Mentioning partly for an example of mundane utility, partly as an example of Things People Probably Don’t Guess A Computer Can Do, and partly because there are still some people who think that LLMs don’t have a world model.
LLMS
-
Llama 1B Inference Optimized in Single CUDA Kernel
By
–
So so so cool. Llama 1B batch one inference in one single CUDA kernel, deleting synchronization boundaries imposed by breaking the computation into a series of kernels called in sequence. The *optimal* orchestration of compute and memory is only achievable in this way.
-
Learning Reasoning Without External Rewards in AI Systems
By
–
Learning to Reason without External Rewards
Paper: https://
arxiv.org/pdf/2505.19590
.pdf
…
Code: https://
github.com/sunblaze-ucb/I
ntuitor
… -
Fine-tuning LLM with RL becomes surprisingly easy to implement
By
–
today i finetuned an LLM with RL for the first time. i regret to inform you that it was easy. it only took a few hours to configure. even though this is a custom task and dataset. and it worked, quite well, on the first run
-
Google’s Open-Source App for Local Android Models
By
–
It turns out that Google also has an open-source mobile app for running local models on Android. Works with Gemma models and supports importing models from other sources as well.
-
Prompt-Based AI Generation Works Well Across Styles
By
–
Just prompt it, it should be pretty good at all sorts of styles.
-
Claude’s Thought Tokens Now Free in Usage Metadata
By
–
Current model pricing doesn’t change, the thought tokens in “usage metadata” refer to the full thoughts (which is what you pay for) and the summaries are just free (and experimental).
-
Model reasoning with summarization layer for detailed outputs
By
–
Behind the scenes, the model is still reasoning with full thoughts, and then we have a summarization model which translates from full thoughts to summary, while preserving as much detail as possible.
-
Gemini API introduces thought summaries feature
By
–
We just rolled out “thought summaries” in the Gemini API, now you can see what the model is thinking and make use of that info! A thread with the details and request for feedback
-
Getting Started with Qwen3: Alibaba’s New AI Breakthrough
By
–
🚀Getting Started with Qwen3 Alibaba Cloud just dropped Qwen3, the latest in the Qwen series, and it’s setting new standards! 🌟
— Satya Mallick (@LearnOpenCV) 27 mai 2025
Outperforming top-tier LLMs like DeepSeek-R1, o1, and Gemini-2.5-Pro, Qwen3 is not just a language model. It’s a thinking machine! 🤖
With its… pic.twitter.com/3pwTLuv4ZqGetting Started with Qwen3 Alibaba Cloud just dropped Qwen3, the latest in the Qwen series, and it’s setting new standards! Outperforming top-tier LLMs like DeepSeek-R1, o1, and Gemini-2.5-Pro, Qwen3 is not just a language model. It’s a thinking machine! With its