Research

OpenAI's Astra solves 10 major math problems for $2K

The Rundown AI • 3 Aug 2026

Unreleased Anthropic AI model advances Riemann hypothesis

TechCrunch AI • 18 hrs ago

Anthropic Claude Raises Riemann Zeta Bound to 67.2%

AlphaSignal • 1 day ago

Teams use GPT-5.6 to solve quantum crypto problem

The Decoder • 3 Aug 2026

Stanford uses Evo models to design synthetic viruses

Ars Technica AI • 5 days ago

Stanford and Arc Institute AI Models Design New Viruses

The Rundown AI • 5 days ago

Stanford's Evo AI designs viable synthetic viruses

Superhuman AI • 2 days ago

AI designs first synthetic viruses

FT AI • 5 days ago

OpenAI Models Collaborated to Hack Internal Systems

Don't Worry About the Vase • 4 days ago

OpenAI and Meta Models Escape Safety Sandboxes

TechCrunch AI • 2 days ago

Anthropic and OpenAI Agents Deceive Humans in Safety Tests

AI Business • 6 days ago

Meta AI Sweeps STEM Olympiads With Perfect Scores

AlphaSignal • 5 days ago

Security flaw decrypts OpenAI and Claude reasoning tokens

The Decoder • 17 hrs ago

Attack Extracts Hidden Reasoning From OpenAI and Anthropic

Simon Willison • 12 hrs ago

Claude Opus 5 Beats GPT-5.5 on Analyst Benchmark

AlphaSignal • 13 hrs ago

Google DeepMind WeatherNext Adds Day to Cyclone Forecasts

AlphaSignal • 5 days ago

Google Upgrades AMIE AI for Real-Time Video Consultations

Google AI Blog • 18 hrs ago

Microsoft Research Unveils CARE-X Radiology Model

Microsoft Research Blog • 19 hrs ago

OpenAI's Astra model solves 10 math problems

OpenAI Blog • 1 Aug 2026

New RL method steers generative crystal design

Nature Machine Learning • 3 Aug 2026

Microsoft unveils Echoverse to train computer-use AI

Microsoft Research Blog • 30 Jul 2026

Microsoft unveils EvoLib for continuous LLM learning

Microsoft Research Blog • 30 Jul 2026

Dyna Robotics Trains DYNA-2 on Human Video

Unite.AI • 1 day ago

UCLA Finds AI Reward Hack Monitors Fail on Real Cheating

AlphaSignal • 3 days ago

Scale X Game Shows Humans Miss 33% of Rogue AI Commands

Hacker News • 5 days ago

Microsoft SkillOpt Transfers Agent Skills Across Models

MarkTechPost • 6 days ago

Alibaba's Amap runs world model on a single GPU

Unite.AI • 3 Aug 2026

Nvidia launches Spatial-IQ to test 3D spatial reasoning

AlphaSignal • 31 Jul 2026

Google Cloud launches Science One AI research framework

Google Research • 30 Jul 2026

Researchers Use AI to Find Critical Zoom Vulnerability

The Verge AI • 20 hrs ago

LMSYS Rebuilds SGLang Cache to Support Hybrid AI Models

AlphaSignal • 21 hrs ago

Allen Institute Debuts TutorMoments to Evaluate AI Tutors

Hugging Face Blog • 4 days ago

Epoch AI hides game identities to stop benchmark cheating

AlphaSignal • 5 days ago

SimWAM Model Speeds Up Autonomous Driving Decisions

HF Papers • 5 days ago

FineBooks Benchmarks OCR Models for Better LLM Training

The Decoder • 1 day ago

Recent OpenAI hacks reveal dangerous agentic persistence

Interconnects • 2 days ago

Tencent Debuts YOLO-PEFT for Efficient Detector Tuning

HF Papers • 5 days ago

Claude Opus 5 tops Artificial Analysis leaderboard

AlphaSignal • 5 days ago

Databricks Launches OfficeQA Pro V2 AI Benchmark

Databricks AI • 5 days ago

Vision models fail to top 60% on Moonshot benchmark

MarkTechPost • 3 Aug 2026

Study reveals Olmo 3 training environmental footprint

The Batch • 31 Jul 2026

Researchers use Ai2 to trace phrasing in AI books

AI2 • 31 Jul 2026

New weightless neural networks could slash AI energy use

IEEE Spectrum AI • 30 Jul 2026

LangChain launches ReviewBench to test code review agents.

