The Archive

ShallowStream optimizes streaming video MLLMs by pruning computation at shallow layers, reducing overhead for embodied AI and autonomous driving.

The Trump administration has intervened in The New York Times' copyright lawsuit against OpenAI, making an argument in favor of the AI lab. The landmark lawsuit, filed in December 2023, alleging that OpenAI unlawfully trained its AI systems on articles from The New York Times and seeks to recoup "billions of dollars" in damages from both Microsoft and OpenAI. This week, the Trump administration filed a statement of interest in the case, supporting OpenAI's argument that it's fair use to train an AI model on copyrighted text. "The New York Times seeks to narrow fair-use doctrine to exclude the...

CodePoisonRAG demonstrates black-box knowledge poisoning attacks on retrieval-augmented code generation systems, revealing supply-chain vulnerabilities.

HyperStyler transfers authorship style with few examples using hypernetworks; narrow NLP application, limited frontier relevance.

Training data attribution via influence-guided rewriting shows intervention effects exceed reweighting; incremental TDA methodology contribution.

Tabular foundation models TabPFN-3, TabICLv2, TabDPT show strong out-of-domain physics priors but lack true noiseless limit understanding.

This post is the third in a series on AI model co-design. It explores how to accelerate LLM inference while maintaining accuracy using speculative decoding and... This post is the third in a series on AI model co-design. It explores how to accelerate LLM inference while maintaining accuracy using speculative decoding and offers five guidelines for selecting draft length and draft mechanism across the Pareto frontier. For a discussion of how model design choices impact both throughput and interactivity without sacrificing accuracy, see AI Model Co… Source

Wonderful's $550M Series C was led by Insight Partners, with Salesforce, Index Ventures, IVP, Vine Ventures, 9Yards and Bessemer participating.

Jio is betting it can turn an aging computer into an AI-ready PC for as little as about $11 for two months.

Factory edge agents selected via retrieval-augmented answer quality outperform parameter-count heuristics for on-premise deployment.

MedMisBench study reveals how LLMs fail under misleading medical context via fabricated evidence; implications for clinical AI safety.

Bilevel coordination game model formalizes orchestrator-worker LLM systems; game-theoretic analysis of reflection and memory.

Repo-To-Skill distills GitHub repositories into compact verified skills for autonomous ML research agents; operational knowledge layer.

HiPoly hierarchical framework for polymer property prediction via G2RINS representation; domain-specific materials science application.

Pooled LLM evaluation reduces cost of comparative retrieval model selection by incrementally expanding judgment pools across new candidates.

MrBeast will feature Gemini, Google Health, and the Fitbit Air in upcoming videos as part of a multi-year partnership with Google. The deal will kick off with a video featuring Jimmy "MrBeast" Donaldson turning to Gemini for wilderness survival advice: First up on September 5 is a new MrBeast video following Jimmy and his crew as they attempt to survive some of the most extreme and treacherous climates on Earth. Teams will race to navigate three of the planet's most unforgiving landscapes - the jungle, the desert, and the Arctic - using Gemini to help them identify dangers, survive brutal wea...

Detection method for Vehicle-to-Infrastructure attacks on Signal Phase and Timing messages in connected vehicles.

Language models control attention dynamically without O(N) proxy scoring, enabling efficient long-context processing.

Comparison of seven LoRA variants for dysarthric ASR adaptation on Whisper and Qwen3 baselines.

LoRA-TSD optimizer treats LoRA updates as tangent vectors on fixed-rank manifolds, achieving 2.8× speedup.

Training-free RVSD framework reduces visual hallucinations in vision-language models via retrieval and sparse decoding.

CORAL agent-driven harness optimizes production recommender retrieval, ranking, and serving decisions at scale.

Google launches Fairwind Program, limited-access cyber defense tools for governments and enterprise partners.

Theoretical analysis of Polyak and Nesterov momentum in large-batch training via kernel regression.

Universal approximation theorem for neural operators approximating convex monotone semigroups via Chernoff methods.

Door-in-the-face psychological technique increases LLM refusal compliance; works on Anthropic Opus 5 (65.8%) but not OpenAI frontier models.

Trace as State reformulates long-context Transformer reasoning by caching reasoning traces as conditional state, reducing memory exponentially in worst case.

HiddenLayer has raised a $100M Series B from Delta-v Capital, Ten Eleven Ventures, Morgan Stanley, Microsoft's M12, Booz Allen Hamilton, and others.

Amazon is adding a scam-detection feature to Alexa for Shopping that can verify whether suspicious emails, texts, and other messages actually came from the retailer.

DKL combines finetuning with retrieval-augmented generation to inject corpus-specific knowledge into instruction-tuned LLMs without requiring massive synthetic data.