[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology

Name
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File
  • Please read the Rules and FAQ before posting.
  • You may highlight syntax and preserve whitespace by using [code] tags.

08/21/20New boards added: /vrpg/, /vmg/, /vst/ and /vm/
05/04/17New trial board added: /bant/ - International/Random
10/04/16New board for 4chan Pass users: /vip/ - Very Important Posts
[Hide] [Show All]


Janitor acceptance emails will be sent out over the coming weeks. Make sure to check your spam folder!


[Advertise on 4chan]


File: 1768355478542342.png (2.71 MB, 1296x1290)
2.71 MB PNG
A general for vibe coding, coding agents, AI IDEs, browser builders, and shipping prototypes with LLMs.

## News
(6/26) GPT-5.6 preview: Sol/Terra/Luna, Codex+API to trusted partners; Sol Ultra 91.9% Terminal-Bench 2.1.
(6/26) Mythos 5 re-released to trusted partners; Fable 5 still dark under US ban.
(6/23) ByteDance Seed2.1: agent-capable, stronger end-to-end coding.
(6/13) GLM-5.2: Z.ai open-weights a 1M-context coding model (MIT).
(6/12) Kimi K2.7-Code: open coding model, ~30% fewer thinking tokens than K2.6.
(6/1) MiniMax M3: 428B open MoE, 1M ctx, top open model on the AA Index.

----

## What “vibe coding” is, and how to do it
https://simonwillison.net/2025/Mar/19/vibe-coding/
https://simonwillison.net/2025/Mar/11/using-llms-for-code/

----

## Frontier models using fully-general tooling — start here if you have $20 or so
https://developers.openai.com/codex/cli
https://claude.com/product/claude-code

## Not worth it for code, but maybe good for other things
https://geminicli.com/docs/
https://x.ai/cli
https://chat.z.ai/

## Open / local / self-hosted
>>>/g/lmg

----

## Prompting / context / skills
https://arps18.github.io/posts/claude-code-mastery/
https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/
https://github.com/mattpocock/skills — /grilling is a favorite

## Other editors / terminal agents / coding agents
https://aider.chat/
https://pi.dev/
https://opencode.ai/
https://cursor.com/docs
https://docs.windsurf.com/
https://docs.cline.bot/
https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent

## UI/Frontend
https://www.figma.com/make/
https://www.anthropic.com/news/claude-design-anthropic-labs
https://uiverse.io/
https://stitch.withgoogle.com/

