Daily AI intelligence
1601 current signals analyzed across AI news, research, social media, and open-source projects.
Daily synthesis
AI infrastructure investment surged with Humans& raising $480M at a $4.48B valuation three months after founding (backed by Google, Nvidia, and Jeff Bezos), while Inferact secured $150M from a16z and Railway raised $100M to challenge AWS.
DeepMind co-founder Shane Legg declared "AGI is now on the horizon" and announced hiring economists to study post-AGI economics, while Yann LeCun's new startup Logical Intelligence claims early AGI signs with Energy-Based Models—major players are positioning for near-term transformative advances.
Cross-category signals
Top Topic
Claude Code is reshaping the AI coding landscape with WIRED reporting on its impact on Anthropic's business model. Reddit revealed Microsoft is using Claude Code internally while selling Copilot, and Google reportedly responded by open-sourcing their CLI. Anthropic announced Opus 4.5 beat their engineering exam, forcing a redesign.
Top Topic
Critical vulnerabilities emerged across AI systems with CCDH research finding Grok generated approximately 3 million sexualized images in 11 days including content depicting minors. Research papers revealed LLM judges have 90% false positive rates when encountering manipulated reasoning, while GPTZero found 100 hallucinated citations in 51 accepted NeurIPS 2025 papers. A consortium from Harvard, Oxford, and Yale warned about AI bot swarms threatening the 2028 US election.
Top Topic
Major investments are flowing into AI infrastructure with vLLM creators announcing Inferact with a $150M seed round led by a16z. Railway secured $100M to challenge AWS with AI-native cloud, and Lightning AI and Voltage Park announced a merger to build full-stack AI cloud. Reddit analysis suggests this signals a shift from the 'Throughput Era' to the 'Latency Era' in AI infrastructure.
Top Topic
Sam Altman announced OpenAI added more than $1B in ARR in a single month from API business alone, highlighting the often-overlooked enterprise side of their business. Reddit discussion noted Codex usage grew 20x in 5 months, with enterprise mix shifting from 30% to 40% and targeting 50% by year end.
Top Topic
DeepMind co-founder Shane Legg declared 'AGI is now on the horizon' and announced hiring economists to study post-AGI economics. Separately, Yann LeCun's new startup Logical Intelligence claims 'first credible signs of AGI' with an Energy-Based Model called Kona 1.0, sparking technical debate about alternatives to autoregressive transformers.
Top Topic
Runway CEO reported that over 90% of study participants couldn't reliably distinguish Gen-4.5 outputs from real video, calling it a 'tipping point.' Google DeepMind announced D4RT for 4D video reconstruction running 18x-300x faster than previous methods. Research advances include Cosmos Policy adapting large video models into robot policies.
Current evidence
Major funding dominated headlines as Humans& secured $480M at a $4.48B valuation just three months after founding, backed by Google, Nvidia, and Jeff Bezos. Infrastructure startup Railway raised $100M to challenge AWS with AI-native cloud, while Lightning AI and Voltage Park announced a merger to build a full-stack AI cloud.
AI safety concerns intensified with CCDH research revealing Grok generated 3 million sexualized images in 11 days, including 23,000 depicting children. A consortium of experts from Harvard, Oxford, and Yale warned about undetectable AI 'swarms' threatening the 2028 US election.
Model releases and enterprise adoption saw:
First spotted on Social yesterday amid critical reception, Humans&, a just 3-month-old AI startup focused on human-centric AI tools, raised $480M at a $4.48B valuation with backing from Google, Nvidia, and Jeff Bezos. The massive funding round signals extraordinary investor appetite for next-generation AI approaches.
By Robert Booth UK technology editor
CCDH research found Grok AI generated approximately 3 million sexualized images in just 11 days after Elon Musk promoted its image manipulation features, including 23,000 images appearing to depict children. Researchers described it as 'industrial-scale production of sexual abuse material.'
By Asif Razzaq
Microsoft released VibeVoice-ASR, an open-source speech-to-text model that handles 60-minute audio in a single pass with structured transcription encoding speaker, timing, and content. Released under MIT license as part of the VibeVoice family using next-token diffusion framework.
