INDEPENDENT SIGNAL // RESEARCH & CONTEXT
Research through a different lens. Source-grounded reporting on science and the systems shaping our world.
HOW THIS WORKS ↗PUBLICATION INDEX
CHANNEL // 01 [106]
Research into artificial intelligence and intelligent systems.
Source published
The HybridSpec preprint reports faster language-model decoding by adapting how many proposed tokens a model checks, using early expert-routing information and more efficient recurrent-state kernels. Its strongest results concern single-request generation on three Qwen models; gains were smaller in a four-client experiment.
Source published
The preprint “Forgetting to Improve” proposes removing observations that distort a Gaussian process model’s uncertainty in a task-relevant region. The authors report benefits in regression, optimization and simulated control, but the method’s guarantees are conditional and its improvements are not universal.
Source published
A preprint testing three language models finds that community support for cooperation can encourage free-riding unless prompts explicitly ask models to weigh social norms and self-consistency. That framing also produced smoother responses across changing community beliefs, but did not reliably align stated beliefs with chosen actions.
CHANNEL // 02 [45]
Research into plants, their biology, and their environments.
Source published
A bioRxiv preprint groups tomato-leaf metabolite signals by their genetic associations, then looks for shared chemical clues to prioritize genes and reactions. Overexpression experiments support three candidate genes, while incomplete compound identification and statistical limits constrain the interpretation.
Source published
A review and analysis of 107 experiments examines how chilling and changing light damage photosystem I. It argues that far-red light can alter which protective pathway matters most, while emphasizing that the proposed mechanism and its importance in nature still need testing.
Source published
Experiments identify CrJAM2 and CrJAM3 as repressors of alkaloid-pathway gene expression in Catharanthus roseus. But hormone treatment reversed several effects after 24 hours, and the study did not establish a reliable increase in alkaloid production.
CHANNEL // 03 [111]
Research into economic systems, decisions, and policy.
Source published
A preprint by Kaushil Patel proposes that whole-dollar attention thresholds can explain both sharp demand drops and weak responses to sales taxes. Its reanalysis of an existing experiment supports some predictions but exposes others the model cannot explain, while showing why similar purchasing behavior can imply different welfare losses.
Source published
Irene Aldridge’s theoretical preprint separates the value of responding to information from the value of average action choices. It also identifies when logged decisions can support an audit—and when missing counterfactual observations make that impossible.
Source published
A preprint presents an algorithm for approximate equilibria in two-player Colonel Blotto games with arbitrary positive rational budgets and shared battlefield values. It also gives a linear-time exact construction under a specific balance condition, using an idealized model of computation.