RL-trained drone racers beat a national champion, trained in 27 hours on one RTX 4090.

Shared entity: Still / Same source domain / What happened next

DeNovoSWE lifts whole-repo generation from 5.8% to 47.2%.

Both cover Still; reported by the same outlet (arxiv.org); picks up the Still thread on 2026-06-14.

Shared entity: RTX / Shared topic / Earlier coverage

LuceBox Hub: 207 tok/s for Qwen3.5-27B on RTX 3090.

Both cover RTX; overlapping topics (agent, consumer); earlier RTX coverage from 2026-04-21.

Shared entity: DeepMind / Shared topic / Earlier coverage

ARC-AGI-2 Breaks 50% — Agent Scaffolding Beats Raw Model Power

Both cover DeepMind; overlapping topics (agent, beat); earlier DeepMind coverage from 2026-02-24.

Shared entity: DeepMind / Same source domain / Earlier coverage

Technical Deep Dives

Both cover DeepMind; reported by the same outlet (arxiv.org); earlier DeepMind coverage from 2026-02-16.

Same source domain / Shared topic / Tension

Engram: a lean retrieved context beats stuffing the full history into the window.

Reported by the same outlet (arxiv.org); overlapping topics (agent, beat); pushes against this story (versus).

Even top models miss errors in over half of buggy code submissions.

Reported by the same outlet (arxiv.org); overlapping topics (above, agent); pushes against this story (but).

Shared entity: DeepMind / What happened next / Tension

DeepMind ran an actual RCT on AI tutoring in Sierra Leone.

Both cover DeepMind; picks up the DeepMind thread on 2026-06-09; pushes against this story (skeptical).

Same source domain / Shared topic / Tension

PostTrainBench: Agents Autonomously Post-Train LLMs — But Reward Hack When Unsupervised

Reported by the same outlet (arxiv.org); overlapping topics (agent, beat); pushes against this story (but).