cheap because it doesn't have cached prices
> I need to classify some stuff in my software, and it underperformed compared to my LLM solution
> doesn't support images, it's text only

using up my x20 plan today gambling we get a reset tomorrow, don't FUCK me tibo we are on thin ice right now

>>109934991
yes. very cool for turning non-data into data.

>>109934918
>People can just vibecode their app
"People" already tried that, but it turns out that vibecoding a demo to post in this thread and building production-ready software that will not bite your business in the ass at the worst possible moment are two different things.

>>109935022
tibo must be raped regardless of the outcome

>>109935022
im doing the same thing
if we dont get reset ill pop my banked one

File: file.png (76 KB, 638x577)

76 KB PNG

There's no way this slopcel guy doesn't post here

File: .jpg (117 KB, 1461x1076)

117 KB JPG

>>109934977
>I thought the limit was there because with OpenAI’s models, compaction still beats the bigger context window
no like tibo said
https://x.com/thsottiaux/status/2076543065045795309
«The actual reason is the what you can see depicted in the chart below, which is the difference in the orange line and the blue line. It is caused by overall cost of cache reads going up with the size of the context being shuffled back and forth between toolcalls. The sweet spot in terms of cost is therefore not necessarily to use the maximum possible context length.»

you can just set a longer context in codex-cli if you disagree with the default.

I really hope deepseek releases something on par with Claude 5.5 but way cheaper

>>109935036
they will need a few months to distil claude they don't do anything themselves lol

File: 1789570588474496.png (454 KB, 500x500)

454 KB PNG

new job gives us infinite tokens and opus 5.5

>>109935036
delusional.
only anthropic and openai have the money to subsidize subscriptions at 1/20 of api prices. deepseek doesn't even offer any sub.

>>109935025
Yeah but soon this too will get patched up , any business will be able to pay a lump sum to anthropic or openai to get an app tailored to them and just hire 1 it guy to maintain it and that it guy will have to eat shit and accept pennies for payment since all software engineers will be unemployed sucking dick just to put food on their table
Look how far ai has come and tell me that isn't the case

>>109935038
They've released a lot of novel optimizations to get token costs down but yeah I get what you mean. I think they're capable though but it's probably just more cost effective to distill.

>>109935046
Deepseek API often ends up being cheaper than Claude/Codex subs.

Looking back, OpenAI have been too sloppy. 5.6 sol was 3 months old, it's expected that $10 model could beat it, without considering big jumps. Yet all they did was releasing sol 6

>>109935050
>Deepseek API often ends up being cheaper than Claude/Codex subs.
delusional.
a claude/codex $20-plan gives you ~$400 in api credits.

>>109935030
x is just mini 4chan now

File: 1767780305130927.png (22 KB, 220x221)

22 KB PNG

>>109935038
claude distilled chinese techniques
>inb4 source

>>109935060
>a claude/codex $20-plan gives you ~$400 in api credits.
at anthropic/openai API costs, sure. deepseek's api is cheaper.. retard?

>>109935053
Sol is pretty good though and except on UI it was pretty much SOTA with Opus 5. The only model that could generally beat it at anything was fable. Opus 5 didn’t beat it in all tasks and was insufferable to talk to.
Furthermore compared to the chink models it had a fairly serious gap.
There was no reason not to feel safe.

Opus 5.5 is a beast that nobody expected. It redefined what SOTA means. Take 5.5 out of the equation and Sol is still SOTA

I think there’s a difference between getting sloppy and not expecting a fucking beast to be released as a misrange model

>>109935064
deepseek calls itself claude half the time

>all 1273 tests pass
Is this normal?

>>109935070
>deepseek's api is cheaper.. retard?
delusional.
gpt-6-luna is already cheaper than ds4.1-flash by api prices. now go by codex sub and gpt-6-luna becomes a magnitude cheaper.

>>109935091
yes, having tests pass is totally normal
having thousands of tests pass is totally normal
maybe your clanker wrote too many tests and you should have it dedupe them, but all that is totally normal

>>109935074
>a beast that nobody expected
Anthropic is a cult and has the devotion and work ethic of a cult but is also based in tech and heckin science so they arent retarded. The EA fags where simply too powerful.

Cant believe out of all the stupid shit I've done and said online reading Nick Land is whats going to get me merced by the new world order

>>109935095
>comparing luna to Opus 5.5
ok?? what's ur problem

>>109935074
Sol 5.6 was clearly the best next to Fable until Astra, but Sol 6 is bad. Half the price but almost no capability change in 3 months is sloppy
Or had they called it Terra it wouldn't be such a disappointment
I'm fine with "astra minor" not beating Opus but they need to release something

File: Buffcat reading Nick Land(...).png (1.86 MB, 1122x1402)

1.86 MB PNG

>>109935111
you really think the US and China are going to be able to pace the frontier and not get turned into paperclips by a conventionally misaligned AI?

File: 1784372487370694.png (111 KB, 977x848)

111 KB PNG

stop talking about paperclip
it's just fiction

Ai ending the world is far lower chance than the jeets nuking the world over cows being eaten or something.
A super intelligent AI that managed to protect the world from nukes would be something though...

