Search the archive

CODE INTERPRETER REASONING

CoRT post‑trains large reasoning models with hint engineering so they can delegate calculations to external code interpreters.

TREERL FOR ON‑POLICY LLM TRAINING

TreeRL uses on‑policy tree search and intermediate supervision to train LLMs without a separate reward model, delivering stronger math and code reasoning than ChainRL.

GROQ JOINS HUGGING FACE INFERENCE

Groq's low‑latency hardware is now a selectable inference provider on the Hub and SDKs, adding fast serverless access to models like Llama 4 and Qwen 32B.

AI USE AT WORK HAS NEARLY DOUBLED IN TWO YEARS

A Gallup survey found that 40% percent of U.S. employees now use AI at work, with white-collar workers driving adoption. Technology workers lead with 50% frequent AI use while production and frontline workers remain stuck at 9%.

See how it works.

CHARACTER AI'S WRITING EVALUATION FRAMEWORK

Character AI outlined a new framework that scores conversational storytelling by combining creative writing heuristics with objective metrics to judge pacing, engagement, and fun.

MODELS.DEV

This site contains an open-source database of AI models that lists AI model specifications, pricing, and capabilities.

TIKTOK LAUNCHES AI-GENERATED PRODUCT MODELS AND REVIEWERS

TikTok's Symphony platform will now generate videos of virtual avatars modeling products and trying on clothes, potentially replacing human influencers.

ASTRAZENECA SIGNS $5B AI RESEARCH DEAL WITH CHINA'S CSPC

AstraZeneca will pay $110 million upfront for CSPC to use AI-driven research to discover small molecule therapies, including an oral treatment for immunological diseases.

Get access to training infrastructure on Baseten

AMD'S AI FUTURE IS RACK SCALE 'HELIOS'

AMD's MI400 will rival Nvidia's Blackwell chips with rack-scale architecture that enables thousands of GPUs to function as unified systems. The company claims 40% better tokens/$ compared to NVIDIA. Its roadmap includes a path to 20x rack-scale energy efficiency by 2030.

GOOGLE, SCALE AI'S LARGEST CUSTOMER, PLANS SPLIT AFTER META DEAL

Meta's $14 billion acquisition of a 49% stake in Scale AI prompted Google to pull its planned $200 million contract for human-labeled training data over fears of exposing sensitive data to Meta. Microsoft, xAI, and OpenAI are also backing away from Scale AI over the same competitive concerns. The exodus benefits Scale's competitors, with Labelbox expecting "hundreds of millions" in new revenue as AI labs seek neutral data providers or move operations in-house.

HAVE LLMS FINALLY MASTERED GEOLOCATION?

Open-source intelligence researchers tested 20 AI models on 500 geolocation challenges using unpublished travel photos to ensure models couldn't rely on memorized training data. OpenAI's latest models outperformed Google Lens by cross-referencing architectural styles, vegetation patterns, and partially visible text, while competitors like Claude often managed only continent-level guesses. However, all models still hallucinated. "Deep research" modes paradoxically performed worse than standard versions.

LOW-BIT QUANTIZATION WITH PARETOQ

ParetoQ is a new training algorithm that unifies binary, ternary, and 2-to-4 bit quantization, achieving state-of-the-art results across all levels.

Schedule a demo to learn more

NEW INSIGHTS FOR SCALING LAWS IN AUTONOMOUS DRIVING

Research from Waymo confirms that, similar to language modeling, increased data and compute resources can enhance the performance of autonomous vehicles. The finding has exciting implications for the development of autonomous vehicles as researchers and developers now know with certainty that enriching the quality and size of the data and models will deliver better performance. It opens up the possibility of devising more adaptive training strategies for planning tasks in robots.

THE AI EVAL FLYWHEEL: SCORERS, DATASETS, PRODUCTION USAGE, & RAPID ITERATION

There was a surprising consistency in the general eval frameworks discussed at the 2025 AI Engineer World's Fair. Most of them involved structuring inputs and how outputs are evaluated and evolving these inputs and evals based on real production usage. The more they can iterate, the better the experience they can deliver, so they try to make eval flywheels as quick and frictionless as possible. One of the key ideas is to have 'playgrounds' to make it easy to tweak a feature and run it against datasets and evals.

HOW ANTHROPIC BUILT THEIR DEEP RESEARCH SYSTEM

In this engineering blog post, Anthropic details its findings in prompt design, tool coordination, and production reliability challenges when orchestrating multi-agent systems. Its design uses an orchestrator-worker pattern where a lead agent spawns specialized sub-agents that search in parallel, far outperforming a single-agent Opus-based method. Token usage alone explains 80% of performance variance, with multi-agent systems consuming 15x more tokens than a regular chat but enabling much more complex research tasks.

SELF-ADAPTING LANGUAGE MODELS

A new training approach enables LLMs to generate "self-edits" that produce persistent weight updates through supervised fine-tuning. The framework outperformed GPT-4.1 despite using a smaller model, but suffered from catastrophic forgetting and required 15x more tokens than standard inference. This addresses the looming data wall and limitations on personalization and memory by enabling models to bootstrap their own improvement through self-generated training material rather than relying on external human-generated text.

GOOGLE'S VEO FOR LIVE-ACTION VIDEOS

Google DeepMind collaborated with filmmakers to produce "ANCESTRA," a hybrid film blending live-action with Veo-generated video. New Veo features enabled personalized, motion-matched generative content for seamless visual integration.

