
Current todo: continuous control, better exploration, more stable latent space. 29/35 isn't 35/35, which is the goal. 4/4

Advancing general-purpose intelligence.

Current todo: continuous control, better exploration, more stable latent space. 29/35 isn't 35/35, which is the goal. 4/4

35 environments (control, mazes, arcade, planning, multi-agent games), all partially observable, one config for all. Ent is reliably good on 29/35 and never fails outright. Every classic RL algorithm fails multiple times and is good on half at max. 3/4

Current RL algorithms usually don’t work out of the box and require heavy adaptation for each task. We aim to change that with Ent - an agent that learns a model of its environment from interactions and uses that model to plan its actions. 2/4

At Regular Lab we're building Ent, a general-purpose reinforcement learning agent meant to learn any environment from scratch, with no per-task tuning. First look at Ent v0.1. regularlab.ai/blog/work-in-p… 1/4

If this resonates, consider joining us: regularlab.ai/join

We believe it’s time to get back to the drawing board. Modern AI looks powerful — until it doesn’t. Systems that excel in one setting still fail in others. Instead of pushing further, we’re stepping back to build systems that actually learn, plan, and generalize. regularlab.ai/blog/back-to-t…

Hallo, we are Regular Lab - a Europe-based AI R&D lab focused on exploring new ideas and approaches to machine intelligence. You can learn more about us and how we think on regularlab.ai