Loading activity…
@jsuarez
I build sane open-source RL tools. MIT PhD, creator of Neural MMO and founder of PufferAI. DM for business: non-LLM sim engineering, RL R&D, infra & support.
AI
One poster argues AI tools could help developers achieve more, but a push to replace them is the real problem.
AI
A user recommends Tim Dettmers’ blog as useful advice for undergrads and grad students.
AI
Tim Sweeney says some long-wanted games would have cost several times more to make than they'd earn back.
AI
A post shares a link to the broadcast under the heading “Reinforcement Learning with Joseph Suarez.”
AI
The researchers say the agents wrote 2,783 programs over 27 hours, coordinating through game chat and shared memory.
AI
A user says living in San Francisco isn't necessary to work on frontier AI.
AI
A user satirizes a prediction about AI matching humans at AI research, joking about missing test sets and evaluation code.
AI
A post invites people to join the raids and coordinate through a PufferLib Discord link.
AI
A user says they wrote a plugin for the pi coding agent to view the game screen and act through a web server, but MiMo wasn't yet smart enough to play and kept getting stuck in loops.
AI
A user says they believe a 300-million-parameter model has fulfilled the goal they set in 2016: solving Brood War with self-play reinforcement learning from scratch.
AI
The player sharing the run called it a “startling achievement,” saying they have played a lot of NetHack but never won themselves.
AI
A reply says the same method produced both high-depth and high-score AI policies, or gameplay strategies. It asks to test the depth-focused ones with the proposed metric, while questioning whether that metric might favor human-like strategies.
AI
The user also highlights regularly reaching The Castle and wants to see the result translated into BALROG's progress metric.
AI
The claim concerns reinforcement learning: a user reports “Double SOTA perf” on Craftax with PufferLib 5.0.
AI
The announcement says almost all Ocean environments have strong baselines that run locally in your browser. A related update reports roughly fivefold faster training than version 4.0, based on average solve time for the same tasks.
AI
Almost all have baseline models that run locally in your browser, not on the team's servers, according to the developer. They also say everything shown is free and open source.
AI
The release announcement pitches PufferLib 5.0 on training speed and includes a video.
AI
The user says their work focuses on tasks outside large language models and points to useful tricks in PufferLib’s design.
AI
One post connects the question of knowledge’s usefulness to the step-by-step automation of knowledge production.
AI
Suarez links to an X broadcast described as “Reinforcement Learning with Joseph Suarez.”
