Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
The first of its kind, this architecture allows data contributors to activate or deactivate their contributions in real-time without retraining the entire model, enabling instant modification. The system trains separate expert modules on private datasets using a frozen public model as an anchor point, then merges them via domain-informed routing without requiring joint training.
Before companies can get to a future where real work happens through AI agents that interact with both data and people, companies have to solve the action interface for human operators. The action interface matters because most operators aren't prompt engineers - their job is to keep the world moving. Operators need a clean interface that lets them react to and confirm actions and a way to proactively ask the system to take actions. This article discusses a new paradigm for building interfaces that creates a seamless experience for operators.
Google has added speech generation to Flow, its AI video tool, allowing users to transform images into talking clips using Veo 3. Alongside this, Flow and Google AI Ultra are now available in over 140 countries, expanding access to advanced video and AI features worldwide.
Sponsored
You'll probably like our flagship newsletter. It's all about tech, science, and programming.
gemini-beta-3.0-pro was just referenced in the latest Gemini-CLI commit.
Elon Musk confirmed that xAI's chatbot Grok will be integrated into Tesla vehicles "next week at the latest."
Sponsored
Visual AI Platform-as-a-Service includes a comprehensive development framework, including ControlNets and IP adaptors.
Enterprise-grade visual AI platform-as-a-service with full IP indemnity. Advanced AI Generation and Editing, Product Shot Suite, Fine-Tuning, Ad Generation, and more.
Benchmark tests on Grok 4 (in early access) show that it will be the lead AI frontier, beating out OpenAI's o3, Google's Gemini 2.5 Pro, Anthropic's Claude 4 Opus, and DeepSeek R1. This will be the first time xAI has the lead AI frontier. The reasoning model is priced at $3 per million input tokens and $15 per million output tokens, which is identical to Claude 4 Sonnet but more expensive than Gemini 2.5 Pro. This post shows benchmarking results for the upcoming model.
Perplexity has launched Comet, a browser built to serve as an AI-powered assistant across both work and personal tasks. It integrates Perplexity's search and reasoning engine to help users organize knowledge, ask questions, and streamline digital workflows.
OpenAI plans to launch an AI-powered web browser soon, challenging Google's Chrome. The new browser will integrate AI agents for tasks like booking reservations, potentially disrupting Google's ad-revenue model dependent on Chrome's user data. OpenAI's browser, built on Google's open-source Chromium, reflects its strategy to directly capture user web behavior data.
Miqdad Jaffer, a product leader at OpenAI, outlines a framework for building AI tools that solve real problems. The 4D method (Discover, Design, Develop, and Deploy) emphasizes user-centric problem discovery, invisible trust-building AI design, resilient development practices, and high-impact first-use experiences.
A hands-on tutorial for experimenting with Gemma 3n and MatFormer, Google's nested transformer model that supports elastic inference via the Mix-n-Match technique.
Sponsored
Dell's Pro Max GB10 designed for AI development [14]lets you train and run massive models locally. The GB10, Built on NVIDIA Grace Blackwell, delivers 1 Petaflop for 200B parameter models, while the Dell Pro Max GB300 powers up to 20 Petaflops and 460B parameters with
Biomni is a general-purpose biomedical AI agent that combines LLM reasoning with retrieval-augmented planning and code execution to autonomously complete research tasks across biomedical domains.
The open-source server lets AI agents query databases and automatically handles connection pooling, authentication, and security for database interactions.
Adding irrelevant phrases like "Interesting fact: cats sleep most of their lives" to math problems causes reasoning models to generate incorrect answers 300% more often than baseline. The query-agnostic phenomenon transfers across model size, with distilled models showing even higher vulnerability rates. The attacks also cause significant computational overhead, with 42% of responses exceeding 1.5x normal token length.
Google's Circle to Search feature now includes AI Mode, allowing users to get advanced reasoning and follow-up responses directly from their visual queries without leaving the current app.
GenAI is set to significantly influence Amazon's Prime Day sales, with Adobe predicting a 3,200% increase in AI-driven traffic. This four-day event expects $23.8 billion in online sales, a 28.4% year-over-year growth. GenAI tools enhance shopping experiences, aiding in research, product recommendations, and deal finding.