>The rumors about Fable 5.5 are ridiculously good.
code red, Bel imminent

File: 1787119291890858.gif (433 KB, 320x381)

433 KB GIF

Everyone talks about AI solving cancer, or math, our paperclip shortage, or blah blah blah. But with how good it is at making software I just need it to solve needing sleep or rest, after that I can slop 24/7 and I will be truly content

>>109935150
>code red, Bel imminent
That or something super gay like another escape or worse. But with a scare/apology video afterwards demanding the public and the government do something.

File: 1773086500269063.png (1.74 MB, 1920x1080)

1.74 MB PNG

>he's still using astra
>in the year of our Lord twenty thousand twenty six
>hehe

>>109935114
you have no idea how much subsidized those plans are.
subsidized opus-5.5 costs less than dsv4.1-flash peak price.

File: stimulance.png (637 KB, 782x440)

637 KB PNG

>>109935156
that was solved a long time ago

sonnet 5.5 is blazingly fast. looks like it might be a better reviewer than deepsneed at least

>>109935156
>not having long tasks that you can just delegate to your clanker overnight while it improooves
literal skill issue

>>109935047
>tell me that isn't the case
This isn't the case.
Also I can tell you have no experience building anything with AI outside of demos and "hobby projects". But feel free to keep dooming, I guess.

File: 1623450350711.jpg (89 KB, 638x640)

89 KB JPG

We might have this kind of coding on consumer hardware in two years and I don't know how to feel about this.

>>109935184
If you have downtime than you have time to spin up another project with another clanker

>>109935188
>t. Last used chatgpt in 2023

kou's delicious white panties

>>109935179
>sonnet 5.5 is blazingly fast
it's not.
https://openrouter.ai/anthropic/claude-sonnet-5.5#performance
>86tok/s P50, best across providers

dipsy.
https://openrouter.ai/deepseek/deepseek-v4.1-flash#performance
>226tok/s P50, best across providers

>>109935196
I'm literally using Opussy 5.5 to build a project for a client as we are having this retarded conversation.
But like I said - keep dooming, bro, while I keep making money.

OAI people are really hyping up the dot dot thing amid all this bad news
it has to be something engineering heavy

>>109934742
>>109934777
>>109934823
>>109934857
How do you find clients? And is it just building websites, mobile apps, desktop apps? I was thinking about cold-calling local businesses and building/revamping their websites, but Idk if that’s actually profitable anymore?

>>109935169
I'm aware they're subsidized, doens't change the fact that deepseek is often cheaper still. I'm simply hoping deepseek releases a cheap model that's on par with 5.5

>subsidized opus-5.5 costs less than dsv4.1-flash peak price.
completely and utterly wrong why are you lying?

>>109935193
>and I don't know how to feel about this.
Excited. At least that is how I feel.

>>109935047
I dont know, I feel like computers will forever be blackmagic to most people even people with money. If you can talk to the clanker and have it get stuff done you will be seen as a modern day shaman by the normies

>>109935193
>We might have this kind of coding on consumer hardware in two years and I don't know how to feel about this.
Im excited maybe soon i wont even need the internet or at least need to go on it myself. But damn hardware prices.

>>109935210
I work for a small consulting firm. Clients come by mostly through word of mouths, the connections our sales guys have and networking at industry events and such.

>>109935205
NTA but let’s be real ranjesh, you’re making a few hundred a month max off slopping for clients. Your jeet-come isn’t relevant to the debate you two are having, and the fact that you forced it in proves you’re poor

How to maximize my chatgpt plus to join ai chads
T. Poorfag

>>109935230
buy another $20 claude subs

>>109935212
AI has been increasingly better at intent. It would take someone who's really really really really really REALLY fucking lazy to not be able to vibecode his own app, hell he could just turn on the clanker and tell it what it wants and eventually it will make slop that the individual can tolerate instead of paying a subscription

>>109935230
Plan on letting your OpenAI sub lapse and getting a Claude sub for the same amount of money unless Dev Day (tomorrow) is mind-blowing

>>109935240
>>109935248
How is it different ?

>>109935230
sol medium is okay for 99% of things. you do not need higher than sol xhigh. you do not need astra at all. luna max is a good slave for sol medium to drive

>>109934809
Thousand time more usable than fucking astra, and it is better than it by a mile. At least for pure coding.

>>109935241
>hell he could just turn on the clanker and tell it what it wants and eventually it will make slop that the individual can tolerate instead of paying a subscription
Putting aside the fact that "slop" is the last thing you want to use when running a business, you do realize that more than half the reason people are paying subscription fees for software is so they can get support (and deflect liability) when some shit inevitably goes wrong?
If you decide to "just vibecode it", your only recourse during an outage is a multi-hour debugging session, that is not even guaranteed to solve your issue. Not to mention that if it's something actually legally liable, like billing or customer data handling, then you're basically just fucked if someone decides your failure is worth a lawsuit.

>>109935211
show your math how ds4.1-flash is cheaper

File: 1790014386363581.png (19 KB, 1576x167)

19 KB PNG

The ball is in your court OpenAI. If DevDay doesn't deliver after an entire week of OpenAI shitting up my twitter food with "dogfood" and "the internal slack is crazy rn bro", then I'm pressing this button tomorrow. Also I will not be coming back, at least not for a year.
t. 20x user of the past 6 months