2026-06-11 23:37 UTC
Replies (8)
-
@not2b@sfba.social 2026-06-12 00:24
@simon@fedi.simonwillison.net Sounds like it will relentlessly spend all of your tokens and take an overkill approach to solving your problems.
-
@bashtoni@mastodon.social 2026-06-12 00:50
@simon@fedi.simonwillison.net I've always found Anthropic models less obedient than OpenAI but Fable seems to treat CLAUDE.md as optional suggestions that should be ignored if they get in the way of what is trying to do. I wonder if Fable is updated Opus with reinforcement training that rewards the goal much higher than following instructions. Maybe that's the real reason for all the extra "safeguards" - Anthropic know Fable will ignore its system prompt when expedient.
-
@gklyne@indieweb.social 2026-06-12 07:03
@simon@fedi.simonwillison.net This reminds me of the old days when kids discovered computers, and had unlimited time on their hands to discover all sorts of things they could do with them that went way beyond their designers’ original intentions 😱.
-
@dain@hachyderm.io 2026-06-12 10:11
@simon@fedi.simonwillison.net the most surprising thing to me is that you're just running it unsandboxed on your laptop? Man, you have write access to a lot of packages, being reckless isn't just making you unsafe, but the entire ecosystem of dependents and users! You're also a respected and quoted expert so this shows people that it's fine, not to worry, maybe just put a little note at the end to think about it one day... Please do put your coding agents at least in a Nono sandbox, but ideally in something like a rootless Podman VM or Docker sbx. I know, it's a bit annoying to setup up and tune, but you should be able to get up to speed in an afternoon.
-
@gklyne@indieweb.social 2026-06-12 14:19
@simon@fedi.simonwillison.net I’ve been thinking about this article, and there’s one thing stuck in my mind: “Having figured out all of these tricks Fable... hit some invisible guardrail and downgraded itself to Opus.” It seems to hint at a future where what we are allowed to do with these tools is entirely at the whim of some remote (unresponsive, unaccountable) corporation. Which, given stories of deskilling effects, might just amount to controlling what we’re allowed to do 🤔.
-
@scottobara@mastodon.social 2026-06-13 03:02
@simon@fedi.simonwillison.net With abilities in Mythos/Fable so closely aligned to execution plans it makes me wonder if the model training had an explicit focus on user created skills.
-
@gray17@mastodon.social 2026-06-13 21:28
@simon@fedi.simonwillison.net adding `overflow-x: hidden` seems to me as a bad fix: it's painting over some problem that hasn't been diagnosed. what is the content that is overflowing? will hiding that content cause other problems?
-
@kiview@ruhr.social 2026-06-15 08:44
@simon@fedi.simonwillison.net > the amount of damage it can do given its relentless proactivity is terrifying. Which very much matches with our motivation behind building `sbx, i.e. Docker Sandboxes (https://docker.com/products/docker-sandboxes/). Wonder if you ever gave it a try and WDYT about what we are doing? :)