Sparse Reward Subsystem in Large Language Models

Guowei Xu, Mert Yuksekgonul, and James Zou report a sparse reward subsystem in LLM hidden states: value neurons encode expected value, while dopamine neurons track reward-prediction error. The study finds these signals are robust across tasks and models and useful for confidence estimation and inference-time search.

Research · Research · Machine Learning · LLMs · AI

A Mind Cannot Be Smeared Across TimeConscious experience, temporal windows, and why sequential computation may not suffice.

Michael Timothy Bennett argues that whether machines can be conscious depends not only on what they compute but when: augmenting Stack Theory, he proves existential temporal realisation over windowed trajectories does not preserve conjunction, so a mind cannot simply be smeared across time.

Research · AI · Philosophy · Research · Consciousness

ADHD as a circadian rhythm disorder: evidence and implications for chronotherapy

A Frontiers in Psychiatry perspective arguing that circadian dysfunction — evening chronotype, delayed melatonin onset, and pervasive sleep disturbance — is a clinically significant phenotype in a large ADHD subgroup, with implications for chronotherapy such as timed light and melatonin.

Research · Research · Health · Neuroscience · Sleep

Cache-to-Cache: Direct Semantic Communication Between Large Language Models

Fu et al. propose Cache-to-Cache (C2C): multi-LLM systems exchange KV-cache semantics directly instead of text tokens, aiming for richer inter-model communication with lower latency and token cost.

Research · LLMs · Research · Machine Learning · AI

Laser Fault Injection on a Budget: RP2350 Edition

Courk details a custom, low-cost laser fault injection platform built to beat the Raspberry Pi RP2350 Hacking Challenge, including die imaging, boot ROM analysis, FPGA and laser driver design, and how a single laser-induced fault bypasses the RP2350's Secure Boot.

Research · Hardware Security · Embedded Systems · Electronics

How web bloat impacts users with slow devicesOriginal article link with an AI-written directory summary

Dan Luu measures how modern websites behave on low-end devices. The analysis highlights CPU costs and responsiveness problems that faster networks alone cannot solve, connecting technical choices with who can use the web.

Research · Web Performance · Accessibility