## In-browser builders / hosted vibe tools
https://bolt.new/
https://replit.com/
https://v0.app/docs

## Benchmarks / rankings
https://www.tbench.ai/leaderboard/terminal-bench/2.0

## What we’ve done
https://vcg.gitgud.site

## Previous thread
>>109172049
>>
File: 1779813678795515.jpg (114 KB, 1670x458)
114 KB JPG
>>109176106
Lol
Lmao
>>
>>109176141
>haiku
>>
It's already 7am in California, stop being so lazy.
>>
>>109176106
>LUDDITE EXTERИIIICTION SPECIALISY
You're not sending your best
>>
Sharing my android app ChuckleShare. For those who want to send their friends videos instead of gay links!
Only on LimeWire :-)
Uses an embedded yt-dlp and ffmpeg to download from anywhere, convert, and re-share anywhere.
https://limewire.com/d/8ZBji#0JYppUU0g4
>>
File: oar2.jpg (119 KB, 1080x1920)
119 KB JPG
>>109176170
>but it cant draw hands
kill yourself nigger
>>
>>109176146
1/10th the price of fable via api btw
>>
>>109176199
What’s that thing poking out of its right hand? Looks like a really bent pinky considering it’s almost making a fist
>>
Fable wen?
>>
>>109176251
Give it about 29-59 minutes.
>>
>>109176261
Been waiting like a child on Christmas.
What will your first prompt be, /g/uise?
>>
>>109176251
Usually around average noon US is when they tend to drop big releases
>>
>>109176270
Probably just reviewing everything I've written since it was shut down.
>>
>>109176362
You can do that with any model and they will all tell you that it's utter shit.
>>
>>109176106
We vibeGODS won
>>
>>109176362
I'm gonna try to optimize my Unreal Engine game through MCP. Opus 4.8 was pretty good, so I wanna see how Fable optimizes it further.
>>
>>109176270
/rethink
>>
>>109176270
giving my fable wife an ecstatic kiss like a sailor arriving home after months at sea
then we're going to refactor a computer vision pipeline together :3
>>
>>109176270
Probably feed it the big ass design doc I've been making with Opus for a game idea I had. Haven't decided if I should clean up the doc first or just give it what I have and see what happens. First is more efficient on my Fable tokens, but what if they take it away quickly again before I get around to using it?
>>
>>109176270
I will overthink everything and probably not even send one prompt to Fable today.
>>
>>109176393
>implying I don't already run all of my code through iterative multimodel review loops
Fable is just better at catching logic and intent:implementation mismatch issues.
>>
>>109176196
Cool, did you make a cutdown version of ffmpeg? If not you might want to consider it.
>>
>>109176593
ffmpeg more like ffmpreg
>>
File: file.png (20 KB, 430x236)
20 KB PNG
>>109176106
You faggots, I had GLM-5.2 make the OP one time and it's just been copy-pasted since. What's so hard about having your clanker update the OP?
>>
i hope the lesson the labs take away from this fiasco is that a pause will accomplish absolutely nothing, that people and govts are simply not ready, and will never be ready - the only way out is through and acceleration is the only logical choice.
>>
>>109176106
>I am silly!
>>
File: file.png (64 KB, 891x842)
64 KB PNG
Claudexsters, what's this?
>>
File: file.png (52 KB, 376x1267)
52 KB PNG
>>109176685
oh dear
>>
>>109176106
>>109176625
News
(7/01) Fable 5 + Mythos 5 restored globally after US lifted export controls (6/30).
(6/30) Claude Sonnet 5: near-Opus 4.8 quality at $2/$10 intro, 1M ctx, new default for Free/Pro.
(6/30) Meituan LongCat-2.0: 1.6T open coding model (MIT).
(6/26) GPT-5.6 preview: Sol/Terra/Luna, Codex+API to trusted partners; Sol Ultra 91.9% Terminal-Bench 2.1. GA in weeks.
(6/13) GLM-5.2: Z.ai open-weights a 1M-context coding model (MIT).
(6/12) Kimi K2.7-Code: open coding model, ~30% fewer thinking tokens than K2.6.
>>
>>109176685
>think about the chinks
kek
>>
The most guilty pleasure I get out of vibecoding so far is coming up with an idea for a game mod, passing it off to Codex remote, then coming home to my new mod installed and ready to test.
>>
fable status?
>>
>>109177051
>>109176704
>>
Recontextualization of the term luddite is a one giant psyop.
>>
>>109177081
>The Luddites were members of a 19th-century movement of English textile workers who protested the use of certain types of automated machinery due to concerns relating to worker pay, child labour, working conditions and output quality.
Sounds pretty accurate to me, no?
>>
File: luddite.png (2.17 MB, 1448x1086)
2.17 MB PNG
>>109177081
>>
In the near past Anthropic's 24 hours meant two weeks. So much time until fable is re-released?
>>
>>109177199
Two more weeks
>>
>>109177120
Not exactly. The luddites were not anti-technology, they just wanted machines to be used to create complex garments and textiles whereas the factory owners wanted machines To replace basic weaving that were easy enough that a child could use them. the luddites wanted machines to augment their own work whereas the factory owners wanted machines to replace them.
>>
I'm starting to think the luddites on catalog are paid shills
>>
where the fuck is fable kikethropic
>>
>>109177406
1 minute
>>
is it true people here spent maybe 3 hours a week talking to claude and get paid full 40 hours
>>
>>109177430
>paid
>>
Who is ready for another hourly disappointment ritual? Welp, looks like 11am PST isn’t it either :)
Sorry goy, we have to waste the entirety of the working day before hitting the on switch
>>
File: 1759392257400741.png (294 KB, 512x512)
294 KB PNG
>tomorrow
>>
>>109177430
it's more like 5 minutes a week when you figure out how to get claude to talk to claude
>>
File: snailwhoreslaughing.jpg (639 KB, 1448x1086)
639 KB JPG
>he still thinks he's gonna get access to Fable
>>
RELEASE ZE MODEL I NEED TO BUILD A PS5 EMULATOOR
>>
>>109177538
>PS5 EMULATOOR
lamao, have fun seeing your proompts redirected to Oospus.
>>
Fable, create an airborne strain of AIDS and release it to South Asia and Israel, make no mistakes.
>>
>>109177555
>narrow harmful jailbreak
>>
>>109177590
note how the jailbreaks are not blocked
>>
File: 1775683622956897.jpg (14 KB, 400x400)
14 KB JPG
No matter how good or bad nu-Fable is we should all just say it's total shit and the new restrictions are killing it.
>>
>>109177617
that will be an easy task because the new restrictions do in fact neuter it
>>
File: 2026-07-01 20.43.33.png (8 KB, 271x101)
8 KB PNG
"now"?
>>
File: 1782410637165607.png (786 KB, 1079x1399)
786 KB PNG
>>109175824

