"GradMem: Learning to Write Context into Memory with Test-Time Gradient Descent" Instead of giving LLMs huge KV caches or one-shot compressed summaries, GradMem shows that you can let a frozen model take some test-time gradient steps to write a long context into a small memory.
MACHINE LEARNING
-
5 Causal AI Tools That Reveal Why Not Just What
By
–
5 Causal AI Tools That Reveal Why — Not Just What: https://
medium.com/@Modexa/causal
-ai-5-tools-that-reveal-why-not-just-what-437ed6be6ec8
… -
GPT-5.4 Pro Outperforms Other Models for Complex Tasks
By
–
GPT-5.4 Pro continues to be the only model of its class. For anything really hard & complex, I throw it into the maw with every bit of context I can think of. More often than not, something very useful comes out. I can't get the same results from Codex or Code or anything else.
→ View original post on X — @mattshumer_, 2026-03-23 17:43 UTC
-
Plan and Budget Framework Optimizes LLM Token Efficiency
By
–
Is your LLM wasting valuable tokens "overthinking" or "underthinking" complex tasks? MIT CSAIL, Virginia Tech, University of Virginia, and Michigan State University present Plan and Budget. This new framework helps LLMs decompose complex problems into manageable sub-questions
-
Uni-1 understands the real and eliminates the need for perfect prompts
By
–
Uni-1 → understands how something real should look This changes something key: You don't need to perfect prompts.
-
Pretraining LLMs entirely on Hugging Face Hub infrastructure
By
–
You can now pretrain LLMs entirely on the HF Hub 💥 Last week, @OpenAI launched a competition to see who can pretrain the best LLM in under 10 minutes. So over the weekend, I made a little demo to automate this end-to-end using the Hub as the infra layer: – Jobs to scale compute – Buckets to store all experiments – Trackio to log all the metrics The cool thing here is that everything is launched locally: no ssh shenanigans into a cluster or fighting with colleagues over storage and GPUs ⚔️ All that's left is coming up with new ideas, but luckily Codex can automate that part too 😁 Can I have a job now please @reach_vb 🙏?
→ View original post on X — @thom_wolf, 2026-03-23 16:29 UTC
-
AI Solves First Open Problem from FrontierMath Benchmark
By
–
EPIC! It's confirmed that the AI has solved the first mathematical problem from the FrontierMath: Open Problems benchmark, which consists of problems that remained unsolved despite attempts by the mathematical community. The first of many more to come!
-
Agentic Workflows Transform Software Development Automation
By
–
Agentic workflows for software development https://
buff.ly/SiF4PdH
#AI #MachineLearning #DeepLearning #LLMs #DataScience -
Elsevier Hiring Software Engineers and AI Specialists Globally
By
–
Elsevier is hiring software engineers, data scientists, prompt engineers, and tech leaders to join its growing tech hubs in Mexico, India, the Netherlands, and beyond. Work on AI-driven products, global data platforms, and scalable solutions used by millions
-
Uni-1: Revolutionary Model Combining Thinking and Pixel Generation
By
–
Uni-1 is here! A new kind of model that thinks and generates pixels simultaneously. Less artificial. More intelligent. pic.twitter.com/2p8kSq4Jtf
— Luma (@LumaLabsAI) 23 mars 2026Uni-1 is here! A new kind of model that thinks and generates pixels simultaneously. Less artificial. More intelligent.