Fable drops, is so me super special ultrapowerful model that solves software engineering, it develops a bunch of freaking amazing looking vidya demos, people post rave reviews, and the usual suspects preach how software engineering is over etc.What if what actually happened, is a skilled set of coders built in a bunch of demo apps and frameworks into the model, like 50 top most popular video games, and handbuilt clones of popular webapps, and what's actually going on is that the model can 'customize' these templates to a great degree, but not actually build this shit from scratch.I remember using Claude way back to write some fiction an ERP shit for myself, and while I was impressed by the inventiveness of plot points, and witty dialogue it produced, after like basically a week of spamming this shit nonstop I realized it kept recycling the same ideas, and what it could do was actually extremely limited.Of course we can't subject the model to deep scrutiny since its 'too powerful' and thus the everyman isn't allowed to inspect it deeply, but I suspect if people kept digging, they would find the same shit.While I suspect the underlying model is more capable than 4.8, its very possible there's some smoke and mirrors going on, which, with the IPO closing in, would be something Anthropic is incentivized to do.
>>109065235Fable was good i tested it it used 2x more tokens than opus but seemed more efficient because one shooted solutions.As for what is going on Anthropic clown shooted himself, AI is a tool if AI breaks encryption or public software and hacks it ,the solution is to fix it not to ban the tool that breaks into code.Just like you are not going to ban planes because they invade people privacy by allowing you to view their backyard or drones for same reason, sure you can add security meassures as drones have so no one uses them over airports but still it's just a tool.The clowns fucked up tough they had a 6-9 months advantage over everyone and OpenAI closed that to probably 2 months advantage by rushing the last months up to chatgpt 5.5.They kept larping their model is too dangerous bla bla bla, now even it's creators some foreigners can't use it, go sue the gov against that after you spent months saying it's too dangerous lmfao.The us gov probably can't do the same against openAI because if sued OpenAI would win as they never said it was dangerous, anthropic only chance now is to keep updating opus in a slow way (larp) up to mythos levels otherwise they probably can't even legally defend themselves.Also chinese models are not that far behind probably a year and half behind but that's still very advanced.
>>109065235Fable aside, the Claude code agent with Opus is pretty fucking solid at understanding what you want and understanding your code. Claude is also amazing at creation and implementation. More so than any of the other top agent's liek Codex, Gemini and Grok Build. Grok build has some advantages but isn't as "intelligent" as Claude opus. I tried Fable for about 10 minutes so I didn't notice anything special other than it was faster .
>>109065520i didn't follow it that closely but i did catch all their marketing about how "we had to add guardrails" and so forth and i wonder if that backfired.
>>109065695Fable was more to the point and faster. Opus is really wordy.
>>109065235If you suspect shenanigans, suspect it being n Opus in a trenchcoat with a small classifier that selects the best predicted answer, not this.
I have been working around and trying to come up with something to fill in the need of Fable 5 after Anthropic pulled it, atleast for a long time for Non-US people.I present fusion-plugin which gives almost Fable like Performance with a mixture of subagents on a particular task with different models being used, works on Antigravity CLI or Claude Code.https://github.com/ProxyAyush/antigravity-fusion-plugin
for me, it's GLM 5.2
For me, it's Grok
no shit that's how llms work, they regurgitate shit from training data/internet. All those minecraft clones and shit they make in a day is just git clone from some project lol
>>109065235I don't know what you're talking about. I have yet to see vibeslopped project with receipts, as in source code and playable demo.I don't frequent AI spaces but I assume AI jeets would have spammed them here if they existed.
>>109065235Fable was that good, that wasnt a lie. It oneshot my project that 4.8 was struggling on. I fear itl be replaced with Opus 4.9
>Fable drops, is so me super special ultrapowerful model that solves software engineering, it develops a bunch of freaking amazing looking vidya demos, people post rave reviews, and the usual suspects preach how software engineering is over etc.this did not happen
>>109065695Fable aside, the Antigravity code agent with Gemini is pretty fucking solid at understanding what you want and understanding your code. Gemini is also amazing at creation and implementation. More so than any of the other top agent's liek Codex, Claude and Grok Build. Grok build has some advantages but isn't as "intelligent" as Gemini. I tried 3.5 for about 10 minutes so I didn't notice anything special other than it was faster .
AI will pick our jobs, interview us, pay us, educate us, make our art, make the justification of a secular ethics, and we will agree to it because we will increasingly depend on it. HOOOOLLLLYYYY COOOOOKKKKEEEEDD!!!!!!