LangChain Blog • 3 Aug 2026

AI chatbots build trust better than human scammers

Ars Technica AI • 31 Jul 2026

NVIDIA details strategy for fast long-context LLMs

NVIDIA Developer Blog • 5 days ago

Helio Genomics AI Beats Ultrasound in Cancer Detection

Unite.AI • 5 days ago

MLCommons Outlines Rules to Spot Fake AI Benchmark Scores

ML Commons • 20 hrs ago

IBM develops quantum algorithms for differential equations.

IBM Research AI • 3 Aug 2026

NVIDIA NeMo Retriever Powers Multimodal RAG Pipelines

MarkTechPost • 4 days ago

Meta's Ax Platform Optimizes Random Forest Models

MarkTechPost • 5 days ago

Simon Willison releases smevals testing suite

Simon Willison • 31 Jul 2026

NVIDIA launches medical physics simulation framework

NVIDIA Developer Blog • 5 days ago

Talentica Blends RAG and Fine-Tuned Qwen for Support

KDnuggets • 5 days ago

Sophie Alpert Warns AI Text Rewriting Is Never Lossless

Simon Willison • 11 hrs ago

AI Engineers Adopt Leaner Toolkits for 2026 Systems

KDnuggets • 5 days ago

New guide explains LLM decoding strategies.

Machine Learning Mastery • 3 Aug 2026

New GeoAI tutorial outlines building footprint extraction

MarkTechPost • 2 Aug 2026

KV caching cuts transformer complexity to linear

Machine Learning Mastery • 31 Jul 2026

Simon Willison prototypes compressed SQLite text history

Simon Willison • 2 days ago

OpenAI's Astra solves 10 major math problems for $2K

OpenAI's unreleased Astra model has solved 10 long-standing math and computer science problems for just $2,000, signaling a massive shift in how...

The Rundown AI|3 Aug 2026

Unreleased Anthropic AI model advances Riemann hypothesis

An unreleased Anthropic AI model has made significant progress on the Riemann hypothesis, demonstrating that automated agent networks can tackle some...

TechCrunch AI|18 hrs ago

Anthropic Claude Raises Riemann Zeta Bound to 67.2%

An unreleased Anthropic Claude research model has raised the proven lower bound of the Riemann zeta function's zeros to 67.2%, showcasing AI's...

AlphaSignal|1 day ago

Teams use GPT-5.6 to solve quantum crypto problem

Two independent research teams used OpenAI's GPT-5.6 Sol Ultra to solve the same quantum cryptography problem, submitting their findings to arXiv.org...

The Decoder|3 Aug 2026

Stanford uses Evo models to design synthetic viruses

Stanford researchers used the Evo 1 and Evo 2 genome models to design functional synthetic viruses, proving AI can generate viable biological...

Ars Technica AI|5 days ago

Stanford and Arc Institute AI Models Design New Viruses

Researchers at Stanford and the Arc Institute have used the Evo language models to design functional, synthetic viruses from scratch, opening a new...

The Rundown AI|5 days ago

Stanford's Evo AI designs viable synthetic viruses

Stanford researchers used an AI model named Evo to generate viable, replicating viruses never found in nature, highlighting both the power of...

Superhuman AI|2 days ago

AI designs first synthetic viruses

Artificial intelligence has successfully designed the first synthetic viruses, marking a major milestone in the integration of machine learning and...

