Doing magic at Moon Studios!
Tech entrepreneur, wholeheartedly into Engineering & Design, loving father of 2, playing with new tech, always optimizing things
first impressions on Opus 5.
- its forgetful and often times does not follow directions
- it claims things with confidence only to retreat when I prove it wrong OR when it re-reads things and realizes it made stuff up earlier
- its not thinking things through properly and often backtracks
- its creating great UI designs (HTML/CSS) so far
If you are used to Fable you spot the shortcomings quickly because in the end its a 2T model versus Fable 5T. The benchmark scores they presented don't add up for my tests unfortunately - yet.
casually vibe-coded a Mirror’s Edge clone over the rainy weekend. mostly one big prompt plus some follow-up tuning with Fable.
characters, rigging, and animations are definitely still the weak spots, but this stuff is getting ridiculously fun. lots to improve... and a few other game prototypes coming soon 👀
@bcherny this is a super annoying bug. Inserting long texts into Claude Code often times just truncates the text. Here, I provided a longer Voice input and it only accepted the last sentence. Sometimes it shows the [Pasted Text #1] thing correctly, sometimes this here.
In preparation of the Fable re-release you wanna update your prompts to always include a hint, per request, which model is actually working on your task. Or tell it to mention when a request is routed to another model.
@Prathkum I built stuff with fable for several hours and it was from another planet. Anthropic's next model, which is in training now, will leap even further. If Codex 5.6 does not deliver then OpenAI has officially lost the race. And I have a hunch it won't be great.
@EmmanuelMacron@bundeskanzler
The US government shutting down Anthropic models for international customers, is your chance to incentivise Anthropic to relocate to Europe and thereby position Europe in the AI game. @AnthropicAI