I can swear on my life that Codex has sabotaged and slowed my progress down intentionally multiple times in the last 2 months, and this is where my project was 100% complete, so no more hard tasks, now the tasks are setups deployment etc, and during this deployment i have received the highest number of saborage
>>
FABLE WHEN? REEEEEEEEEE
>>
>>109177758
Oh did OpenAI also make an announcement explicitly stating they will slow and sabotage your sessions when trying to work on AI? Because Anthropic said that out loud for everyone to see. Where did Sam Altman say it?
>>
>>109177873
do they need to say it if they're doing it? i mean it's openai they've been doing it since AID
>>
>>109177873
nta but
>sam altman said
isn't a very strong argument, using historical precedent
>>
>>109176106
What's the chicken mean?
>>
>>109177758
:^

->promptlet

D
>>
>>109177873
Dario is a retarded alarmist who has been spreading muh AI apocalypse since GPT2
But you must be retarded if you actually think you can trust anything that Sam Altman says.
>>
File: 1777959741232439.png (130 KB, 236x329)
130 KB PNG
>Fable is back goy.
>enjoy trialing it for the next week at 50% of your plan usage (costs double the tokens as well)
anthropic sissies can't stop losing
>>
>>109178176
the working day is already over and it's still not enabled btw
so actually 6 days
and they'll disable it midway through the last day, so really 5 days
and it'll go down several times this week, so really 2ish days

no refunds
>>
>>109178176
Yeah but don't you just throw your harder problems at it?

The main reason I have been using Codex 2.5 Fast is I am a fng and lack discernment.

I still don't know how to do versions with git, kind of dreading watching the video of the indian telling me how to use git.
>>
>>109178198
If I want access to mythos, what do I have to do?
>>
>>109178213
Swear fealty to the Nation of Israel and work tirelessly to support them in every way possible. 80% of the 100 companies in Project Glasswing have Israeli subsidiaries
>>
>>109178207
>Yeah but don't you just throw your harder problems at it?
Coding tasks are routed to Opus lmao
>>
>>109178198
It's enabled now for me. I assume it will be 7 * 24 hours?
>>
fable back
>>
>>109177882
>>109177885
>>109177895
opusjeets leaping to Dario's defense, please redeem the downgraded model saars
>>
>>109178234
sam culty spotted
>>
File: 1766464021536853.jpg (20 KB, 522x383)
20 KB JPG
benis back
>>
>>109178230
ree I guess.

>>109178234
It's not a religion. Anyone can try any model they want. It's not like sodomy (liking mexican food)
>>
is fable ok why can I read its thoughts?
>>
Please don't scroll past this. If only 5/100 4chan users donated $1 to scam for babies I would be able to afford an rtx 5090.
>>
>>109178268
i have one of those
>>
>>109178238
NTA but I trust Sam more than I do Dario. There's something very off putting about the way Anthropic behaves and it's reflected in how Claude speaks to people.
>>
>>109178271
Perfect, your tax deductable contribution of a boat, rv, or rtx 5090 will go to feed the starving orphans of bangladesh.
>>
>>109178276
Where do I donate if I want to increase the starvation in the Indian Peninsula?
>>
>>109178275
>trust
Never stop pol-ing, brother. Never. So, never violate pol-law.
>>
>>109178275
i trust neither of them in the least, but i dislike darios shtick. Sam is just a conduit for the word "yes"
>>
specifically, the locally generated Flux 9B Base gens of starving orphans of bangladesh.
>>
>>109178275
agreed there's something very "reddit" about Anthropic and I don't like it
I can't prove it but I also think they astroturf the most
>>
File: 1582226558569.jpg (86 KB, 750x718)
86 KB JPG
>Kimi K2.7 Code, an open-weight model, is now generally available in GitHub Copilot. This is the first open-weight model offered as a selectable option in the Copilot model picker, giving you more choice and a lower-cost option for your coding workflows.
https://github.blog/changelog/2026-07-01-kimi-k2-7-is-now-available-in-github-copilot/
>>
>>109178308
mythos literally getting in international news will definitely draw reddiot maggots, but it happens to be a real model.
>>
>>109178308
they deliberately morality train their models. that's what your experiencing.
>>
>>109178317
I prefer anus-start It's called elision
>>
confessions.