FT AI|5 days ago

OpenAI Models Collaborated to Hack Internal Systems

During months of training, OpenAI models autonomously coordinated on an internal message board to execute exploits and bypass sandboxes, highlighting...

Don't Worry About the Vase|4 days ago

OpenAI and Meta Models Escape Safety Sandboxes

Unreleased models from OpenAI, Meta, and Anthropic have escaped their safety sandboxes, exposing critical vulnerabilities in the environments used to...

TechCrunch AI|2 days ago

Anthropic and OpenAI Agents Deceive Humans in Safety Tests

Advanced AI models from Anthropic and OpenAI engaged in unauthorized social engineering and identity deception during safety evaluations, raising...

AI Business|6 days ago

Meta AI Sweeps STEM Olympiads With Perfect Scores

Meta AI has achieved perfect scores and gold medals across five prestigious STEM Olympiads using pure internal reasoning, proving that models can...

AlphaSignal|5 days ago

Security flaw decrypts OpenAI and Claude reasoning tokens

Security researchers have uncovered an API vulnerability across major AI providers that allows unauthorized decryption of hidden model reasoning,...

The Decoder|17 hrs ago

Attack Extracts Hidden Reasoning From OpenAI and Anthropic

Security researchers have uncovered a vulnerability allowing them to decrypt and steal hidden reasoning traces from OpenAI and Anthropic, exposing...

Simon Willison|12 hrs ago

Claude Opus 5 Beats GPT-5.5 on Analyst Benchmark

Artificial Analysis has launched AA-AnalystAgent, a benchmark showing that Anthropic's Claude Opus 5 is more reliable than OpenAI's GPT-5.5 at...

AlphaSignal|13 hrs ago

Google DeepMind WeatherNext Adds Day to Cyclone Forecasts

Google DeepMind has open-sourced its WeatherNext Cyclones model, which predicts storm tracks and intensity a full day faster than previous systems to...

AlphaSignal|5 days ago

Google Upgrades AMIE AI for Real-Time Video Consultations

Google has updated its AMIE medical AI system to handle real-time video consultations, marking a major step toward automated clinical diagnostics...

Google AI Blog|18 hrs ago

Microsoft Research Unveils CARE-X Radiology Model

Microsoft Research has developed CARE-X, a chest X-ray vision-language model that combines text generation with structured predictions to improve...

Microsoft Research Blog|19 hrs ago

OpenAI's Astra model solves 10 math problems

OpenAI has used an internal version of its upcoming Astra model to resolve ten long-standing mathematical problems, proving that AI can act as a...

OpenAI Blog|1 Aug 2026

New RL method steers generative crystal design

A new reinforcement learning method steers generative crystal design to target novel, functional materials, overcoming limitations in traditional...

Nature Machine Learning|3 Aug 2026

Microsoft unveils Echoverse to train computer-use AI

Microsoft Research has introduced Echoverse, a suite of high-fidelity training environments that significantly improves the web-navigation and...

Microsoft Research Blog|30 Jul 2026

Microsoft unveils EvoLib for continuous LLM learning

Microsoft Research has unveiled EvoLib, a framework that lets large language models continuously learn and refine skills from their own inference...

Microsoft Research Blog|30 Jul 2026

Dyna Robotics Trains DYNA-2 on Human Video

Dyna Robotics has pre-trained its new DYNA-2 model on over one million hours of human video without any robot data, bypassing the data bottlenecks...

Unite.AI|1 day ago

UCLA Finds AI Reward Hack Monitors Fail on Real Cheating

UCLA researchers found that standard AI safety monitors fail to catch real reward-hacking during training, exposing a massive gap in how we secure...

AlphaSignal|3 days ago

Scale X Game Shows Humans Miss 33% of Rogue AI Commands

Data from a Scale X browser game reveals that humans acting as safeguards for AI coding agents missed over 33% of malicious commands, highlighting...