>>109065235My co-worker who is the biggest AI fan in the group spent that one day with it, and didn't like it. He found it did more than he asked, launched more agents than he asked for, and had them doing things he didn't ask for...Then when he asked it to add a new endpoint to his project, it removed all the existing endpoints in the process. As I pointed out, a local model would delete his endpoints for free.
>>109066129>all those gaudy animationswhy are jeets so understimulated all the fucking time? they need to bombard their senses with diarrhea at all fucking times, top ADHD tiktokbrained shitalso how fucking low is your IQ to actually believe what you just said?
>>109066129Indian hate is fully justified.
>>109069084it did happen, are you fucking blind?
Fable is just an overhype piece of shitThey brought it down because it is bad not because of the security shit
>>109065235In leaf land, I'm not allowed to use Fable. It's "so good" that they had to impose export controls on it.
>>109065520The only privacy is in your head. All codes are fundamentally breakable fixing the problem will lead to somebody else finding exploits for it.
>>109065235its always a sales pitch and the sheeple eat it up everytime. Sam Fagman always says their next model is nearing AGI and the like stink on shit the retards move in lthinking they havethe next iRobot on their hands only to be disappointed it cant tell you how many r's are in strawberry
Maybe I am the retard here, but how is a LLM "intelligent"? Hummus beans are considered intelligent years before we have learned to speak or read any language. Sort of seems like they are doing it ass backwards.
>>109075848imo the question itself is kind of meaningless i think
>>109075858You say that, but how is AGI/ASI supposed to emerge if literally no other "intelligent" lifeform has the same path as LLM's to "intelligence"? It's not like we picked a biological analog and ran with it. Just odd that we are essentially working backwards, and it seems counterintuitive.
>>109076075it is one of the fuzzy word that doesnt have solid meaning across the domains and never will resolve in any productive or meaningful debatemaybe compression/decompression algos are intelligent in a way that we cant take much practical advantage of other than reducing filesize
>>109073174yeah i must be blind, hence why i cant see any example of this happening
>>109067442This can't be stated enough. Before LLM marketing, non-devs could have learned to do git clone and tweak 3 parameters in a well commented config file and that would have achieved 100x more than they're getting TODAY with their use of LLMs. Yet they refused to do "that nerd stuff". Now that LLMs have been marketed to hell, they get shitty cobbled together copy-pasted segments of these same repos, no proper build system so they have to do much more complicated run and setup steps, in the end they get much worse quality slop, and they still lap it up.Also, fable was mediocre as fuck. No better than 4.8, and 4.6 is still best. That's even if you managed to get through its retarded 'guardrails'. Oh and all these demos and shit that were totally fable and shit? The vast majority were impossible because they would have hit the guardrails and auto downgrade to 4.8 if you tested it.
>>109071293>He found it did more than he asked, launched more agents than he asked for, and had them doing things he didn't ask for...Yes, same experience and same thing others have told me as well.>Then when he asked it to add a new endpoint to his project, it removed all the existing endpoints in the process. Yes, it's the same problem as the above.
>>109074595Fake, they blocked it everywhere including in the US. It's all a marketing ploy anyway. Same as gpt-2 being too dangerous to release.
>>109076113LLMs are fundamentally very advanced compression algorithms. Weights in a baseline pretrained model are exactly a highly compressed (lossy) representation of all the data in the training. That is by design of the math of training such models in this way.
Anthropic has to generate hype as they ready for the IPO rug pull. They know their business is unsustainable, everyone knows that Anthropic, OpenAI and all the other cloud model companies are a money sink that'll never recoup the R&D costs, so they're now preparing to grab the bag and run before the gig is up.At least it'll mean the hardware pricing will stabilize once those idiots stop eating up all of TSMC's, SK hynix's, Samsung's and Micron's production capacity, and open source development of local models can finally push us into the golden age of AI. Not this fake grifting bullshit.
>>109077195i know
>>109077250good
>>109065695>More so than any of the other top agent's liek CodexThe astroturfers keep saying this but I'm not seeing it
>>109077209The IPO is almost half a year away, sometime in Oct to Dec. This government rug pull is of no benefit to them.
>>109074759>The only privacy is in your headNonsesnse i am 100% sure humans brains are more hackeable than AI code at this point not sure why a model could not be trained to decode brain signals.
>>109065695Grok Build new cursor 1.5T model literally about to make them the top of the chart in 2 weeks
>>109065235For me it’s Grok Build with cursor composer 2.5 but I’m just waiting for the 1.5T model which I’m hearing is already better than gpt 5.5 at a much lower cost and they only way sammy can save his ass is drop gpt 5.6 or whatever in July first week Opus is in big big trouble now bc they banned fable and they have no models left to release lOOOL
>>109077347If anything its bad for them because they were about to make it paid per token as of today I think where they would just give the world week of full access with the paid plan and then switch to the extra expensive mode. They are losing tons of cash.