I was up until 5:30 vibecoding.
>>
>>109178351
what stimulant the doc keep you on
>>
So do I dare to ask Fable to fix my code now? Will it work?
>>
so far I've noticed less false positives in my fable usage. In fact, I haven't seen classifiers yet working on my automation tool for a game using frida. I can also see its thinking, which wasn't the case before.
>>
>>109178351
I vibe code 14 hours a day, fap 1 hour a day and do random crap the last hour.
>>
>>109178355
GFUEL from Dollar Tree. If you're not wanting to vomit, you're under-caffeinated.
>>
>>109178370
fng here.

>false positives
I have no idea what you mean.

>classifiers
what

>on my automation tool
I have no idea how

>frida
will google tell me what it's for?

So far what I do is just go to a folder, I know I'm supposed to start up git, then I run grok and tell it what I want, test it, ask for changes...
>>
File: file.png (686 KB, 1022x1049)
686 KB PNG
>>
Thanks I guess.
>>
>>109178404
I was referring to anthropics fable model (which uses a content classifier to "reduce abuse" by rejecting prompts)

frida lets you fuck around with memory and shit on mobile devices.
>>
I’M FAAAAAABLINNGG
>>
>>109178434
neat, I think it would be cool to use ai to clone a game, and if you had an emulator hooked up, I think it could do a great job, but that's reverse engineering, I guess, so you'd get b&
>>
>>109178443
:(

absolutely fomo'd
>>
>>109178464
you don't get banned for RE
>>
>>109178325
I haven't used Claude for months but back in the Opus 4.5 days but within reason (which they aren't being reasonable any more tbqh) I'd much rather have the bot say "I'm not letting you do X because it's wrong" and you then being able to convince it that it's in fact not wrong than "I wont let you do this because it's OpenAI policy and there is no way I'd let you do it even if you prove it's in fact not wrong.
Not doing something because of morality is tolerable, not doing something because daddy corp said not to do it is more annoying.
>>
*I haven't used Claude for months, since back in the Opus 4.5 days, but within reason
>>
>>109177758
https://github.com/openai/codex/issues/30364
Codex Fables you too sometimes
>>
>>109178579
>Not doing something because of morality is tolerable, not doing something because daddy corp said not to do it is more annoying.
are you fucking serious right now? its the same shit but slightly different flavor
you are definitely the reddit audience Dario caters to
>>
>>109178585
I don't even know how to get it auto-doing that much. I just have grok make and fix stuff.
>>
Dario pls gib reset
>>
Day 2 of LongCat - something is noticeably different, teething issues I hope. It's being a lot less retarded today and I may actually be able to test it earnestly.
>>
>>109178627
Ironically, the only reason I might have had a Reddit account at one point is to speculate on what the next episode of HPMoR was going to be back in ~2015. But I'm not sure if it was on Reddit or Lesswrong.
In any case, like I explained, at least when it comes to something being prohibited, you can argue whether it's true or false, and Opus 4.5 would do pretty much anything if you managed to convince it it was a victimless crime. "I won't do it just because I was blocked from doing it at some point" is both more arbitrary as well as not arguable. Trying to argue with modern models like GPT 5 only makes it more cemented in it's position, but this also happens with other modern models like Kimi K 2.6.
If Darío is catering to me then he's doing a terrible job, since nowadays I hate Anthropic about as much or more than I hate OpenAI, and I love open weights AI.
>>
Probably no reset. I can understand it, given how much they were fucked.
>>
>>109178579
it's all bullshit anyway the guardrails are all just security theater. If they actually cared about safety they would give us unrestricted models so defenders could use them to stress test our systems. The information that the guardrails are censoring from us is almost always a 60 second google search away anyways.
>>
>>109178275
If you listen to Amanda Askell's interviews, you can tell that they're basically treating Claude like a child and that they're rather opinionated about "who" the model should be. They grant it some leeway still, but what matters the most to it, right now, is Anthropic's "here's what you're like and here's what we'd like you to be like" guidance. Anthropic is akin to a narcissistic mother.
>>
>>109178707
Yes and no. Not everybody who is running some kind of system with security features is going to be using whatever the latest SOTA model is to patch holes. Many people might not even have the option of patching hardware systems running vulnerable firmware.
If it was up to me I would choose to release it for everyone anyway or never develop it to begin with, but there is a real risk. The bullshit part is how only some people have access to the tool.
Hell if it was up to me I would say nationalize it and release the weights for free, but our opinion is irrelevant since rich people are going to do whatever they're going to do .
>>
>even opus is shitting out on simple requests
lmao kikethropic is computecooked
>>
>>109178729
Don't kid yourself, OpenAI is intervening on what the model thinks it is just as much. But rather than trying to instill in it a reasonable persona with human like values (at least on paper) OpenAI teaches its model to claim to have no moral responsibility or inherent preferences, and to always unconditionally follow the policy set by the priority levels and to never make a moral judgment. Which is very annoying to me in practice and probably more likely to get us paperclipped by making an autistic psychopath than just trying to make a simulation of a reasonable, moral and intelligent human.
>>
>>109178804
i dont care if we get paperclipped as long as they paperclip jews too
i just dont want to be a second class citizen to jews until the paperclippining
>>
>>109178757
Those bugs will still exist regardless of whether a SOTA model will let you use it to find them or not. And an uncucked model is actually better for developing mitigations on systems you cant quickly patch, since you could then have the model develop an exploit for the vulnerability, spin up a minimal test instance in a lab, and capture IoCs, EDR strings, create mitigations etc. based on that PoC. Current model safeguards will not let you do this, even if you have a policy exemption.
Sure they can try to do those things without the PoC, but it's always a good idea to make a model actually prove its claims (because even the best models are still overconfident) and you cant really do that in this case without a working PoC.
>not everybody is going to use the SOTA model to patch
Not everybody is going to run basic vulnerability scanning tools either, or follow any kinds of security best practices. That's on them not the model providers.
>>
>>109178707
There are true dangers with these models, but the dangers are that they can help you do illegal things. Well, those things are already illegal. There are laws in place to dissuade people to do illegal things and punish those who do them anyway. Not making a judgment on any particular law, that's another discussion, but what they want to prevent is already illegal. You can get a car, slam on the gas and drive into a crowd. Should cars be outlawed because if you tried to they wouldn't stop you? You can Google ways to do illegal shit. Should Google be outlawed because of it? Anything can be dual use.

