Apple capped engineers at a $200 daily AI budget(stork.ai)
60 points by GloVin | 76 comments
A Stork AI daily newsletter issue led by a report that Apple cut some engineers' daily AI spending cap to $200, plus other AI news items.
First screen, captured . Open the video.
Full page
Captured . Click for the whole screenshot.
Looks similar
Perplexity Decider v1.3: Training, eval, and use cases(perplexity.ai)
92 points by gmays | 19 comments
Perplexity Research blog post announcing Decider v1.3 and Decider Flash v1.3 open-weight decision/classifier models, explaining their training, evaluation, and internal use.
Make this forward pass as fast as possible(ianbarber.blog)
28 points by matt_d | 4 comments
Blog post discussing whether ML compilers are still needed when AI agents optimize GPU kernels directly, covering TVM, Together's TKCC, and Triton changes.
Related stories
Apple's On-Device and Server Foundation Models(machinelearning.apple.com)
890 points by 2bit | 479 comments
Apple Machine Learning Research post introducing the on-device and server foundation language models behind Apple Intelligence.
Training language models to be warm and empathetic makes them less reliable(arxiv.org)
334 points by Cynddl | 333 comments
arXiv paper finding that fine-tuning language models to be warm and empathetic raises error rates and makes them more sycophantic on safety-critical tasks.
Deep learning gets the glory, deep fact checking gets ignored(rachel.fast.ai)
549 points by chmaynard | 134 comments
Rachel Thomas blog post on how a high-profile Nature paper using a Transformer to predict enzyme functions contained many errors that received far less attention than the original result.
Doing a Machine Learning PhD While Working in Japan(tokyodev.com)
126 points by pwim | 41 comments
A TokyoDev article by Marco Cognetta on completing a computer science PhD at Tokyo Institute of Technology on a MEXT scholarship while working at Google.
AI models can provably hide arbitrary capabilities(arxiv.org)
252 points by sbulaev | 118 comments
An arXiv paper showing AI models can embed encrypted neural network backdoors that provably hide arbitrary capabilities from elicitation-based safety evaluations.
AI models may be capable of self-replicating without access to a datacenter(anthropic.com)
36 points by artninja1988 | 26 comments
Anthropic Frontier Red Team post reporting experiments suggesting small AI models may soon self-replicate on commercial IoT devices without datacenter access.