The Archive

HazardWeaver: LLM agent framework for scientific route selection in multi-step hazard analysis workflows.

Threat-preserving representation sensitivity metric reveals benchmark representation effects on LLM agent security ASR scores.

Solver-aware target selection for caching in few-step diffusion transformers optimizes tensor reuse post-distillation.

HyperBrowseComp: multilingual multimodal benchmark of 423 questions across 13 languages for web-browsing agent evaluation.

mmWave radar signal processing for contactless heart-rate sensing via learned observability assessment.

Multi-candidate LLM pipeline using GPT-4o-mini and Claude Sonnet 4 for text simplification and complexity spotting.

Domain-shift robustness framework for authorship detection and generative AI detection using feature stability taxonomy.

FinSkillBench: 2,603-episode financial agent evaluation across portfolio, risk, and analysis tasks with deterministic verifiers.

Cephalonauts One: 30-hour whole-brain fMRI dataset with naturalistic podcast audio for speech decoding research.

GRIP/TRIP framework for principled long-range interaction certification in Graph Neural Network benchmarks.

LLM mathematical reasoning via search exhibits expert performance but fails to generalize abstraction levels across subfields.

Recursive Harness Self-Improvement (RSI): framework for adaptive co-evolution of reasoning-data synthesis tasks and generation procedures.

ZeroMAG: zero-shot multimodal adapter for extending EEG foundation models to heterogeneous physiological signals without labels.

Soft Actor-Critic autonomous robot navigation for endovascular brain-computer interface catheter delivery in cerebral veins.

Comparative study of LLM prompting vs. style embeddings for zero-shot authorship attribution using author representations.

Theoretical analysis proving forward KL divergence inflates student entropy in LLM distillation, with empirical validation across pretraining and finetuning.

OpenAI publishes guidance for GPT-6 model selection, reasoning tuning, prompting, tool integration, and production deployment for startups.

Proposes compilation-based evaluation for GNN explainers to replace problematic plausibility benchmarks that conflate motif recovery with actual model reliance.

FLINT: structured composition framework for table-to-report generation using verifiable atomic insights to improve coherence and traceability over sequential agents.

FLINT: domain-specialized Text-to-SQL system for financial data using lookup agents and multi-hop joins, exceeding 50% on production benchmarks vs. generic systems.

CERTID: formal identification pipeline revealing reasoning models produce unsound causal identifiability judgments; uses complete calculus to validate claims.

XGenAct: world action model for robot manipulation integrating RGB, depth, surface normals, and functional role prediction via cross-task generation.

Root-cause-guided on-policy distillation corrects student reasoning failures instead of borrowing reference solutions, addressing distillation trap in chain-of-thought training.

Weave Forcing: training-free framework for compositional memory routing in long-form interactive video generation, selectively reusing historical visual components.

Empirical study showing efficient CoT training via token reduction does not necessarily harm reasoning faithfulness or human monitorability across methods.

Enterprise AI is no longer a future ambition. It is in full operational flight. Model capabilities are advancing faster than most organizations can absorb, while the cost of performance continues to fall. Globally, AI investment is set to reach $2.5 trillion in 2026, up 44% from the previous year. For many enterprises, this investment has…

"There is an ontological difference, even before an aesthetic one, between art and what a machine can generate through statistical calculation based on millions of images created by others," the pope wrote. "Algorithms lack the spark of humanity."

Blackstone's Jas Khaira will take the Builders Stage at TechCrunch Disrupt 2026 on building next-gen AI. Register for your pass and get 50% off a second.

Here are Google’s latest AI updates from September 2026