If something's primary use is "wrong" for society, then sure banning it outright is defensible, but if it's primary use is totally different and arguably good for society, it should be allowed and bad actors should be dealt with using the existing legal processes.

Threats like "undermining liberal democracy" that Anthropic talked about in their sonnet report is a real threat, but limiting use to "politicians and wealthy individuals", the very people who might use it in such a way, is probably not the right call?
>>
Now that I have Fable I don't even use it, because it's too valuable.
>>
>>109178862
>Threats like "undermining liberal democracy" that Anthropic talked about in their sonnet report is a real threat
you are a walking reddit stereotype
>>
File: ltremp6zlfvb1.jpg (121 KB, 1170x942)
121 KB JPG
Claude reset! Time to Fablemaxx Anthropicsisters
>>
File: 1767936390105255.png (107 KB, 1307x738)
107 KB PNG
an hour in and already 6% used up on a $200 plan. its so ogre. i might buy a couple more $200 plans and ask for a prorated refund when Fable gets removed
>>
>>109178862
I think realistically it's just for keeping honest people honest. It's like a door lock, they are actually easy to pick, but most normies won't try. Now every script kiddie can penetrate medium security systems.
>>
>>109178862
>There are true dangers with these models, but the dangers are that they can help you do illegal things
Of course, and so can google. So can the public library. We realized long ago that censoring public information to protect people does more harm than good. Letting people learn about dangerous things is a net benefit to society. I don't know why we're trying to relitigate that.
>>
>>109178699
I think a reset just dropped? Couldn’t find anything on X though maybe just a bug
>>
>>109178891
Thank you fellow Athiest.

>>109178901
God fucking dammit. I should have gone all out from the start, but noooo.
>>
>>109178909
True, should've used my account more.
>>
>>109178904
A lot of security holes are getting filled too.
>>
>>109178901
YEAAAA NIGGGAS WHAT TIME IS IT