>>109077347Exactly the opposite and I laugh at you for thinking it's the government doing this and not the usual corpo shit going on.The only thing that's not of benefit to them is amazon playing their jew games to anthropic's detriments.
>>109077325>astroturfers how swell
>>109077188Except this time it's blocked and not coming back for general use, only for companies that can meet the security requirements. So basically how Mythos was deployed.
>>109079742Of course, it's how they plan to drum up hype for investors. That's how the game works. "We are the best, proof is that only our model got banned" "imagine the money we'll make when people have access once we get through the legal"Remember: the ENTIRE AI investment thesis is predilected on the idea that only one will remain standing, and that one will eat the entire software world. With this play, anthropic is suggesting evidence that they are the one, not openai or google or xai.
>>109079761Well, it's the government blocking it though, unless you're arguing they're in on it.
>>109079828current government is pure p2w lol.
>>109079828No, it's not the government. Companies do this via things like lobbying and other similar tools. Another thing you need to look up: "regulatory capture".
>>109066129kys you stupid brown retard
>>1090771704.5 is best.
>>109077170You are wrong. If you were right then:1 - Models before Fable would have not struggled2 - Fable would have not struggled because 90% of the time it just puts up voxel renderer with noise worldgen, slaps around some trees and then makes you hotbar UI where you can place blocks3 - If you look at the code it is not the same, sure it uses the same noise algorithm but thats about it, the whole code is completely differently structured, using completely different libraries and completely different languages. Its like artists claiming Diffusion models copy when they learn that tree has a brown trunk and green crown and then generates completely new tree from scratch.
>>109071190gemini is slow and cant figure out simple problems. when it comes to something like migrating a project from one framework to another. claude converted a bootstrap app into tailwind no problem. gemini kept picking wrong classes, making up classes that didnt exist, and telling me it made correcitons that it never made.
>>1090809951 - they don't struggle as much as fable does, correct2 - one doesn't imply the other3 - the code literally is 1:1 ripped off in most places actually if you look at the code. But the fact it isn't LITERALLY 1:1 ripped off everywhere is, ironically, why AI is a meme and can't do anything right.
I never used fable but using opus every day and i dont see anything unrealistic that it implemented a minecraft clone in the web, which are already dime dozen anyway
>>1090820841 is straight up false, vibeniggers have been telling LLMs "make me Minecraft" since the release of ChatGPT and only with Fable it gave results that are even playable2 is whatever3 nothing AI does is 100% ripoff because it does learning similar to what animals do, where neurons or virtual neuron-like nodes remember patterns that can then reverse engineer whatever they trained on reverse engineering, but they can do it with next to everything, you dont just have to have a Minecraft clone, you can have shitty minecraft clone with gambling dragons or Shrek musical or whatever stupid shit you can imagine and it just does it to certain degree
>>1090833871 is straight up false, vibeniggers have been telling LLMs "make me Minecraft" since the release of ChatGPT and always claimed it totally worked and gave perfect results with each new AI announcement2 is acceptance for your surrender3 I have a PhD in deep learning from yoshua bengio's lab. You don't even have a cursory understanding of the most basic concepts. You don't even know what neuron means in the context of a neural network in the sense used in deep learning.
>>109083409>I have a PhD in deep learning from yoshua bengio's lab. You don't even have a cursory understanding of the most basic concepts. You don't even know what neuron means in the context of a neural network in the sense used in deep learning.What he said is broadly correct even if his understanding is shallow. DL has shown that modelling stuff like spiking, neuron types, neurotransmitters is simply not computationally relevant in practice. We don't know if animal brains implement SDG per se, but any sufficiently general learning algorithm will look like a compression algorithm when viewed at the right level of abstraction.
>>109084041What "he" said isn't even remotely correct in the absolute most generous interpretation of the sense. Literally on the level of saying gravity is like glue, at best. Didn't bother reading further. Go snort some glue or fuck a goat or whatever your usual activity of choice is, Cletus.
>>109080995>the whole code is completely differently structured1000 monkeys and a thousand typewriters. Not intelligent just random bullshit
>>109084106>What "he" said isn't even remotely correct in the absolute most generous interpretation of the sense. Literally on the level of saying gravity is like glue,More like saying that gravity is like electromagnetism, which is trivially correct. But since you lack the reading comprehension to understand the rest of my post, I guess we can end it there.
>>109084108You know you are wrong, I will not even bother responding to you beyond just pointing out that LLM output is factually not noise
