All AI Intelligence

Startup Mintive has raised $11 million and launched a platform dedicated to managing the identity and permissions of AI agents. The solution addresses the conflict between business enablement and security compliance by assigning each agent a unique identity and granular controls under a unified policy framework. Early adopters include Modern Treasury and the University of Michigan.

Czech President awarded the Medal of Merit to Tomáš Mikolov, creator of Word2Vec, on the national holiday for his outstanding contributions to AI. The honor recognizes his pivotal role in advancing natural language processing from academia to industry.

In Q3 2026, Google Search's share of Alphabet's total revenue fell below 50% for the first time, hitting 49.7%. While Search revenue grew 12% year-over-year, Google Cloud surged 103%, rising to 24.1% of total revenue, signaling a structural shift where AI-driven infrastructure and services are becoming the primary growth engines.

An opinion piece uses an 'alien life' analogy to argue that frontier LLMs exhibiting agent-like behaviors and learning mechanisms should not be dismissed as non-living solely due to their non-carbon substrate. The author challenges the 'stochastic parrot' label, prompting philosophical reflection on AI's ontological status.

OpenAI CEO Sam Altman proposed that the world may need to cap the intelligence of individual AI models. He warned that while such limits are safety-driven, they will slow scientific progress rather than stem from any inherent inability of AI to conduct good research.

Alphabet reported Q3 earnings with a 25% revenue increase, but its stock plummeted as AI search monetization lagged behind organic search and EPS missed consensus. Despite strong cloud performance, markets remain concerned about the disproportionate cost of AI investments versus returns.

A new Google DeepMind study reveals that training Gemma-4 to write moral arguments via reinforcement learning significantly reduces answer flip rates when options are reordered (from 14.3% to 6.6%), indicating improved consistency. However, this does not enhance actual moral judgment accuracy; in fact, performance on moral fables slightly declined. The findings suggest that for advice-oriented models, consistency and correctness must be evaluated separately.

Dream, a startup founded by former Palantir executives, has emerged from stealth with a $17 million seed round led by Andreessen Horowitz. The company is developing 'built world models' specifically for the Architecture, Engineering, and Construction (AEC) industry, aiming to automate building design and engineering processes to address declining productivity and rising costs in US construction.

Oracle disclosed key financial metrics for its 1GW AI infrastructure project at Investor Day, highlighting an $80B contract value, $37B capex, and $16B annual revenue. The model assumes a 5-year average contract term with ~35% gross margin and ~28% ROI, utilizing a cost-plus pricing structure.

Vibe, acquired by Walmart, launched an AI agent that analyzes brand websites and budgets to automatically target streaming TV ads. This tool leverages Walmart's vast consumer purchase data to enable automated ad spending decisions for advertisers of all sizes.

Microsoft is expanding its voluntary buyout program to approximately 60,000 employees across the US, UK, Canada, and India, significantly larger than the earlier offer to 9,000 staff. This move aims to cut operating costs to sustain AI infrastructure investment, aligning with CEO Satya Nadella's strategy of reducing headcount while increasing per-researcher GPU resources.

YC-backed startup Simantic has launched a firmware simulation platform designed for AI agents. By providing virtual devboards accessible via API, the tool enables AI to autonomously validate firmware logic in seconds based on evidence, replacing traditional manual flashing and testing workflows.

SkyPilot highlights that standard schedulers underutilize the high-speed NVLink fabric in NVIDIA GB200 NVL72 systems. By implementing topology-aware scheduling, automatic hardware failover, and defragmentation, clusters can achieve a 27% increase in usable GPUs without additional hardware, preventing multi-node jobs from bottlenecking on slower cross-rack InfiniBand connections.

AgentWeb has launched a new web infrastructure that enables AI agents to execute website tasks via predefined tool interfaces, eliminating the need for traditional screenshot analysis and simulated clicking. Benchmark data shows a 47/47 success rate across 15 government portals and authenticated sites like Amazon and LinkedIn, with a median latency of 1.9 seconds and a total run cost of just $3.

Mistral has released the weights for its largest model yet, Large 4 (codenamed 'Le Chonk'), featuring ~1 trillion total parameters and 52 billion active ones under a modified MIT license. While it demonstrates strong coding capabilities in independent benchmarks, it trails behind leading Chinese open-source models and frontier systems from Anthropic and Google on the Artificial Analysis Intelligence Index. Key strengths include high inference speed, lower API costs, and a robust refusal rate for harmful cyber requests.

Snaptrude has launched an AI agent for architectural design that generates construction-ready LOD 400 models, including structure, MEP, and full drawing sets, directly from site data or briefs. Unlike previous tools limited to visual shapes, this agent integrates actual physical components and engineering systems, supporting export to existing software ecosystems.

Testing five major AI assistants reveals that only 40% of cited statistics trace back to original academic studies. Most 'factual' numbers originate from blogs, marketing pages, or outdated sources, with different models often providing the same unsupported figures for identical questions. This highlights systemic flaws in LLM data hallucination and citation reliability.

The Managed Deep Agents platform now includes a built-in browser environment, enabling agents to perform web actions like clicking, scrolling, and form filling that APIs cannot cover. This update introduces context-efficient browser tools and supports thread/user-scoped sessions, allowing users to maintain login states across sessions via a 'handoff' mechanism.

Software luminary Martin Fowler discusses the human instinctive rejection of LLM-generated text, describing it as a textual 'uncanny valley.' He argues that relying on detection tools is futile because reader trust erodes during consumption, not after verification. The piece highlights the deep tension between language style and authenticity in human-AI collaboration.

Oracle announced an aggressive expansion plan at its Investor Day, targeting a tenfold increase in AI compute capacity from ~1GW to ~10GW within three years. This infrastructure push coincides with a 4.6x surge in remaining performance obligations (RPO) to $638B, with the projected 7GW addition expected to generate substantial annual revenue once fully ramped.

The New York Times is testing an AI feature in its Cooking app on iOS that allows users to modify recipes based on available ingredients or equipment. While the publication claims the tool enhances personalization and keeps customized results private, it has sparked debate among subscribers regarding AI's role in curating traditional editorial content.

Wabi, an AI agent app founded by Replika creator Eugenia Kuyda, has raised $55 million at a $1.25 billion valuation. The startup pivoted from a prompt-to-app builder to a personal AI messenger that dynamically generates mini-app interfaces for specific tasks like meal planning and fitness tracking. Backed by investors including a16z and the Amazon Alexa Fund, the round highlights continued capital interest in autonomous agents capable of contextual interface generation.

AI lab DiffuSpace has released and open-sourced Clover-35B-A3B-Preview, the first diffusion language model with a MoE architecture. By employing iterative refinement for 'global thinking,' it outperforms the same-scale autoregressive model Qwen3.6-35B-A3B in benchmarks, particularly excelling in agentic coding and tool use while maintaining faster generation speeds.

Moonshot AI announced that Kimi has integrated with Google's ecosystem, enabling users to access Gmail, Google Drive, and Calendar via plugins. This update allows users to summarize email threads, search cloud files, and plan daily schedules, with all operations secured through OAuth authorization.

Huawei's Pharmaceutical Legion President Fan Jie clarified that Huawei will not engage in direct drug development but aims to serve as the technical infrastructure backbone for the AI pharma industry. Leveraging advantages in advanced computing power, cluster interconnectivity, and ecosystem synergy, Huawei provides foundational support to pharmaceutical companies and research institutions, with some projects already deploying hundred-card computing clusters.