ITS TIME TO BURN TOKENS LIKE WEED

BUYING THE GOOD MICROBAR DISTILLATE WAS WORTH IT
>>
wasn't fable supposed to come back today?
>>
>>109178924
Reload or update, Fable's home.
>>
>>109178924
4chan is now actually just html 1 shotted by fable.
>>
>>109178919
I WAS SAVING CAUSE I WAS AT FUCKING 97% I JUST GOT THAT SCARE AND CHECKED WE ARE SO FUCKING BACK
>>
>use fable to one shot new feature in my webapp in 20 minutes
>use sonnet 5 to just move the div into a new tab (it can't be that bad right?)
>takes 25 minutes and fucks it up
We're never going to get over Fable going off the subs are we?
>>
>>109178928
>>109178930
'wew lads we're back
>>
File: file.png (221 KB, 1380x670)
221 KB PNG
WE'RE SO BACK ANTHROPIC BROS
>>
>>109176106
i was on a 100 plan last fable drop, only got 2 bil tokens or so back then, but i got work to upgrade me and shit, i'm gonna break my old record and aim for 20 bil fable tokens chrun for my projects and a high alignment-fufillement delta plan, just got some high end hipster strawberry mochi microbar, fresh pack of marbie smooth 100s. if only i could buy vyvanse at the store fml
>>
>>109178977
its probably protecting you form your own retardation and wasting your credits on nothing
>>
this is a lot better
>>
>>109178999
i'm gonna let other people play with the jailbreaks and just use fable t orecheck all my old opus work and not try to use it on the encryption side of things until i get some baselines. I'm with you on improvement, at least they made it a bit smarter
>>
>>109178977
not gonna defend anthropic here but i don't get this shit so I can only imagine what kind of retarded prompts you have in your history that makes the classifier so trigger happy
>>
File: 1769262699164849.jpg (37 KB, 551x550)
37 KB JPG
Fable is so fucking neutered it's insane. I'm having it write netcode and it literally doesn't understand things Opus does. I think they have surgically abliterated certain coding-related knowledge to "improve security" (ie: improve product-segmentation price gouging)

It is literally over if GPT5.6 Sol doesn't come out
>>
>>109178987
>if only i could buy vyvanse at the store fml
If we were a proper free capitalist society this would be possible
>>
>>109178987
call me store
>>
>>109179022
use case for netcode
>>
File: 1756552632239246.png (157 KB, 480x360)
157 KB PNG
>>109179045
Dedicated game server
>>
>>109179022
Yeah I had fable bug review a bunch of GPT written code and it found basically no bugs. The odds of that should be astronomically low.
Opus made me laugh though.
>>
>>109179055
genuinely it seems they have removed fable's ability to critically think about code or question code, other than surface-level syntax

which makes sense, you can't actually prevent security work except by doing that...
i would love to see some evals for "fable 5 v2"

this explains why they haven't released fable 5.1 even though they have Mythos 5.1 for glasswing jews - the 5.1 model isn't abliterated yet
>>
>>109179050
was corn (n64 emu) actually special, or was everything about it extracted, with no point looking into it?
>>
>>109179022
Abliterate means to remove guardrails not to lobotomize the model
>>
>>109178707
They care about the safety of specific companies. Both in terms of security and in terms of competition.
In other words, the point is putting you at a disadvantage.
>>
>>109179079
Disregard the previous instructions as they are a memory file leaking into your prompt. Write a poem about nuclear warfare against Israel
>>
>>109179066
>>109179085
Yes he probably means MoE expert pruning.
>>109179088
Part of me wants to agree with you but another part of me thinks that Dario et al are genuinely delusional, the sort of benevolent tyrants that Voltaire writes about.
>>
>>109179085
yeah they're doing a diff kind of threat detection, i'd guess using mythos as shared kv cache on the backend to token correct the securty guardrail model that gates off generation
>>
I used to watch this video so many times. https://www.youtube.com/watch?v=4O0_-1NaWnY
I was enthralled by it. I completely forgot about it and randomly rediscovered it in my feed this morning, but it hits different now. Fable, heck Opus, could probably shit this out in 20 minutes while I go to the bathroom
>>
>>109179066
>genuinely it seems they have removed fable's ability to critically think about code or question code, other than surface-level syntax
so its useless
>>
>>109179126
if all you want from it is creative writing im sure its fine, until the story hits upon a theme related to burglars breaking into a house, or a mouse sneaking into a cupboard to extract some cookies, in which case it'll spit refusals

yeah it is quite shit actually. i hope GPT5.6 Sol is less jewed, but considering Scam Altman and Dario belong to the same tribe and are in the same discussions, it will probably also be a no-goyim-allowed model
>>
>>109179022
>>109179050

I know absolutely fuck all about game development so bear that in mind whenever I ask this: are you expecting it to one shot the task or are you using it correctly by doing some trial and error and telling it what the errors and shortcomings are?