FEATHERLESS AI ON HUGGING FACE

Featherless AI is now available as an inference provider on Hugging Face. It offers serverless access to a wide variety of models from Meta, Qwen, DeepSeek, and others.

AUTOMATED ISSUE RESOLUTION DATA PIPELINES

SWE-Factory provides automated training and evaluation pipelines for GitHub issue resolution tasks. It is powered by LLM-based multi-agent systems.

lightweight SDK. Get your free API key

GOOGLE TESTS AUDIO OVERVIEWS FOR SEARCH

The feature converts existing AI Overviews text summaries into audio format using NotebookLM's podcast-style functionality.

INSTITUTIONAL BOOKS 1.0 (HUGGING FACE REPO)

Harvard Library and Google Books released 242 billion tokens from almost 1 million public domain books as a high-quality training dataset.

THE CLAUDE BLISS ATTRACTOR

Claude, when talking to copies of itself, will eventually turn the conversation into a discussion on spiritual bliss, Buddhism, and the nature of consciousness.

AN INTRODUCTION TO GOOGLE'S APPROACH TO AI AGENT SECURITY

A recent paper from Google describes key risks involved in deploying AI agents and the company's framework for securing them.

🔗 GET STARTED NOW

BYTEDANCE TOPS VIDEO GENERATION BENCHMARKS WITH SEEDANCE 1.0

The TikTok parent's new model tops leaderboards for both text-to-video and image-to-video tasks, beating Google's Veo 3 and OpenAI's Sora. Seedance generates 5 seconds of HD video in 41 seconds and natively supports multi-shot storytelling though it lacks audio generation.

SCALE AI ANNOUNCES NEXT PHASE OF COMPANY'S EVOLUTION

Scale has officially announced its next phase. With Meta investing $14.3B, Scale's founder Alexandr Wang will join Meta's AI initiatives. Chief Strategy Officer Jason Droege will step in as Scale's Interim CEO.

NETFLIX BUILDS UNIFIED DATA ARCHITECTURE

Netflix introduced UDA to address challenges from fragmented data models across its platforms. UDA provides a conceptual foundation that connects and enforces consistency across systems, reducing duplication, improving terminology, and enhancing data quality.

HOW CURSOR IS BUILDING THE FUTURE OF AI CODING WITH CLAUDE (30 MINUTE VIDEO)

Cursor's leadership discusses tackling AI coding's biggest unsolved problem: verifying generated code matches not just functionality but organizational style and unwritten architectural knowledge. The company's background agents work autonomously in isolated environments, representing a shift toward AI that codes independently while developers focus on higher-level design decisions.

A16Z'S 16 CHANGES TO AI IN THE ENTERPRISE

Enterprise AI budgets have grown 75% beyond already high expectations, with OpenAI, Google, and Anthropic emerging as clear leaders. A survey of 100 CIOs shows companies now use 5+ models strategically by use case, with AI-native apps like Cursor achieving 90% code generation rates that are driving a fundamental "buy vs build" shift in enterprise software procurement.

real time—on the edge—using Dell's 7875 Precision workstation

with NVIDIA RTX PRO GPU acceleration. READ THE SCIENCE BEHIND THE

SPEED →

AGENTIC CODING RECOMMENDATIONS

Agentic coding is rapidly evolving - workflows today may look dramatically different tomorrow. However, it is clear that integrating agents into development processes can unlock significant productivity gains. Developers should keep experimenting - while tools and techniques will evolve, the core principles will remain essential. The goal is to write better, more maintainable, and resilient code, not just to leverage agents to write code faster.

HOW I USE CLAUDE CODE

Claude Code's flat pricing allows it to become a daily driver for many tasks. This article discusses how it can change how you approach many programming tasks. While it's not perfect, there are real productivity gains - the key is finding the right balance between automation and human oversight. Developers should experiment with different approaches.

BETTER VISUAL GROUNDING FOR LVLMS

ReVisiT enhances decoding for large vision-language models by guiding generation using internal vision tokens.

META AI APP IS A PRIVACY NIGHTMARE

Users are publicly posting deeply personal queries about tax evasion, medical conditions, and legal troubles through Meta AI's social feed, apparently unaware that their conversations are visible to everyone.

WINDSURF LAUNCHES AI-INTEGRATED BROWSER

The browser automatically shares tab content and DOM access with Windsurf's Cascade AI to eliminate the need to copy-paste when referencing documentation or debugging web apps.

NVIDIA ‘CLIMATE IN A BOTTLE' OPENS A VIEW INTO EARTH'S FUTURE. WHAT WILL WE DO WITH IT?

Nvidia's new generative foundation model enables simulations of Earth's global climate with an unprecedented level of resolution.

GOOGLE HAS A NEW AI MODEL AND WEBSITE FOR FORECASTING TROPICAL STORMS

Google DeepMind and Google Research have launched a site called Weather Lab to share AI weather models, like its new experimental model for forecasting cyclones.

THE DARWIN GÖDEL MACHINE: AI THAT IMPROVES ITSELF BY REWRITING ITS OWN CODE

Sakana AI created a coding agent that reads its own codebase, proposes modifications, and then tests whether those changes would improve its performance on coding tasks.