We just released an oss version of PorTAL, including training scripts and links to the paper’s models on Hugging Face: github.com/ramp-public/porta… Looking for alpha testers and feedback, give it a spin and let us know what you think!
Average bjork beat on homogenic
Ancient Roman music is the most unsettling shit I’ve ever heard in my life
Readers added context they thought people might want to know
The music is not from a surviving score. It is new music based on guesswork.
en.wikipedia.org/wiki/Synaulia
1
6
483
many are saying this
Memory is what allows Claude to learn from experience. In a new Engineering Blog post, we share how “dreaming” helps Claude organize what it’s learned, correct its mistakes, and find what it needs more easily the next time. anthropic.com/research/compo…
1
66
31,028
🦍 Introducing Memorilla: memory as a modality for LLMs
Memorilla turns everything a frozen LLM needs to remember into just 16 memory embeddings
⚡ It uses up to 4,212× fewer tokens and is 27× faster than full context inference while more accurate
🎯 Pushes the Pareto frontier over RAG and agentic memory in personalization, long-document and knowledge QA
A thread 🧵:
Humans form short-term memories throughout the day and consolidate them into long-term memory while we sleep. Dreaming plays a huge role in this process, replaying details from the day in different contexts and scenarios.
I think LLMs should have a similar process for long-term memory consolidation. In this setting, dreaming is self-study. We create a self play game that identifies gaps in existing knowledge and can generate targeted experiences that strengthen memory in those areas. Please read more about our work on this below
Reframing cartridge training as a self-play game cuts the training steps needed by more than half, just by asking better questions.
This post continues our series on cartridges, compressed KV caches of long documents. Here we focus on self-study, the method used to train them. Self-study teaches a model reading the cartridge to answer the way it would with the full document in context. Currently, its questions come from fixed, generic prompts, so many of them ask about things the cartridge already knows.
Recent KV cache compression research has focused on small models with dense attention. We scaled these techniques to GLM 5.3, a frontier sparse attention model.
We adapted KVzip and Attention Matching, reducing KV cache size by 80% while retaining over 90% of full-context accuracy on the QuALITY benchmark.
RL for continuous learning is slow and can lead to catastrophic forgetting; OPSD does not work reliably. At Ramp Labs we’ve been taking an alternative bet to continuous learning, one that involves efficient accumulated memory in the KV cache over changing parametric knowledge.
This blog is the start of a multi part series on techniques my coworker Jeff and I have been exploring over the last few months to bring these techniques to the frontier. Watch this space 👀
Readers added context they thought people might want to know
The video is an animation by artist R.J. (@RJ16848519) from his Screen Time series, shared here without credit.
x.com/rektober/statu…
x.com/connoissurreal…
x.com/kikanicolela/s…
there are many many people who studied physics and came to the conclusion that working on machine learning would probably create more scientific progress than studying physics including many of the original chatgpt crew like john schulman, liam fedus, amodei & jared kaplan at ant















