[bottlepalm]

I find it funny that AI keeps getting bigger, and the mental gymnastics needed to trivalize the progress get bigger as well - ie the government shutdown an AI model twisted into now even the government is being tricked.

Everyone is tricked except me. Only I know AI isn't as smart as everyone thinks it is.

[throwaway74628]

“Too dangerous to release” has been exploited for marketing.

A sizeable plurality of the informed public know as much.

Regulatory capture is a thing.

[lazide]

Or you could use it, and see the massive disconnect between hype and reality yourself. It’s not hard.

The market is built on hype, so of course it’s going to get hyped everywhere.

[bottlepalm]

I've seen Fable reverse engineer binaries like nothing I've used before - Fable/Mythos is far from marketing hype.

On top of that I think it's just stupid to think anyone in the marketing department at Anthropic has any part in the system card for a model. That kind of thinking just screams cope.

[mikojan]

It is beyond absurd to assume a company dependent on unprecedented sums of investor money is NOT deeply integrating its marketing department in its operations.

[christoph]

I’ll dream of a world where even 1% of that marketing money goes to customer support.

[ikiris]

The ai psychosis is real.

We've played with it a good bit, it in no way matches the ridiculous hype.

[IndeanCondor]

This statement needs qualifiers.

Are you claiming you have a raw binary to Fable and it just reverse engineered it by reading it? Or are you claiming (like for every other model released in the past 1.5 years) it's using an integration with Ghidra or BinaryNinja to assist - in which case I completely disagree even a 30B model can do that with those tools.

Also an FYI, AI advancement and Anthropic are not synonymous. Someone asking Anthropic to back up their claims is not coping about AI, especially as independent benchmarking of Fable is giving equivalent or slightly above par results to GPT 5.5.

The system card does not use any of the benchmarks used in the previous Opus 4.5+ system cards. All the scores are in Anthropic owned benchmarks. I find it extremely hard to believe the marketing department of the company was not involved in a material release to the public - which is the marketing departments literal job.

[bottlepalm]

Yes with assist tools Fable was able to figure things out Opus 4.8 and ChatGPT 5.5 were unable to. Like significantly better.

[SepiaSapient]

I'm sorry that I think that "Our LLM is the missing element for a group to develop nukes or bioweapons" is marketing hogwash.

I'll guess we will see when or if the IPO happens. The more probable claim (Trump just wants money) will be proved if Amodei buys Truth Social or something and pulls a Tim Apple. My (not very probable) tinfoil hat theory is sadly unverifiable, but very funny. Anthropic bribed some Trump minion to ban Fable and lock in the honeymoon period until just before the IPO.

[reassess_blind]

Not as smart as everyone thinks it is, maybe, but a model like Fable 5 without safeguards against offensive cyber attacks would be a nightmare. There are millions of improperly secured web applications that, in the wrong hands, would be easily exploited by these models.

[lillesvin]

There have been millions of trivially exploitable vulnerabilities out there for decades — many of which could be easily discovered by using simple scanning tools or manual probing. This is hardly a new situation and LLMs really aren't that impressive at pentesting — even with these simple exploits. Maybe they are if you're not a pentester, but then ZAP, Burp, Nessus, SQLMap, etc. are likely also impressive if you put a little effort into learning how to use them, but many AI-advocates aren't interested in learning skills themselves.

It's the same situation as with vibe coding. Everyone and their grandma can have an LLM spit out a web application without any programming experience, but if you're a programmer, you'll likely quickly see some issues with maintainability and further development of the code base.

[zomiaen]

>LLMs really aren't that impressive at pentesting

The point is that Mythos apparently is quite capable and has developed novel exploits on its own.

[lillesvin]

That's the claim, yes. Has any proof been made available yet? (Genuinely asking here because I haven't been paying that close attention.)

[tayo42]

In a substantially different way then how it is now? You can put something listening on 22, 80 and 443 and log how much stuff tries to get in.

[reassess_blind]

Yes, it is substantially different. A targeted, relentless attack by a state of the art cybersecurity model is far more likely to find obscure vulnerabilities than a traditional automated attack/fuzzer. These models are so much better at finding security holes than anything we've seen before.

[ianm218]

I feel like it is strange seeing some really smart people go full conspiracy theory tin foil hat. Half these threads think that Anthropic is playing some 5D chess game to purposefully get nationalized.