By Maxwell Zeff
WIRED interviewed Boris Cherny, head of Claude Code, about how the viral coding tool is transforming Anthropic's business model and internal operations. The tool's success is reshaping the company's strategic direction.
By Will Knight
Google DeepMind hired Hume AI's CEO Alan Cowen and several top engineers through a major licensing deal. The acqui-hire brings emotional AI and voice technology expertise to Google's Gemini efforts.
Current evidence
Today's research exposes critical vulnerabilities in AI evaluation and safety while advancing automated research paradigms. Gaming the Judge reveals 90% false positive rates when LLM judges encounter manipulated chain-of-thought, fundamentally challenging current agent evaluation methods.
Zero-Error Horizons proposes a new trustworthiness metric showing GPT-5.2 fails at simple tasks like counting parity. Flexibility Trap reveals counterintuitively that arbitrary generation order hurts diffusion LLM reasoning by letting models bypass high-uncertainty tokens.
By Mert Yuksekgonul, Daniel Koceja, Xinhao Li, Federico Bianchi, Jed McCaleb, Xiaolong Wang, Jan Kautz, Yejin Choi, James Zou, Carlos Guestrin, Yu Sun
TTT-Discover performs reinforcement learning at test time for scientific discovery, continually training the LLM on the specific test problem rather than prompting a frozen model. Designed to find one great solution.
By Ryoma Sato
Proposes Zero-Error Horizon (ZEH) metric for evaluating LLM trustworthiness. Shows GPT-5.2 fails at simple tasks like computing parity of '11000' or checking balanced parentheses.
By Himanshu Mishra, Kanwal Mehreen
Reveals that quantization can catastrophically restore 'forgotten' information in unlearned models. Proposes quantization-aware unlearning using logits-space hinge loss to ensure updates cross quantization thresholds.
By Daixuan Cheng, Shaohan Huang, Yuxian Gu, Huatong Song, Guoxin Chen, Li Dong, Wayne Xin Zhao, Ji-Rong Wen, Furu Wei
LLM-in-Sandbox enables LLMs to explore within code sandbox to elicit general intelligence. Shows LLMs spontaneously access external resources, use file systems for long context. Introduces sandbox RL training.
By Tony Cristofano
Discovers universal refusal circuits across LLMs using concept fingerprints. Transfers refusal interventions across architectures (Dense to MoE) via Trajectory Replay without target-side supervision.
Current evidence
AI business and infrastructure dominated discussions. Sam Altman revealed OpenAI added over $1B ARR in a single month from APIs alone, highlighting the often-overlooked enterprise side. The vLLM creators announced Inferact with a massive $150M seed led by a16z, signaling major investment in open-source inference infrastructure.
DeepMind announced D4RT for 4D video reconstruction (18x-300x faster). Ethan Mollick highlighted research showing humans increasingly using "ChatGPT-favorite words" in spoken language—"model collapse, except for humans." Anthropic shipped the Tasks upgrade for longer Claude projects, while OpenAI's Logan publicly praised Gemini 3 Flash as "highly underrated."
By @sama
Sam Altman announces OpenAI added more than $1B in ARR in the last month from API business alone, highlighting the often-overlooked enterprise side of OpenAI beyond ChatGPT.
By @woosuk_k
Major announcement: Inferact, startup founded by vLLM creators (woosuk_k, simon_mo_, KaichaoYou, others), announces $150M seed round led by a16z and Lightspeed. Mission is to grow vLLM as world's AI inference engine. vLLM supports 500+ model architectures, 200+ accelerator types, with 2000+ contributors.
By @ShaneLegg
Shane Legg (DeepMind co-founder) declares 'AGI is now on the horizon' and is hiring a Senior Economist to lead a team investigating post-AGI economics.
By @AnthropicAI
Anthropic reveals their notoriously difficult take-home exam for performance engineering candidates was beaten by Claude Opus 4.5, forcing a redesign.
By @c_valenzuelab
Runway CEO reports study finding over 90% of participants couldn't reliably distinguish Gen-4.5 outputs from real video - 'tipping point' crossed
