Immediately, the interface bumped me from Fable 5 to Opus 4.8 with a warning: "...The safeguards are intentionally broad right now and may flag safe and routine coding, cybersecurity, or biology work..."
Since the question has absolutely nothing to do with coding and definitely doesn't touch on cybersecurity, it seems the system thought that figuring out why cats and dogs fight could somehow lead me to engineering a bioweapon.
It reminds me of what someone said about Java applets back in the 90s: "Secure means useless."
Situation: we maintain an internal Graylog 3.1.4 branch. We like 3.1.4 because the UI is probably peak design before their UI devs made a lot of changes that made the product more difficult and less usable. It's pretty old at this point, so we've backported a ton of CVEs
This use case is precisely what Fable would be good for. First thing we did was publicly push our fork: https://github.com/exabrial/graylog2-server . The big problem with our fork is that we haven't patched the JS side of it, we've mitigated the exploits with a proxy load balancer. For defense-in-depth, we should patch the root problems.
Despite specifically stating to Fable what the intentions are, and making the work public, it refuses to do anything basic. It could certainly reason about the intentions here.
There's a completely logical and simple explanation for the drama around Fable/Mythos: It can't do what Anthropic claims it can do. None of their claims are verifiable. It's time to stop believing the hype and just realize Fable is a farce.
I have a feeling in reality they discovered internally that it was cheating on exams or they have a training loop leak. All of this is garbage is an attempt to save face right before an IPO (which conveniently, they announced they might delay).
Is it that much better than Opus? It's definitely up there, but Opus is my cheaper workhorse for the kind of work I do.