I used Fable nonstop while it was out and am now back to Opus, and I don't notice a difference
If you told me that there was a bug with my Claude Code for those three days, and I was on Opus the whole time, I wouldn't be surprised
I am skeptical about all the claims of how much better people find it. Not because I don't think it's better. I trust Anthropic evals. But because I think our guts are poorly calibrated to sense differences in intelligence at this level
My guess is that it's a lot like blind taste testing of wine: it's orders of magnitude harder than you'd think it is. It's easy to fool yourself that you can tell the difference
Which I guess we can turn into a challenge: when Fable comes back online, I can make a "blind taste test" app to give people a chance to see if they can tell which is which. I'd be very impressed with those that can! I'd love to learn your ways!
Notes
- @annarmitchell @cognition Best ad ever
- @DavidKPiano @simonw what do you call something that most of the time gives the same answer? feels like maybe we want a name for that but yeah, i agree with you, people mean "it has deterministic parts" or "it has some non-deterministic parts embedded in a deterministic workflow"
- @DavidKPiano lol! glad to know i'm not the only one who cares about definitions it's you me and @simonw
- Everyone is using the phrase "one shot" incorrectly, including me! One-shot means that you give the LLM one example before asking it to do the task We mean zero-shot: you give the model a description of what you want done, and no examples
- https://www.val.town/
- new homepage!! https://x.com/stevekrouse/status/2067692500811231532/photo/1
- @bradnoble @MaineFrameworks huge hug!
- i love teaching client-server-database architecture to non-technical folks. such a hug unlock. it's a powerful mental model https://x.com/stevekrouse/status/2067680038141485190/photo/1
- @bradnoble @ValDotTown on the roadmap! thanks for the vote :)
- @Ian_Costa18 lol
- I am always and forever insufferably pointing people to things I've already written down 😅
- if i can't remember the difference between these two buttons, i can't image how confusing they are for non-tech folks https://x.com/stevekrouse/status/2066923766857212078/photo/1
- @ericciarla @firecrawl to be fair, claude is who caught it https://x.com/stevekrouse/status/2066921086273400852/photo/1
- @firecrawl very cool! ps - that's invalid js; drop the trailing comma ;)
- @jesusnotw @ValDotTown sorry about this! we had trouble with our authentication provider for a short time. resolved now
- RT @fw3d: what i was missing in my ai stack: @ValDotTown it literally gives wings to everything i build
Reposted by Steve
- RT @CharlieMolthrop: New @ValDotTown @Vapi_AI template! https://x.com/CharlieMolthrop/status/2065488664927379511/photo/1
Reposted by Steve
- https://luma.com/49gi6yz4
- if you or someone you know in nyc wants to vibe code: come to the val town office next thursday! https://x.com/stevekrouse/status/2065544997236068698/photo/1
- wake up babe, new level of abstraction just dropped https://x.com/rauchg/status/2065520041894756480
- RT @nicholasbs: I've deleted more code with AI than I've written with it. I help maintain a 15-year-old Rails codebase for the @RecurseCen…
Reposted by Steve
Text may be partial · Read original
- It's annoying when AI agents need you to press buttons in an app, because the MCP doesn't cover something What if we added every possible route from our app's internal unstable frontend API to our MCP server? Then we could guarantee that your agent would never need you to press a button again! MCP = Unstable Frontend API Both MCP and Unstable Frontend API are: 1. Machine readable: json, not ui 2. Quickly changeable: it's ok to ship breaking changes, because there's inference (human or agent) on both sides of the API call – unlike with deterministic code calling a stable REST API – to handle anything unexpected
- Human understanding is the bottleneck What an exciting time for people who care about making things legible! Still the best example is https://glench.github.io/fuzzyset.js/ui/ https://x.com/garybernhardt/status/2065330891963219981
- @garybernhardt https://glench.github.io/fuzzyset.js/ui/
- after hours of nonstop fable, i finally triggered it! https://x.com/stevekrouse/status/2065192296929231206/photo/1
- RT @ValDotTown: sure, we had blobs, but not scoped blobs, it's very important to scope blobs to vals because of agents and vibecoding, the…
Reposted by Steve
Text may be partial · Read original
- our goal is +20% monthly arr - got some work to do! https://x.com/stevekrouse/status/2065078356094181632/photo/1
- @runarorama Cool!
- https://x.com/stevekrouse/status/2064780947141079378/photo/1
- @danshipper embarrassing answer for me
- @danshipper how???
- http://val.town/x/templates/stripe-mirror
- ARR from stripe is surprisingly hard Things become a lot simpler when you mirror your stripe data into a database Introducing Val Town stripe-mirror get yo stripe data into a sqlite!!! Then saaspocalypse the shit of your revenue analytics https://x.com/peer_rich/status/1701291062927503499 https://x.com/stevekrouse/status/2064467877768450299/photo/1
- https://simonwillison.net/2026/May/28/claude-opus-4-8/
- https://claude.ai/share/c4e47af3-10e0-49ef-8e85-396d9b52ce94
- vs Opus 4.8 https://x.com/stevekrouse/status/2064462712554311804/photo/1
- fable kinda bombed pelican bench??? @simonw - what do you think? https://x.com/stevekrouse/status/2064462710088065293/photo/1
- obligatory fable launch tweet https://val.town/townie https://x.com/stevekrouse/status/2064451576735957339/photo/1
- @firecrawl @WorkOS I asked Codex to try this out and it replied: I can fetch http://firecrawl.dev/auth.md, discover the metadata, and call /agent/auth, but this Codex session does not expose any tool to mint the required WorkOS ID-JAG. The missing piece is an OpenAI/WorkOS-signed identity assertion…
Text may be partial · Read original