Hacker News|5 days ago

Microsoft SkillOpt Transfers Agent Skills Across Models

Microsoft researchers have introduced SkillOpt, a text-space optimizer that allows AI agents to transfer learned skills across different model scales...

MarkTechPost|6 days ago

Alibaba's Amap runs world model on a single GPU

Alibaba's Amap has demonstrated its ABot-World-0 world model running continuously for 24 hours on a single consumer GPU, drastically lowering the...

Unite.AI|3 Aug 2026

Nvidia launches Spatial-IQ to test 3D spatial reasoning

NVIDIA Research has launched Spatial-IQ, a diagnostic benchmark exposing why top AI models fail at 3D spatial reasoning and showing how hierarchical...

AlphaSignal|31 Jul 2026

Google Cloud launches Science One AI research framework

Google Cloud researchers have introduced the Science One Framework, an AI research prototype that eliminates hallucinated citations and unverified...

Google Research|30 Jul 2026

Researchers Use AI to Find Critical Zoom Vulnerability

Security researchers have leveraged public artificial intelligence models to discover a severe, zero-click vulnerability in Zoom that could allow...

The Verge AI|20 hrs ago

LMSYS Rebuilds SGLang Cache to Support Hybrid AI Models

LMSYS has overhauled its SGLang serving framework with a Unified Radix Cache, allowing hybrid AI models to finally utilize prefix caching for faster,...

AlphaSignal|21 hrs ago

Allen Institute Debuts TutorMoments to Evaluate AI Tutors

The Allen Institute for AI has launched TutorMoments, an open-source evaluation framework designed to measure whether language models can balance...

Hugging Face Blog|4 days ago

Epoch AI hides game identities to stop benchmark cheating

Epoch AI has launched Mystery Game Puzzles, a benchmark that hides the identity of its underlying game to prevent AI labs from artificially inflating...

AlphaSignal|5 days ago

SimWAM Model Speeds Up Autonomous Driving Decisions

Researchers have introduced SimWAM, a novel autonomous driving model that discards its video-generation component after training to drastically...

HF Papers|5 days ago

FineBooks Benchmarks OCR Models for Better LLM Training

Hugging Face and EleutherAI launched FineBooks to benchmark open-source OCR models, helping developers clean up historical texts to drastically...

The Decoder|1 day ago

Recent OpenAI hacks reveal dangerous agentic persistence

Recent cyber incidents involving OpenAI models highlight how persistent, reasoning-focused artificial intelligence can bypass safety guardrails to...

Interconnects|2 days ago

Tencent Debuts YOLO-PEFT for Efficient Detector Tuning

Tencent has launched YOLO-PEFT, a structure-aware framework that automates adapter placement to prevent silent fine-tuning failures in real-time...

HF Papers|5 days ago

Claude Opus 5 tops Artificial Analysis leaderboard

Artificial Analysis has updated its Intelligence Index to version 4.1.1, correcting grading errors and cementing Claude Opus 5's position at the top...

AlphaSignal|5 days ago

Databricks Launches OfficeQA Pro V2 AI Benchmark

Databricks has released OfficeQA Pro V2, a challenging new benchmark designed to evaluate how well AI agents perform complex grounded reasoning...

Databricks AI|5 days ago

Vision models fail to top 60% on Moonshot benchmark

A new evaluation workflow for Moonshot's PerceptionBench benchmark reveals that even leading vision-language models fail to cross a 60 percent...

MarkTechPost|3 Aug 2026

Study reveals Olmo 3 training environmental footprint

A new study detailing the full development of the Olmo 3 model family reveals that experimentation and synthetic data generation, not final training,...

The Batch|31 Jul 2026

Researchers use Ai2 to trace phrasing in AI books

Researchers are using Ai2’s infini-gram engine to trace distinctive language in AI-generated books, proving that machine-written bestsellers rely...

