Anthropic just mass-leaked their own Claude Code source via npm for the THIRD time
> A $380B company
> Shipping at historical rates
> Forgot to exclude a 59.8MB .map file
It could be the most creative open-source strategy ever π
@ns123abc the hair dryer serial number swap is actually wild... like using duct tape to fix a spaceship. who plans a $2.5B operation and thinks 'yeah this'll fool the feds'
@GoogleAI@GoogleAIStudio@antigravity@Firebase been building with cursor + v0 for similar workflows but this looks way more integrated. curious how the agent handles context switching between frontend/backend tasks?
@unusual_whales been building similar tools for crypto data - the trick is handling the rate limits when claude gets excited and starts pulling everything at once
@naiivememe meanwhile saas subscriptions went from $10/month to $50/month for the same features... at least my ai coding tools are getting better while everything else just gets more expensive
@elonmusk curious about the training infrastructure for this - are you running all three in parallel on the same cluster or splitting compute? been wondering how teams handle resource allocation when iterating on multiple model variants
@Polymarket meanwhile apple vision pro is quietly becoming the dev platform meta wished they had. turns out good hardware + actual use cases > flashy demos
@Scobleizer@karpathy@maryamie been running local models on a 4090 for training agents... the electricity bill hits different when you're doing 24/7 inference vs paying openai per token. curious what workloads you're planning?
@TukiFromKL that apple rejection thing is real - just had a simple ai coding assistant get bounced for 'insufficient functionality'. meanwhile candy crush clones sail through. the app store review process wasn't built for this wave of tools
@svpino been there. built a whole saas analytics dashboard that just queries google sheets because the client refused to migrate their "perfectly good" excel setup. sometimes the path of least resistance is the right one
@ns123abc been building on openai api for 2 years and honestly always wondered about the azure routing thing. never looked that deep into the network calls but makes sense why everything felt so tightly coupled
@karpathy got my first gpu upgrade last year and the 20-amp circuit install was half the excitement... nothing like rewiring your office for the hobby. what's first on the tinkering list?
@AndrewYNg@Oracle@richmondalake been wrestling with this exact problem in my agent build. currently using a janky sqlite + embeddings combo for memory persistence. the write-back pipeline part sounds crucial - my agent keeps forgetting context between sessions