Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Fable 5 is Back

twitter.com

Post reply on HN
Print view View original

301–310 of 438 posts

Re: Fable 5 is Back

#301

Post by themgt »

skerit wrote: post #89
gregw134 wrote: post #31

Earlier quoted context omitted.

I'd imagine they want to do price segmentation. Sell the best model for $50k a year to corporations willing to pay full price, keep the rest of us on a lower tier. Gotta pay for that infra somehow.

I just don't really understand the entire strategy behind this. Or their horrible, horrible communication. Because right now it's as if Fable/Mythos 5 is "the end of the line". It's as if this is the best their models are ever going to be. So what the hell are we going to get next? All of their models will forever inch closer to Fable, but never reach it? That doesn't make any sense. It all seems so dramatic. Instead…

afaik there's somewhat painful economics. Not sure back-of-napkin but something like:

   • 150-500B: Sonnet
   • 0.9-2T: Opus
   • 3-5T/10T: Fable / Mythos
So if bigger model is "smarter" but you effectively wind up with a "shared hosting" model where a coherent inherence node(s) that cost $2m or something can run max 10x customer workloads simultaneously ... not sure what that can be priced at.

If it turns out a $10m/10x shared node can host even smarter models, then what?

Re: Fable 5 is Back

#302

Post by neuronexmachina »

noosphr wrote: post #267
himata4113 wrote: post #194

I think this is as good as time as any to bring up that fable/mythos weights are one mistake (malicious or not) away from being leaked to adverseries or available in a random torrent. Imagine this, fable weights are likely distributed to hundreds of datacenters with likely thousands of people directly or indirectly having partial or full access. I just don't quite buy that a 'world ending' fable/mythos model would be…

We've been told models are too dangerous since gpt2. There comes a point where you not only want the boy to stop crying wolf, but hopefully be eaten by one.

That example is brought up a lot, but in retrospect the concerns about Gpt2 were pretty valid: https://techcrunch.com/2019/02/17/openai-text-generator-dang...

>OpenAI said its new natural language model, GPT-2, was trained to predict the next word in a sample of 40 gigabytes of internet text. The end result was the system generating text that “adapts to the style and content of the conditioning text,” allowing the user to “generate realistic and coherent continuations about a topic of their choosing.” The model is a vast improvement on the first version by producing longer text with greater coherence.

>But with every good application of the system, such as bots capable of better dialog and better speech recognition, the non-profit found several more, like generating fake news, impersonating people, or automating abusive or spam comments on social media.

Re: Fable 5 is Back

#303

Post by megous »

tuwtuwtuwtuw wrote: post #69
levkk wrote: post #58

I cancelled Claude. The harness is kinda broken, GPT 5.5 is good, and GLM 5.2/Deepseek is good too (with pi, especially). Just not worth the trouble. And I'm not going to pay two subscriptions.

How is the harness broken?

Recent update resulted in 100% permanent CPU load in the background, when doing nothing with the app. Basically unusable on a notebook (hot+noise, unexcusable for do nothing "api/command coordination" terminal app). Dunno if they fixed it, since I could not use it on the notebook and I do not have this issue on my workstation.

Re: Fable 5 is Back

#304

Post by janalsncm »

SparkyMcUnicorn wrote: post #269
janalsncm wrote: post #261

I never strongly considered open weight models before getting throttled all the time from Claude. I bought a 1 year pro subscription but my bet is I won’t renew it. The Claude “usage” UX is very bad. At the most basic level, there is no way to know what you’re actually paying for if you buy a “pro” plan or a “max” plan. Dario will take your $200 but he will give you a secret number of tokens in return. I have no way…

If you use Claude Code, there are plenty of projects/tools that let you see what your usage would have cost at API rates, then you can cross-reference that to your "usage" in Claude. I run through enough token in a few days for the subscription to pay for itself. But I agree that advertised visibility into this would be nice, so we could compare the different providers up-front.

1) There is no way to cross reference actual usage because Anthropic won’t tell you how much they think you used.

2) It’s not a nice-to-have, it is a basic accounting question of how much of a product you are getting.

3) Even if it did work, it should be a first-party feature not a third-party add on.

Re: Fable 5 is Back

#305

Post by replwoacause »

madamelic wrote: post #172
pkoird wrote: post #144

I gave it a book on human consciousness I was writing and it flagged it. This model is hilariously bad. Anthropic has defanged this model to the point of malice. No way am I paying to use something that is basically useless.

Today I told Sonnet (!) to use a browser MCP to enter a username and password for the project it is working on, it told me that it can't do that because it violates its security protocol. This worked fine before. I love Claude, I have stuck with it even through people saying Codex is better but this is definitely getting to be the last straw. It's completely absurd I am paying them $200+ per month along with pushing…

I think the 5 series of models suck kinda. I'm sticking with 4.6 for as long as I can. But GPT 5.5 fixes me up nice like too.

Re: Fable 5 is Back

#306

Post by replwoacause »

matheusmoreira wrote: post #124
logancbrown wrote: post #91

Earlier quoted context omitted.

Amusing to see OpenAI being the "good guy" in the end

They didn't become the good guy, Anthropic became the bad guy instead. The good guys are the chinese firms releasing open weight models.

Yeah this is wild stuff. Seeing Anthropic free fall after building up such good will is shocking.

Re: Fable 5 is Back

#307

Post by ofjcihen »

pkoird wrote: post #280
ofjcihen wrote: post #182

Earlier quoted context omitted.

It’s incredibly ridiculous that it won’t help with that for me either sometimes but yet I’m also sitting on 3 surefire ways of jailbreaking Opus 4.8 that I use for cybersecurity assessments and pentesting

I'm not saying you are on a list now, I'm just saying if you were now to be on a list, I wouldn't be surprised.

Nah they definitely know of these methods there’s just not a way to keep people from doing it unless you neuter the models à la Fable

Re: Fable 5 is Back

#308

Post by Breza »

noosphr wrote: post #267
himata4113 wrote: post #194

I think this is as good as time as any to bring up that fable/mythos weights are one mistake (malicious or not) away from being leaked to adverseries or available in a random torrent. Imagine this, fable weights are likely distributed to hundreds of datacenters with likely thousands of people directly or indirectly having partial or full access. I just don't quite buy that a 'world ending' fable/mythos model would be…

We've been told models are too dangerous since gpt2. There comes a point where you not only want the boy to stop crying wolf, but hopefully be eaten by one.

And Guinness thought the t-test was too dangerous to announce publicly over a century ago

Re: Fable 5 is Back

#309

Post by cjbgkagh »

dakolli wrote: post #262
himata4113 wrote: post #215

Earlier quoted context omitted.

Cyber-warfare is pretty world-ending these days. Our relatively peaceful world is built on top of mutual destruction. See: https://news.ycombinator.com/item?id=48633023

Why would you believe anything an intelligence agency says? Half their job is spreading fear and obfuscating the truth.

Half? What’s the other half?

Re: Fable 5 is Back

#310

Post by steve-atx-7600 »

petra wrote: post #26

Maybe, for some projects, instead of generating code with it, it would be useful to generate a plan and the loop(tests/formal verification),because those take much less tokens than a full project, and than use the loop using the older models ?

Did that today on a 2 repo affecting project of the kind where I already set the right design for one major use case and I needed Claude to create a superset of that use case that was not substantially different: after plan I had about 10% of 5h context left for fable 5 and this was the only thing I worked on. Hard to generalize this of course.
Post reply on HN