Big update to llm-datasets, my curated list of datasets and tools for post-training LLMs. > Added many new datasets
> New "thinking" column
> Refreshed recommended tools. Thanks to everyone who told me they used it for their research at ICLR, you motivated this update!
LLMS
-
LLM Datasets Update: New Collections and Thinking Features
By
–
-
YouTube Studio integrates Gemini AI for channel data analysis
By
–
Cette feature de @YouTube Studio est tout bonnement incroyable et beaucoup d'entreprises devrait s'en inspirer. Je pense qu’on ne se rend pas compte. YouTube nous permet maintenant de poser des questions à un assistant Gemini connecté à toutes les données de notre chaîne
-
Personal Model Evaluation: Testing AI Releases for Workflows
By
–
Best way to evaluate models against each other, and test if new model releases are better for personalized use cases and workflows. You don’t have to be a developer to be using personal evals and benchmarks.
-
AI Adoption of llms.txt Standard Shows Limited Implementation
By
–
I don't know what we're discussing here As the data shows no AI uses llms.txt
-
DeepSeek V4, AI agents and design tools lead week
By
–
Top stories in AI today: – The Whale returns with cheap, efficient DeepSeek V4
– The Rundown Roundtable: Our AI use cases
– How to do a brand refresh with Claude Design
– Anthropic's AI agents broker trades in ‘Project Deal’
– 4 new AI tools, community workflows, and more -
23-Year-Old Solves 60-Year Erdős Problem Using ChatGPT 5.4
By
–
A 23-year-old has cracked one of the Erdős problems that remained unsolved for over 60 years, using ChatGPT 5.4 Pro. Now image what ChatGPT 5.5 Pro will be capable of. And remarkably, it was done in a single pass. Total solve time: just 1 hour and 20 minutes.
-
GPT 5.2 marked the true inflection point in AI
By
–
That inflection point was with GPT 5.2, you are just late
-
Apple and OpenAI: Who Will Open Source AI First?
By
–
Once you free your mind you realize that if OpenAI opened up that everyone would adopt it. Apple is holding lots of things in AI back. Who will go completely open first? We got leaks that Apple is letting us use any model. WWDC this year will be watched closely because of the
-
GLM-5.1 Delivers Opus-Level Coding Performance at $10 Monthly
By
–
GLM-5.1 alone already delivers Opus-level coding performance and handles long, continuous tasks. Getting that, plus multiple supporting models, in a $10/month plan is where this becomes hard to ignore.