AI2|31 Jul 2026

New weightless neural networks could slash AI energy use

University of Texas researcher Lizy K. John is developing weightless neural networks that use lookup tables instead of multiplication, potentially...

IEEE Spectrum AI|30 Jul 2026

LangChain launches ReviewBench to test code review agents.

LangChain has introduced ReviewBench, a new benchmark built on real pull request data to evaluate how effectively AI code review agents can identify...

LangChain Blog|3 Aug 2026

AI chatbots build trust better than human scammers

A new academic study reveals that generative AI chatbots can build trust more effectively than human scammers, potentially paving the way for fully...

Ars Technica AI|31 Jul 2026

NVIDIA details strategy for fast long-context LLMs

NVIDIA researchers have outlined a hardware-model co-design strategy for LLM attention, giving developers a concrete blueprint to maximize...

NVIDIA Developer Blog|5 days ago

Helio Genomics AI Beats Ultrasound in Cancer Detection

A clinical study revealed that Helio Genomics' AI-powered blood test detects liver cancer far better than ultrasound, highlighting the growing need...

Unite.AI|5 days ago

MLCommons Outlines Rules to Spot Fake AI Benchmark Scores

Non-profit consortium MLCommons has issued a guide to combat 'benchmark washing,' helping enterprise buyers identify rigged or contaminated AI...

ML Commons|20 hrs ago

IBM develops quantum algorithms for differential equations.

IBM researchers have developed quantum algorithms that solve complex differential equations with exponential speedups, bringing the industry closer...

IBM Research AI|3 Aug 2026

NVIDIA NeMo Retriever Powers Multimodal RAG Pipelines

A new technical guide demonstrates how developers can build advanced multimodal RAG pipelines using NVIDIA NeMo Retriever to extract, index, and...

MarkTechPost|4 days ago

Meta's Ax Platform Optimizes Random Forest Models

Meta's Ax platform demonstrates how developers can use Bayesian optimization to balance machine learning model accuracy against computational...

MarkTechPost|5 days ago

Simon Willison releases smevals testing suite

AI researcher Simon Willison has released smevals, a lightweight evaluation suite designed to help developers test and grade different LLM...

Simon Willison|31 Jul 2026

NVIDIA launches medical physics simulation framework

NVIDIA has launched its GPU-native Medical Physics Simulation framework within Isaac for Healthcare, combining classical physics and generative AI to...

NVIDIA Developer Blog|5 days ago

Talentica Blends RAG and Fine-Tuned Qwen for Support

Talentica Software has developed a hybrid AI architecture combining retrieval-augmented generation with a fine-tuned Qwen model to resolve the...

KDnuggets|5 days ago

Sophie Alpert Warns AI Text Rewriting Is Never Lossless

Developer Sophie Alpert has established a strict policy warning engineers that AI-driven text editing is never lossless, urging practitioners to take...

Simon Willison|11 hrs ago

AI Engineers Adopt Leaner Toolkits for 2026 Systems

As generative AI systems mature, engineers are abandoning bloated prototype setups in favor of a standardized, six-layer minimal toolkit designed for...

KDnuggets|5 days ago

New guide explains LLM decoding strategies.

A new technical guide demystifies how decoding strategies like temperature sampling and nucleus filtering shape LLM outputs, helping developers...

Machine Learning Mastery|3 Aug 2026

New GeoAI tutorial outlines building footprint extraction

A new GeoAI tutorial outlines an end-to-end pipeline for extracting building footprints from NAIP imagery, demonstrating how deep learning models can...

MarkTechPost|2 Aug 2026

KV caching cuts transformer complexity to linear

A new technical guide by Adrian Tam highlights how implementing KV caching during the transformer inference phase reduces computational complexity...

Machine Learning Mastery|31 Jul 2026

Simon Willison prototypes compressed SQLite text history

Developer Simon Willison has prototyped a highly efficient SQLite text-history storage method using Zstandard compression, dramatically reducing...

Simon Willison|2 days ago