Sponsored
You'll probably like our flagship newsletter. It's all about tech, science, and programming.
Students can now access lecture transcripts from Panopto and peer-reviewed Wiley content directly within Claude conversations, while new Canvas LTI support enables seamless integration into coursework.
Nvidia passed Apple's previous record of $3.9 trillion before closing at $3.97 trillion.
OpenAI has finalized its $6.5 billion acquisition of Jony Ive's hardware startup, io Products Inc.
YouTube framed the change as a minor update of existing rules, but the timing reflects growing concerns over AI-generated slop flooding the platform, damaging their credibility and advertising appeal.
Sponsored
How does OpenAI manage billing for millions of users across thousands of enterprise contracts?
Replit and Microsoft have partnered to deliver natural-language-based enterprise app development, integrating with Azure services to let business users create and deploy production-ready software without coding experience.
Mistral is negotiating to raise $1 billion from investors like Abu Dhabi's MGX fund and French lenders such as Bpifrance SACA. The French AI startup, known for its Le Chat chatbot and open-weight LLMs, has already raised $1.19 billion, reaching a $6.51 billion valuation. Mistral is collaborating with MGX and Nvidia to build Europe's largest AI data center, aligned with the UAE's €50 billion AI investment in France.
A Harvard Law graduate who argued before the Supreme Court fed his case briefs into Claude 4 Opus and had it answer the same questions the Justices posed to him. The AI delivered what he called an "outstanding oral argument" with coherent answers and clever responses he hadn't considered, leading him to conclude that AI lawyers could soon outperform even top human advocates at oral argument.
Sponsored
Join a team making transformative AI breakthroughs with unmatched resources and scale. From our AGI SF Lab to our Amazon Nova foundation models, we're combining pioneering research with real-world impact, backed by Amazon's industry-leading infrastructure and commitment to long-term innovation.
OLMo 2 is a family of fully-open language models. OLMo 2 32B is the most capable and largest model - it is the first full-open model to outperform GPT3.5-Turbo and GPT-4o mini on a popular suite of benchmarks. OLMo 2 7B and 13B are competitive with open-weight models from Meta and Mistral on English academic benchmarks. OLMo 2 1B, the smallest member of the model family, surpasses peer models like Gemma 3 1B and Llama 3.2 1B.
Hugging Face's SmolLM3 is a fully open 3B-parameter language model that supports six languages, strong reasoning capabilities, and long-context processing. It targets high performance in the small model segment.
An open-source memory layer for AI agents that enables long-term, personalized interactions by efficiently storing and retrieving user context across sessions, reducing token costs and improving response accuracy.
A fully open-source, LlamaCloud-backed alternative to NotebookLM, this project uses LlamaCloud for document processing, OpenAI for content generation, and ElevenLabs for voice synthesis.
In the few states where it's legal, in-person psychedelic therapy can cost thousands of dollars per session, gated by lengthy waiting lists. As a cheaper, faster alternative, users are turning to AI tripsitters. Companies are following suit too: Mindbloom, for example, is already integrating AI into its ketamine programs. Experts doubt that AI can provide the emotional attunement they consider essential for safe psychedelic experiences.
OpenAI has reportedly poached high-level engineers from Tesla, xAI, X, Meta, and other companies. The engineers include Tesla's Vice President of Software Engineering David Lau, X and xAI's head of infrastructure engineering Uday Ruddarraju, fellow xAI infrastructure engineer Mike Dalton, and Angela Fan, an AI researcher from Meta. This article looks at the new hires and their backgrounds and specialties.
Sponsored
You'll probably like our flagship newsletter. It's all about tech, science, and programming.
After alleging that DeepSeek improperly copied its models using distillation techniques, OpenAI implemented "information tenting" policies, biometric access controls, and offline isolation of proprietary technology.
The companies are launching the National Academy of AI Instruction with teachers' unions as schools struggle with AI's role in education.
LangChain is an AI infrastructure startup that provides tools for building and monitoring large language model-powered applications.
Sponsored
Rime offers the most realistic and expressive AI voices on the market today, creating agentic experiences that sound like everyday people [4], not robots or voice actors — while driving double digit increases in conversion for brands like Dominos.