The Archive

The president posted on Truth Social that he wanted to appoint an "AI czar" to lead a new "AI force." He made the announcement amid growing calls from across the political spectrum and even within the industry to pump the brakes on AI development. He posted that his administration "will not in any way hinder or stifle the Growth of this incredible Industry. Rather, we will cherish it, help it, and watch over it, as it grows!" In the rambling post, Trump said that data centers were "wealthy and prestigious" and claimed without evidence that they raised salaries, lowered taxes, and made for saf...

Current ticket pricing ends in 6 days on Sept. 25 at 11:59 p.m. PT. Join 10,000+ founders, investors and tech leaders at Disrupt and save up to $200 on your ticket until then.

Before recent high-profile hacks raised the specter of AI possibly "killing all humans," our energy systems were already disturbingly vulnerable to cyberattack - and the risk is growing. "We were always prey. We were just kind of surviving at the appetite of our predators," Joshua Corman, executive in residence for public safety and resilience at the Institute for Security and Technology (IST), told me last year. At the time, I was preoccupied with a Department of Homeland Security warning that Iranian actors and sympathizers could target the US with cyberattacks. Last week, I called Corman u...

Method for discovering hidden thermodynamic ontologies from anonymous experimental measurements using symbolic regression and identifiability theory.

Fine-tuning Google Research's TimesFM foundation model for real-time feature extraction in particle physics detector data acquisition systems.

Flow matching approach for Bayesian filtering in high-dimensional physical systems using test-time training to preserve full posterior distributions.

Identifiability theory and polynomial-time algorithm for recovering physical ontology and measurement types from controlled experiments without labeled ground truth.

Training-free leaky-integrator method that stabilizes error accumulation in recursive differenced time-series forecasting without architectural changes.

Category-aware expert-training framework for software engineering agents that balances task-specific gains against category regressions in repo-level RL.

Orthogonal knowledge distillation method for open-vocabulary audio-visual event localization that selectively weights teacher supervision by temporal-boundary reliability.

Evidence ladder framework organizing healthcare RL progress from retrospective policy identification through prospective evaluation and lifecycle monitoring.

Machine-Interpretable Information (MII) protocol for compiling documents into fixed-bandwidth agent-readable states, solving index-payload schism in retrieval-augmented systems.

Verification-driven fine-tuning approach enabling LLMs to generate syntactically correct FORM code for symbolic particle physics calculations with zero-shot failure.

Formalizes information preservation in recurrent models via predictive quotient theory, balancing stability against memory destruction.

TicTacBench evaluates coding agents on RTL timing closure, a critical gap in existing hardware design benchmarks.

Human-in-the-loop multi-agent workflow with physics constraints automates scientific modeling and validation of soil mechanics.

Low-rank factorization reduces pitch estimation model size 36% on edge devices without accuracy loss.

Stochastic reconfiguration acts as statistical filter in overparameterized neural quantum states, improving finite-sample generalization.

World model predicts digital health campaign conversion curves in-flight using latent patient dynamics.

Systematic benchmark of long-tail loss functions on single-cell foundation models reveals architecture-agnostic failures on rare cell types.

Graph-theoretic analysis of LoRA adapter co-occurrence in Alibaba production diffusion inference workloads using 26 million inferences.

CSC detector addresses LLM-era social bot camouflage via calibrated multi-modal conflict awareness between text and graph signals.

Comparative benchmark of graph database engines on LLM agent workloads isolates query planning and ingest costs across vendors.

ValueDiff proposes value-geometric KV cache eviction for LLMs with weak attention sinks, ranking tokens by value-vector L2 deviation from cache mean.

OMST optimizes strata construction for A/B test stratified sampling using multi-way decision trees to improve statistical power.

Comparative evaluation of EmbeddingGemma and fine-tuned variants for semantic job-candidate matching via hybrid retrieval with reciprocal rank fusion.

Formal analysis of A* search tie-breaking strategies on final f-layer, showing unbounded expansion-count gaps between standard strategies.

Count Flow Map generates high-dimensional count data via learned stochastic transitions using Poisson births and Binomial deaths in one-step generation.

Latent Telepathy enables decentralized multi-robot teams to communicate learned perceptual latents rather than kinematic data under partial observability.

RA-MMEA framework assesses visual reliability and adaptively improves image representations for robust multi-modal entity alignment.