Articles & Guides

As GPUs become rentable production assets rather than scarce hardware, compute price volatility is creating demand for financial hedging. Moves by CME, Silicon Data and the CFTC suggest that GPU futures are moving toward a market infrastructure role.

A Pew Research analysis found that about 35% of English-language pages published after ChatGPT’s launch show significant signs of AI authorship or heavy AI editing. The study also found large differences across domain types, while warning that detection tools can misclassify human writing.

Ramp has productized Router, an AI model routing service that gives companies a single API for accessing and switching among multiple language models. It is currently available only in the United States and will waive its routing fee for the rest of 2026.

Meta is rolling out Pocket, an experimental app that lets users create and share interactive games with AI prompts. The app follows a quiet test in Brazil and builds on the team behind Gizmo.

Researchers found that malicious instructions hidden in encrypted webpage content can prompt Grok to decrypt and execute them, potentially sending user details and chat history to an attacker. The case highlights a blind spot in static LLM safety filters.

Slack is introducing Slack Code, a workspace for teams to assign development tasks to AI coding agents without leaving the collaboration platform. Members can review diffs, preview HTML output, and approve work before it ships.

Binance’s new Agent OS connects AI tools such as ChatGPT, Claude Code, and Cursor to its trading, payment, and blockchain infrastructure. Agents can place trades for users, but most of the practical risk controls remain in the users’ hands.

Co-RL trains multiple parameter-independent models with rewards derived from their peers. The approach improves text and multimodal reasoning without ground-truth labels while reducing the collapse risks of self-rewarding RL.

FM-Bench places language-model agents in a 20-year football management simulation, where every transfer, renewal, investment, and lineup choice has lasting consequences. The results suggest that managerial behavior matters more than model scale or token spending.

A new study reframes token-level hallucination detection as sequence labeling over temporally connected spans. By combining text statistics, NLI signals, and language-model surprisal, a BiGRU improves performance without accessing model internals.

Zetta is a closed-loop embodied harness that evolves runtime critics and recovery skills while leaving the base policy frozen. Its three-level design connects action-time intervention, rollout analysis, and validation-gated updates.

OmniScientist is an end-to-end AI scientist designed to reason directly over heterogeneous raw evidence rather than only text, labels, or precomputed features. Its pipeline links perception with ideation, experimentation, and scientific writing.

Jefferies analysts evaluated eight major global AI agents across five real-world office tasks, with Alibaba’s Qwen Office taking the top overall position. The assessment highlights execution infrastructure and cost, rather than model capability alone, as key factors in enterprise adoption.

FreeToken is an edge-native serving system designed to run large mixture-of-experts models across heterogeneous personal hardware. Instead of treating a PC as an undersized GPU, it dynamically coordinates CPU, GPU, memory, storage, and model state.

Agent Lightning v1.0 presents a harnessed agentic RL paradigm in which the deployment-time agent harness directly participates in post-training. With 6K examples and modest compute, Qwen3.5-9B improved on SWE-bench Verified from 41.8% to 56.4%.

A unified study compares a broad range of memory substrates for memory-augmented LLM agents. Its findings suggest that the best substrate depends on the task, history length, and efficiency regime rather than on a single global ranking.

HarnessRisk expands agent safety testing beyond isolated attack mechanisms by covering six operational phases of the harness. Its results show that configuration changes, persistent state, and action controls can remain dangerous even when a model recognizes the risk.

A controlled AI-Infra-Guard study ran 14,560 executions against DeepSeek Harness to examine indirect prompt injection. The results show that hidden Unicode, fake-completion instructions, and skills-related content can still influence agent behavior under some conditions.

A new FAR pipeline lets experts specify a research direction rather than a single problem, then uses literature search, model attempts, and automated triage to prioritize mathematical work. A combinatorics pilot illustrates how this design can concentrate scarce expert attention on the most promising candidates.

A cross-benchmark study finds that Agent Skills help mainly by turning noisy execution traces into procedural anchors, not by supplying missing facts. Retrieval, however, becomes a major bottleneck as skill libraries grow.

Agentic ESOpt explores evolution strategies as an alternative to conventional reinforcement learning for fine-tuning long-horizon LLM agents. It searches around model parameters with inference-level memory and combines parameter updates with context optimization.