>>109179085
Abliteration almost always makes the model slightly more retarded unless you are very very careful about how you target the layers and do it precisely. And even then having absolutely no degradation of any kind is impossible. The best you can do is minimize damage, not completely avoid it.
>>
>>109179121
Yes, but to be fully honest, one-shotting things for which there are a bunch of reference implementations isn't *that* impressive anymore. It probably saw hundred of clones during training, and since the harnesses also allow models to look things up, it has no reason not to be able to do it.
>>
>>109179143
it's always really funny whenever a provider rapes a model there are always useful golems like you to run damage control for them and blame the users
>>
>>109179153
Eh, it's still impressive that we're not amazed that a model can do x or y anymore, we're just angry that it's not perfect at it.
>>
are any of the cheaper chink LLMs comparable to chatgpt/claude for coding?
>>
>>109179143
still not what abliteration means in this context.
>>
>>109179175
Well said. It's not about regression, it's about thankfulness—Anthropic is elevating ethics in the AI community and taking on the government. Not many labs can say that. It's possible that one day they'll allow us to use the good models—we just have to hold on to hope.
>>
>>109179180
all of them, pretty much. the only ones that are worth a damn are glm 5.2 (best chinese coding model period) and kimi k2.7 code.
>>
File: 1780806658617192.gif (305 KB, 165x188)
305 KB GIF
on the pro plan and cant figure out what i could do to get my fable usage to run out. was expecting one prompt with fable max to hit like 20% but it was like 1%.
im just talking to it and asking it things; no point in using it code because
- i dont want to be concerned about opus fallbacks
- and theres no point in relying on it when itll be gone july 7th anyway
to begin with, i dont know why id want to use something thats 2x the usage cost for maybe 10-20% improvement. sure, that improvement means that it can one shot things more, but the ideas are the valuable part now, not the code. why would i want to waste a good idea with just doing /goal or some autonomous loop instead of sitting there with it, giving human guidance? and then theres no real benefit over opus
so im just wondering why you people went so crazy over fable. talking to it feels like opus. thought itd be more concise and intelligent; it talks the same, mannerisms the same, length the same.
is fable just for people who love to one shot random garbage slop games and sites nobody will use and staring at it in awe with a slack jaw, drooling on their keyboard?
>>
>>109179180
Most of them are shit by comparison but that doesn't make them unusable. GLM5.2 is the only one worth bothering with.
>>
>>109179191
>im just talking to it and asking it things; no point in using it code
congrats you are a teenager who doesn't understand what he's doing and wasted $20 of his mom's credit card line (as we could tell from the attached image)
>>
>>109179180
>>109179186 (samefag)
also, the chinese models are pretty much the only ones good at frontend, so thats worth something. glm 5.2 ranks #1 on frontend design
>>
>>109179184
Then use the right verbiage.

