True if you round-trip through OCR to markdown, that's where the token cost piles up. But PixelRAG skips that. The screenshot gets embedded straight into a vector for retrieval, no image to text step in the pipeline. The VLM only reads pixels at the end, on the few tiles it
RESEARCH
-
Small random recurrent net contains function, train only output layer
By
–
A small randomly connected recurrent net tends to bubble around chaotically and often contain the function you are looking for, and you can find it by training only an output layer. That's the basic idea
-
Two-month pinned threads and recall of important elements
By
–
Yeah! I had pinned threads that continued for more than two months and codex still remembers all the important elements. Otherwise, it searches for important elements by scanning the session itself.
-
GPT-5.6: a great positive surprise ahead
By
–
I have the feeling that GPT-5.6 will be a great positive surprise. Let us recall the information about GPT-5.6: "The company is separately preparing to launch a new AI model, codenamed 5.6, which will represent a 'significant improvement' over the flagship"
-
Google’s X, Brain, and Cerebra before DeepMind integration
By
–
google had x, brain and cerebra before it all got deepminded
-
Andrej Karpathy’s 2018 talk on Building the Software 2.0 Stack
By
–
The talk is "Building the Software 2 0 Stack (Andrej Karpathy)" 2018
-
Karpathy: PhD sleep lost over models, industry over datasets
By
–
There is this 2018 talk of Andrej Karpathy comparing his PhD time spent vs Tesla old days. PhD: sleep lost over massaging models and algorithms
Tesla(industry): sleep lost over massaging datasets In reality it is always about data and evals/your mechanism of evaluating the -
Offline agentic exploration develops robot skills
By
–
Great work using offline agentic exploration to develop robot skills! https://t.co/9JOJDpLohI
— Ken Goldberg (@Ken_Goldberg) 20 juin 2026Great work using offline agentic exploration to develop robot skills!
-
Betting on GLM 6 with new pre-training optimizations
By
–
I am betting big time on GLM 6 There are many recent papers with great pre-training optimizations (e.g. DSv4) Now, if Zhipu uses some of that (+ their own novel research), and top it off with their current post-training regime, we're looking at an amazing SOTA in the making