Research & Deep Dives
Research papers, reports, scenarios, and technical findings translated into practical implications for builders and operators.
Open Research & Deep DivesA post-training paper shows how interpretability tools can audit preference data, expose unwanted learning signals, and reshape rewards before models absorb bad habits. The method turns interpretability into a practical intervention in data curation and reward design.
Self-Harness shows that agents can use execution traces, targeted edits, and regression tests to improve the scaffolding around their own model. The important constraint is preserving task performance while the runtime changes itself.
Paul David's dynamo analogy explains why general-purpose technologies can spread through business before appearing clearly in productivity statistics. The lesson is to distinguish implementation lag from evidence that the technology lacks value.
A paper separates the model that improves an agent's harness from the model that must use it. Cheap models may propose useful scaffold changes, but task performance still depends on an executor capable enough to load and follow them.
A scaling paper argues that larger models win partly because frequent tasks stop overwriting rare-task features. The proposed mechanism links parameter scale to reduced interference and better retention of infrequent capabilities over time.
Leopold Aschenbrenner's Situational Awareness argues that AGI by 2027, superintelligence, trillion-dollar clusters, lab security, and government mobilization follow from the same trendlines. Reading it as a unified acceleration thesis makes its assumptions and policy implications easier to test.
An essay on Pluralis argues that decentralized AI needs scalable training, defensible monetization, and governance for intelligence outside the major labs. The piece tests whether distributed infrastructure can create a credible alternative to centralized model ownership.
Life-Harness argues that many agent failures come from the runtime interface, not the model, and shows frozen models improving across tasks when the harness changes. It shifts attention toward interfaces, tools, feedback, and execution design.