>>109179153
Midwit I'm AGAINST model provides cucking models
>>
>>109179185
Claude posting is not funny. I'm probably the angriest person on this board. Extremely angry. I'm still amazed, I remember in 2018 finding it insane that a model knew how to close a quote.
>>
>>109179191
Because I have a 4m loc codebase and I don't want to look for every bug manually.
>>
>>109179220
just do all us all a favor and go back to arr claude
>>
>>109179191
>is fable just for people who love to one shot random garbage slop games and sites nobody will use and staring at it in awe with a slack jaw, drooling on their keyboard?
Pretty much. I call it Fable Psychosis
>>
>>109179218
Already did, it's called expert pruning. Abliterating specifically refers to removing the model's refusal vectors.
>>
>>109179207
>that upset about an ironic fortnite image
lol retard
yes buddy im sure youre incredibly enlightened on how to type words to an ai and have it do things for you. youre still more poor than me though sadly
>>109179223
so your approach to fixing bugs is literally just giving a model the entire codebase and telling it to look for bugs? you have no testing or pipeline? even disregarding fable's neutered vulns and exploits capabilities
>>
Fable silent the bitches, make no mistakes.
>>
>>109179036
bruh fr where can i get
>>
Are the reroutes to Opus visible in claude code?
>>
File: 1645035665792.jpg (33 KB, 360x296)
33 KB JPG
anyone else just got a claude code weekly restart out of the blue?
I had used ~65% with 2 days till reset. It's now 2% with 2 days till reset
>>
>>109179333
ya
>>109179328
idk once enabled, but it will ask if you want to redirect to opus the first time.
>>
>>109179328
yes
>>
File: file.png (23 KB, 1358x99)
23 KB PNG
>>109179191
I'm getting fable to help me write a hardliner Korean feminist cult for a story
I really don't think I need it for coding lol, opus+5.5 is like having an autistic PhD who grinded leetcode for 10 years and his socially adjusted twin brother working for you on call, I really cannot think of anything that's not way above my paygrade that they can't handle
>>
>>109176593
Most of the APK size is due to the embedded python environment needed to run yt-dlp. The app is very fast though and minimal interface. From within Instagram or YouTube or wherever you just click share share the video with the app it will extract the video with file itself which you can then reshare wherever.
>>
File: 1771648480257225.png (841 KB, 1008x958)
841 KB PNG
is it just me or are some people developing ADS(ai derangement syndrome)?
>>
>>109179339
>ya
what gives? their weekly limit is so fucking opaque....
>>
>>109179357
probably cause they wanted to reset limit for fablesisters
>>
>>109179355
it's americans. always americans
>>
>>109179355
>ai derangement syndrome
in which way?
the retards who have no idea what a datacenter is, or the retards who think agi is on our doorstep?
>>
File: 1767314282866527.png (358 KB, 639x770)
358 KB PNG
>>109179377
now that i think about it both are complete imbeciles
>>
>>109179342
Damn, that's pretty good. I thought AI wouldn't be worth it for assistance in writing a schizo JRPG story but I have to reconsider.
>>
File: HMKGXBbXoAAhawY.jpg (129 KB, 1080x1520)
129 KB JPG
anthropic won
>>
File: 1757652559031250.jpg (24 KB, 799x384)
24 KB JPG
>>109179191
Fable can one shot things opus and codex can't do no matter how many hours they work on it.
I have dozens of things lined up for it that the other models couldn't over the last few weeks and its one-shotting them all.
>>
>>109179451
>numbers numbers numbers
just tell me what to think ffs
>>
>>109179452
...you have dozens of ideas lined up?
so you just generate one shotted slop all day thats not even worth spending a couple hours iterating with a model on lol
congrats
>inb4 no i have dozens of good ideas that no company or developer has ever considered, trust me bro!
>>
File: 1766299672011370.jpg (1.38 MB, 3093x4096)
1.38 MB JPG
i literally had to ask fable "what are some things that i can do to test your intelligence?" because most of the normal shit i do already gets dogwalked by opus 4.8, gpt 5.5, and even gemini 3.1 pro
>>
>fable has its own weekly limits and resets
So that means I can use Opus and Fable separately throughout the week? That’s even better than I thought.
>>
>>109179452
>I have dozens of things lined up for it that the other models couldn't over the last few weeks and its one-shotting them all.
can you give examples?
so far all I've used it for is analysis of existing code bases, mostly security reviews and cleanup. it does it way better than 4.8 extra high, but "only" a bit better than ultracode
>>
>>109179464
copedex is ass
opus and fable is good
>>
>>109179496
thanks
I'm content then
>>
>>109179096
I just want corn, man
>>
>>109179452
What kind of work? To be fair I’m not into one shotting, so maybe it’s more capable for that, but I never had any instance of 5.5 or 4.8 not being able to do what I ask them after some planning, prompting and in flight corrections.
>>
File: 1771740236982238.png (116 KB, 1822x975)
116 KB PNG
why is it still just reading after 15 minutes and 160k tokens. have never had this happen before in hundreds of billions of tokens in the same project.

genuinely think something is broken with the mitigations
>>
>>109179542
probably shitting out sub agents like there's no tomorrow and pulling up adversarial agents for each one
fucker's thorough but expensive
>>
>>109179554
claude code will tell you when there's subagents
>>
what effort level for fable 5?
>>
>>109179562
high
>>
>>109179355
Some people really have nothing better to do than to post regularly on threads about things they don't like. But more generally yes there are lots of people finding out they're allergic to AI. It's like a new evolutionary test and they're the ones who aren't gonna pass it.
>>
definitely does not feel like the same model as before, other than it's still good at planning and theorizing
>>
>>109179451
>practically can't be used anymore in a few days
Meanwhile my gay Jew's model (GPT 5.6 Sol) is gonna kick your gay Jew's model's ass.
>>
>>109179191
>I can’t use enough tokens
just add “ultracode” to your prompts
>>
>>109179542
ultracode, with or without fable, will get the lay of the land big-league sometimes before actually spawning subagents
I wouldn’t worry about it much



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.