@zephyr_z9 On the topic of Moonshot affording compute, there's a risk… I think the ultra-sparse approach is failing to handle logical reasoning and problem solving like frontier models. Made a new benchmark digging deeper:
@alexjc
-
DeepSeek V4 Open Models Lag Behind Frontier AI
By
–
@teortaxesTex About DeepSeek V4 being able to compete with the frontier, I made a new benchmark that suggests open models (particularly the new ultra-sparse ones) are qualitatively worse at problem solving and logical reasoning:
-
Open Weights Model Performance: MoE Sparsity Challenges
By
–
I will add the remaining two (?) models rumoured for release early next week and finalize it with a blog post… In the meantime, if you have any theories why open weights struggle (my theory is that it's MoE/sparsity induced) — let me know!
-
Open Weights Models Lag Behind Frontier on Logical Reasoning
By
–
PREVIEW: The Joy Of Benchmarks (Q1'26) My new #AI benchmark on out-of-domain programming languages (joy) suggests that open weights models are qualitatively *far* behind the frontier on logical reasoning and problem solving… The newest models: GLM-5, Minimax M2.5, and Kimi
-
Does Runtime Definition Imply Walktime Existence?
By
–
Does the definition of a "runtime" imply the existence of a "walktime" too?
-
AI-Generated Code Quality and Iterative Design Improvement
By
–
Somewhere in between hand-written source and machine slop… there is a new kind of code. It starts out kinda functional but its design feels very far below average, takes multiple complete rewrites and many more review rounds (pushing back against regression to the mean), and
-
Joyfl v0.6: Runtime Language Server with JSON Protocol
By
–
joyfl — v0.6: Runtime Language Server Lots of features small and big, most important is a simple language server + protocol (JSON over stdin/stdout) for easier integration into modern tools. Also migrated over to Codeberg, for all the reasons: https://
codeberg.org/creativeai/joy
fl
… -
Tauri App Compatibility Issues on Older Intel Mac Hardware
By
–
@conductor_build Trying the app on an older Intel mac, is it possible the current build doesn't run there? Not exactly up-to date but Tauri supports it… Thanks!
-
Style-conditioned AI models with prompt engineering capabilities
By
–
Oh yes. I liked the original Ai2 project for that reason, it's contained to a single codebase! So it can learn your style… I bet it'd be relatively easy to make it style-conditioned model as well, like a promptable feel/patterns/quality!
-
Kimi 2.5 Video-to-Website: Multimodal AI Breakthrough in Design
By
–
People saying there is no obvious use case for video-to- website: it's a hint how the model was (post-)trained. And it worked well for Kimi 2.5, top in the design arena by far! https://t.co/Tg4189Y6ef pic.twitter.com/gUqtDBk6P0
— Alex J. Champandard 🌱 (@alexjc) 28 janvier 2026People saying there is no obvious use case for video-to- website: it's a hint how the model was (post-)trained. And it worked well for Kimi 2.5, top in the design arena by far!