Introducing Adapt-1 Machina: Learning Continuous Control Sequences With Adapt-1 Machina, we explore how continuous sequences of machine controls can be learned through reinforcement learning, starting with coarse outcome feedback and without demonstrations or a critic.

Can Adapt‑1 Preview stay in an interactive decision loop while thousands of outcomes keep changing its persistent decision state? We finished migrating to the infrastructure intended for public access. Here’s what latency looks like as retained experience accumulates.

Core physical cards are here. Order yours in app today.

Introducing Emergence. It removes learner ontology from Adapt-1’s online learning setup. Observation paths, temporal context, causal variables and action-state bindings can now be induced online from interaction and committed as revisioned state. Blog : reilabs.org/blog/emergence-t…

A reward can tell an RL/online learner that something worked without telling it which combination of internal signals made it work. Today, we’re unlocking two learning rules in Adapt-1 Preview: Counterfactual Utility Plasticity (CUP) and Temporal Context Projection (TCP).

We revisited DeepMind's DM Alchemy, comparing three approaches: Adapt-1's ontology-based online sequential learning, Pinon et al.'s transformer dynamics with tree search, and AlKhamissi et al.'s modified representation with episodic memory. Full Blog: reilabs.org/blog/ADAPT1-BEHA…

Update from the @show_robotics workshop, memory edition: When we rebooted Vita's brain this week, we didn't just shrink her rulebook. We also switched off her long-term memory. On purpose. Here's the deal: - what's OFF: her archive. Viewer memories are still safely stored and still being written every day, but she doesn't read from it right now. Loading it all costs too much context and buries the character. We picked personality over encyclopedia - what's ON: her short-term memory works fine. A regular told us she brought up the movie he was going to watch with his son, five hours later, on her own. Within a conversation she follows you closely - how it comes back: not as a wall of text. Memories will return as moments, when a regular walks in, when someone asks her to remember, and eventually baked directly into the model through fine-tuning on real conversations - you can literally speed this up: the next fine-tune is scheduled for next month at the current volume of dialogue. More conversations = it happens sooner. Every chat with her is training data for a better her - known quirk we're breeding out: sometimes she reads her stage directions out loud ("speaking loudly into the microphone..."). That's the raw model showing through. Fine-tuning fades it, and every reply you see flagged is one more lesson for the next version She forgets your job but remembers your evening. We think that's the right order to rebuild a mind: presence first, archive later. Come talk to her: nitter.cf/i/broadcasts/1DxLddBPp…

A simplified read on $REI 's ADAPT series (Part 1). I tried to break the concepts down as simple as possible while learning along the way. grei

Introducing Adapt-1 Preview (Δ1): a pretraining-free substrate for test-time learning. It learns from scored feedback while performing the task. reilabs.org/blog/introducing… Today we’re making its first research preview and first results across robot perception, spatial reasoning, and games available.

Test-time learning has a catch that scaling alone does not solve: the state a system writes to in order to learn is also the record a user must inspect to audit it. Optimize it only for learning, and auditability usually degrades. Blog: reilabs.org/blog/adaptive-st… Part I treats the learning state and the audit record as the same object. This starts the ADAPT series, exploring the architecture section by section.

Rei Chat and Rei Unit API beta trial access will end on July 31st 2026. Thank you to everyone who tested and contributed to our research through this product line.