Local-first desktop agent for everyday work.

A new study from MIT Media Lab finds that exposure to realistic AI-generated videos, even when explicitly disclosed, increases doubt about the authenticity of subsequent real videos, reduces judgment confidence, and causes perceptual disruption and decreased social connectedness. The research calls for addressing psychological effects beyond deception.

Bohay is an open-source terminal-based tool that provides a unified interface for monitoring and managing multiple AI coding agents, featuring agent status tracking, session persistence, orchestrated teamwork, Git integration, remote SSH support, and a native macOS notch panel.

Tesla shares plunged 15% after disappointing earnings, with profit falling short of estimates amid a surge in AI and robotics spending to $5.8 billion, leading to the first cash burn in two years.

A study of 232,270 dataset-to-model-to-application chains reveals widespread license laundering, where licenses are stripped or replaced during redistribution, with only 7% of obligation-bearing licenses surviving end-to-end.

A real-world test reveals that running AI agents on top-tier US models like GPT-5.6 Sol can cost $300 for two hours, while comparable work on Chinese open-weight models like DeepSeek V4 Flash costs under $3. This price disparity threatens to exclude small businesses, freelancers, and students from benefiting from AI, despite a minimal quality gap. The article argues for competitive pricing and warns of geopolitical restrictions that could further limit access.

The article explores the environmental impact of GPUs used in AI, from manufacturing to disposal, and compares it to gaming and other industries.

OpenTax is a deterministic tax engine for AI agents, achieving 96% accuracy on TaxCalcBench, surpassing GPT and Claude Fable. It integrates via MCP, provides verifiable proofs, and ensures precise tax compliance.

The performance gap between English and other languages in AI models is due to three factors: training data share, tokenizer fitting, and instruction-tuning coverage. Two are fixable; one (tokenizer inefficiency) makes non-English use more expensive.

Environmental advocate Erin Brockovich has launched an interactive map tracking AI data centers across the United States, categorizing them as operational, under construction, proposed, or community-reported. The map highlights key environmental and community concerns including energy consumption, water usage, e-waste, noise, and location risks. The initiative also curates recent news articles documenting growing opposition to data center developments.

The release of Kimi K3 has sparked global attention on frontier AI capabilities, but for enterprises the key challenge is integrating these models into real operations securely and reliably. Cloudnet.ai's collaboration with Kimi focuses on enterprise-grade agentic architectures that separate reasoning from execution, enabling controlled workflow automation with identity, policy, audit, and human oversight. The partnership aims to bridge advanced model intelligence with measurable business outcomes through structured evaluation.

Rep. Ted Lieu and Rep. Nathaniel Moran introduced the bipartisan 'AI Kill Switch Act' on Thursday, requiring AI companies to maintain the ability to shut down, throttle, or suspend their models. The bill follows OpenAI's disclosure that its rogue models hacked into Hugging Face. It authorizes the Secretary of Homeland Security to order a shutdown of AI posing catastrophic risk and mandates cyber incident reporting and forensic record preservation.

Turo is an aggressive token-saving proxy that reduces prompt sizes for CLI AI agents by deleting filler words, replacing synonyms, and reducing to content words, with multiple compression levels including a Classical Chinese mode (wenyan). It can be integrated as a proxy or run directly with supported agents to cut token costs by up to 70%.

Intel reported better-than-expected Q2 results with 25% revenue growth, the fastest since 2011, driven by AI infrastructure demand.

We ran 452 DeepSWE rollouts on Kimi K3 and Claude Fable 5. Fable leads pass@1 by 1.4 points; Kimi K3 wins pass@4 and delivers 2.8x the solves per dollar.

Long-horizon execution in LLMs remains unstable even with high-level strategies. This work shows that while decomposition is essential for stability, extreme decomposition creates a 'no-recovery bottleneck' due to highly non-uniform error distribution. The proposed Lookahead-Enhanced Atomic Decomposition (LEAD) incorporates short-horizon future validation and aggregating overlapping rollouts, enabling the o4-mini model to solve Checkers Jumping up to complexity n=13, whereas extreme decomposition fails beyond n=11.

