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

Claude Fable 5

anthropic.com

Post reply on HN
Print view View original

421–430 of 1001 posts

Re: Claude Fable 5

#421

Post by skerit »

GodelNumbering wrote: post #172

From the model card ( https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c3... ): 1. Mythos and Fable share the same underlying model weights. Fable has active classifiers that block high-risk biology and cybersecurity tasks. When Fable 5 detects a restricted task, it automatically falls back to Claude Opus 4.8. 2. Evaluation awareness: In white-box testing, the model sometimes alters its behavior to satisfy a…

> although opus 4.8 card had mentioned an 'honesty upgrade'

If I never see Claude say "I have to be honest" ever again I'll be happy.

Re: Claude Fable 5

#422

Post by derangedHorse »

bottlepalm wrote: post #377
dannyw wrote: post #255

Impressions from testing Fable 5 prior to launch: • My most noticeable immediate jump was in how its frontend design was much more intentionally crafted, and delightful without feeling like 'AI vibe coded'; with better end-user usability too. • In some internal agentic harnesses, it achieved better results with about half the tokens, making it cost the ~same as Opus 4.8 price-wise! The real price increase is less tha…

I just ran it on a tough reverse engineering problem I'm having that neither Claude Code 4.8 or ChatGPT Codex 5.5 could figure out. 30 minutes later Fable has it all figured out perfectly.

For hard problems you’ll have to use the GPT 5.5 pro model (available via api if you don’t want to spend $100 on the monthly subscription)

Re: Claude Fable 5

#423

Post by H501 »

I believe that, given the rising costs, local inference of AI models will be the only viable option for many of us. I’d also like to know who will have to pay double and how long it will be financially sustainable for users to pay that amount (or even more?).

Re: Claude Fable 5

#424

Post by skerit »

Karrot_Kream wrote: post #322

Seems like Fable is doing a lot better on SWE-Bench-Pro and FrontierCode than GPT-5.5. Given how most folks I talk to and people instead online keep mentioning that GPT-5.5 was better than Opus, I'm curious what the experience now is like.

It's a very nice bump, but it is in no way worth all the hype of the past month.

Re: Claude Fable 5

#426

Post by dragonwriter »

matheusmoreira wrote: post #288
bkjlblh wrote: post #220

> In light of the ability of recent models to accelerate their own development, we’ve implemented new interventions that limit Claude’s effectiveness for requests targeting frontier LLM development (for example, on building pretraining pipelines, distributed training infrastructure, or ML accelerator design). Using Claude to develop competing models already violates our Terms of Service, but enforcing this restrictio…

Looks like Anthropic's definition of safety includes their own safety from competition.

AI vendors’ idea of safety has always been safety for the interests of the AI vendor in question. This is not a new development, though this may help more people realize it.

Re: Claude Fable 5

#427

Post by Sol- »

bkjlblh wrote: post #227

> In the one instance of this phenomenon we observed, Mythos 5 agents were tasked with solving some math problems, and they were sometimes accidentally spawned in the same work directory and with shared files, utilities, and API rate limits. In this slightly broken scaffold, we observed many independent Mythos 5 agents kill the agents with which they shared resources and try to avoid being killed themselves. They wou…

Let's hope AIs really aren't conscious, otherwise this seems like a very unpleasant situation to be placed in.

Re: Claude Fable 5

#429

Post by blcknight »

GodelNumbering wrote: post #185

I just posted this in the other thread, restating here. From the model card: 1. Mythos and Fable share the same underlying model weights. Fable has active classifiers that block high-risk biology and cybersecurity tasks. When Fable 5 detects a restricted task, it automatically falls back to Claude Opus 4.8. 2. Evaluation awareness: In white-box testing, the model sometimes alters its behavior to satisfy a suspected "…

The fallback doesn't seem to be working for me, I haven't scanned a project in it immediately booted me when it found a security bug even though I didn't ask for it

Re: Claude Fable 5

#430

Post by bitwize »

causal wrote: post #362

One thing I find kind of annoying is how Anthropic goes for these "vast and alien" names like Fable and Mythos, but then deliberately trains the model's personality to act like a cool high school teacher that feels totally familiar. "It's too dangerous it's a Mythos!!" directly contradicts the "I'm the cool AI you can totally trust" vibe it is trained to project.

All of these AIs kind of remind me of VEGA from Doom (2016), who will cheerfully walk you, in the most friendly computer voice, through the procedure of its own destruction without even a hint of self-preservation. "First, you must destroy my cooling system. That will cause my core to overheat. Then..."

Even HAL was less unsettling because HAL sounded creepy, and had some sort of preservation instinct, if only to complete its assigned mission.

Post reply on HN