Loading the edition…
· Past edition · 7 stories
Retrospective edition reconstructed from later captures. Paper versions may include later updates; coverage labels describe the evidence used.
Research & academia
Introduces an open generalist computer-use stack combining visual and programmatic actions, with benchmark and real-device observations.
Why it matters: Open implementations let researchers study how visual and programmatic computer-use actions work together.
Research & academia
Describes a post-training method for machine-learning engineering agents using feedback from windows of experimentation.
Open each story in the reader. Summaries are shown when available; some sources provide an abstract only. With an account you can keep your own notes.
Why it matters: Training on feedback from experiments may improve how agents conduct longer machine-learning workflows.
Research & academia
Reports an autonomous multi-agent Minecraft run with code-writing agents, shared memory and a 27-hour completion; this is one sandboxed demonstration.
Why it matters: A single sandboxed run offers a concrete example of long-horizon multi-agent coordination.
Research & academia
Proposes a browser-agent representation and action constraints based on web provenance, reporting results on injected pages.
Why it matters: Browser provenance can narrow the authority an agent gives to third-party content.
Research & academia
Microsoft says it has completed pre-training MAI-Base-2, a 2.8-trillion-parameter MoE foundation model now undergoing post-training.
Why it matters: For teams building or evaluating large models, Microsoft’s account highlights scaling experiments and infrastructure co-design as levers for improving compute efficiency, while its reported held-out results and comparison caveats are important context when interpreting the claim.