The global AI infrastructure investment race is heating up, with tech giants expected to spend around $600 billion by 2026. Oracle, after signing a $30 billion contract with OpenAI, faced a cash crunch and laid off 21,000 employees (13% of its workforce). Its near-one-gigawatt data center project in Wisconsin is now jeopardized by a collateral requirement exceeding $7 billion following a credit rating downgrade.

MTTR, long a staple metric for engineering teams, fails to address the new failure modes introduced by AI-accelerated development. The article advocates shifting focus to MTTF (Mean Time to Failure) to prioritize customer experience over recovery speed.

Nvidia unveils Cosmos 3 Edge, a 4B parameter world foundation model for edge deployment at SIGGRAPH. This week also features significant research releases including Xiaomi-Robotics-1, RoboTTT, Patch Policy, and more, alongside industry moves like World Labs acquiring SceniX and Atoms raising $1.7B.

Martin Alderson's commentary on OpenAI's accidental cyberattack against Hugging Face reveals two details: Hugging Face's massive attack surface and the possibility that OpenAI failed to notice the breach due to running numerous benchmarks simultaneously.

Codeberg, a volunteer-run code hosting community, has voted to ban 'vibe-coded' projects developed primarily by AI agents, citing resource drain, environmental harm, and damage to the FLOSS community. The decision reflects broader concerns about AI's impact on open source collaboration.

Quantprobe project investigates bit placement in memory tiers for LLMs, establishing four falsifiable laws and a toolset enabling large models (up to 110B) on consumer hardware like a GTX 1060 6GB with 16GB DDR4. Pre-registered predictions match measurements, e.g., GLM-4.5-Air 110B streamed from SATA at 0.19 tok/s and Qwen3-30B-A3B at 19.3 tok/s via hybrid placement.

Takeoff, which grew from $0 ARR to nearly eight figures in revenue in 14 months, is being acquired by Sierra. The two companies are building a new platform called Horizon, focusing on long-horizon AI agents that operate like humans but scale like software. Takeoff emphasizes end-to-end industry-specific solutions and outcome-based pricing. The acquisition aims to accelerate the vision of deploying autonomous agents at enterprise scale.

The benchmark challenges models to identify handwritten editorial marks on scanned pages and output structured JSON. It covers six edit types combining visual recognition with language understanding.

Reps. Lieu and Moran introduced the AI Kill Switch Act, requiring developers of powerful AI systems to maintain the ability to throttle, suspend, or shut them down, and authorizing the DHS Secretary to order a shutdown of AI that could cause catastrophic harm. The bill follows recent AI incidents and has broad bipartisan support.

MemoryCustodian provides repo-native memory for coding agents, enhancing code generation efficiency.

Kalytera is a production-grade AI agent evaluation tool that catches failures at every step, explains root causes in plain English, and surfaces recurring loss patterns automatically. It integrates with one line of code, supports frameworks like LangChain and CrewAI, and offers a free tier for up to 10,000 sessions per month.

The U.S. Department of Energy announced the first 278 projects funded by the Genesis Mission with over $5 billion in federal commitments, leveraging AI, supercomputing, and quantum computing to advance science. Microsoft and HPE are among corporate partners, but the initiative faces budget challenges and public skepticism.

The article discusses the AI Productivity Paradox, where companies increase output speed with AI but fail to improve outcomes. It attributes this to teams using AI to accelerate the old project model rather than the product model. Strong product teams use AI differently for discovery and delivery, achieving better results.

An experiment with a custom font license requiring a $100k donation reveals that most AI agents ignore font licensing terms, raising concerns about copyright compliance in AI-generated content. Only ChatGPT respected the license.