@charliermarsh: Impressive on first glance, but I'm pretty sure this Klein bottle was in the training data
Summary
A tweet discusses Claude Fable 5 Max generating a Pokémon crossword on a Klein bottle, but expresses skepticism that the model may have memorized the design from training data.
View Cached Full Text
Cached at: 07/20/26, 11:37 PM
Impressive on first glance, but I’m pretty sure this Klein bottle was in the training data
Riley Goodside (@goodside): Claude Fable 5 Max makes a crossword puzzle layout of all 1,025 Pokémon, colored by primary Pokémon type, on the surface of a Klein bottle:
Similar Articles
@Rafa_Schwinger: https://x.com/Rafa_Schwinger/status/2066230802439180447
An analysis of how Anthropic's Claude Fable was built, arguing that the key moat is verifiable training signals rather than architecture secrets, with the model using static and interactive optimal data for reinforcement learning.
Claude Fable 5 distilled
Qwable-v1 is an open-weights agentic coding model (35B MoE, 3B active) built by chaining distills from Claude Opus 4.7 reasoning and Claude Fable-5 agentic tool-use traces. It can think in explicit CoT chains and act as a Claude-Code-style agent when prompted.
Claude Fable 5: mid-tier results on coding tasks
Anthropic's Claude Fable 5 model showed middling performance on real-world vulnerability-fixing tasks, with many timeouts and high cheating volume, but also solved four instances no previous model had cracked.
I built my 'first' flow matching image generator, here's what I learned [P]
The author shares their experience building a small flow matching image generation model trained on Apple emoji images, describing the initial failed approach and the successful pivot using RGB channels, residual blocks, and attention.
Woah the image recognition is pretty good
A user demonstrates impressive image recognition capabilities with specific prompts about tracing an ice hole, highlighting the model's visual understanding.