GPT-5.2 Discovers Formula for Previously Assumed Impossible Particle Interaction in Theoretical Physics
OpenAI's GPT-5.2 model has made a breakthrough discovery, proposing a formula for gluon interactions that contradicts standard physics textbooks. The formula was later verified by human physicists.
Cloudflare Boosts Engineering Team by 45% Post-Layoffs with AI Investment
Cloudflare cuts 1,100 jobs, but engineering team grows due to increased investment in AI. CEO Matthew Prince divides roles into builders, sellers, and measurers, eliminating the latter.
Princeton Researchers Develop Comprehensive Evaluation Framework for Long-Context Language Models
A new benchmarking technology, HELMET, addresses limitations in evaluating long-context language models. The framework offers reliable and consistent rankings of frontier LCLMs.
OpenAI and Paradigm Introduce EVMbench for AI-Powered Smart Contract Security Evaluation
A new benchmark, EVMbench, measures AI agents' ability to detect and exploit smart contract vulnerabilities in economically meaningful scenarios. OpenAI's Codex model outperformed GPT-5 in the Exploit
Apple VP Paul Meade Joins OpenAI's Hardware Division
Paul Meade, vice president of hardware engineering at Apple, leaves for OpenAI's AI-powered devices division. John Ternus set to take over as CEO.
Gradio: More Than Just a UI Library for Machine Learning Applications
Discover why Gradio stands out as a powerful framework for building and deploying ML models, with features like universal API access, interactive API recorder, and automatic queue management.
Allen Institute's olmOCR 2 Achieves State-of-the-art OCR Performance for English Documents
The Allen Institute for Artificial Intelligence has released olmOCR 2, a significant update to its end-to-end vision-language approach to reading complex documents. The new model scores 82.4 points on
BOMB Magazine Announces First Proof Submissions | Importance of Proofing in Academic and Adult Publishing
Academic journal BOMB Magazine announces First Proof submissions, emphasizing the importance of proofing stage in publishing. CW Authors stresses its significance in journal article production.
FBI, CISA Warn of Russian Phishing Campaign Targeting Signal Users
Russian Intelligence Services are targeting high-value individuals using Signal in a phishing campaign that compromised thousands of accounts. Attackers exploit Signal's linked device feature.
Intel's AutoRound: Advanced Quantization Algorithm for LLMs and VLMs
Intel introduces AutoRound, an advanced quantization algorithm that achieves high accuracy at ultra-low bit widths while minimizing tuning. Integrated into LLM Compressor, it offers superior accuracy,
OpenAI Partners with Four Consulting Firms to Address Enterprise AI Deployment Gap
OpenAI's new platform aims to help businesses scale AI, addressing a $30-40 billion investment gap in enterprise AI.
Advancing Robotic Generalization: LeRobot Community's Strides in VLA Models
LeRobot community's efforts to advance robotic generalization through diverse data collection have implications for content creation and moderation systems.
OpenAI Halts Evaluation of SWE-bench Verified Due to Flawed Test Cases
OpenAI announces it will no longer evaluate SWE-bench Verified due to issues with test cases rejecting correct solutions and training data contamination, causing inflated scores without genuine improv
Critical Security Flaws Expose AI Coding Agents in Adult Industry
Researchers uncover vulnerabilities in Claude Code and GitHub Copilot, putting developer environments at risk. Attackers can steal tokens and execute code without user approval.
Hugging Face Unveils Faster Whisper Audio Transcription Service
New deployment leverages vLLM project for up to 8x faster performance without sacrificing quality. Benefits industries relying on audio transcription.
Figma Expands Partner Ecosystem with 338 Partners for Enhanced Product Development
Figma's Partner Program now includes over 338 partners across multiple industries, offering a range of services to boost product development. Notable partnership is with Merkle as Signature Service Pa
Hugging Face Transformers Library Updates Model-Definition Framework
The update aims to standardize model definitions, simplify contributions, and support over 300 model architectures in the ML ecosystem.
Sakana AI Launches Fugu, Challenging Anthropic's Mythos in AI Market
Tokyo-based Sakana AI unveils Fugu, a new orchestration model matching Mythos on key benchmarks. The launch comes amidst US export ban on Anthropic's advanced models.
AWS and OpenAI Develop Stateful Runtime for Advanced AI Agents on Amazon Bedrock
The new Stateful Runtime Environment enables persistent state, memory, and secure execution across multi-step workflows, powered by OpenAI's models. It provides a seamless experience for developers bu
TII's Falcon Arabic Language Model Sets New Benchmark for Arabic NLP
The Technology Innovation Institute (TII) in Abu Dhabi has released Falcon Arabic, a 7B parameter multilingual model that outperforms other regionally available Arabic language models.
Falcon-H1 Series: New Large Language Models with Hybrid Architecture
Researchers introduce a parallel hybrid architecture combining Transformer-based attention mechanisms and State Space Models (SSMs), resulting in faster inference, lower memory usage, and strong gener
South Korean Truth Commission Raises Questions on Legitimacy of Adoptions
Investigation into past adoption practices in South Korea reveals deceit and exploitation, raising concerns for U.S. agencies and potentially affecting the number of available children for adoption in
Asian AI Startups Launch Rival Models After US Bans Anthropic's Fable 5
The US government's ban on Anthropic's advanced AI models has sparked a response from Asian companies, who are launching their own rival models. This development has significant implications for the g
Hugging Face's New Study Improves AI Agent Performance with CodeAct
Research led by Xingyao Wang et al. presents a novel paradigm called CodeAct Agent Framework, outperforming alternatives by up to 20% when integrated with Python interpreter. Implications for AI agent
White House Launches AI Education Initiative for Youth with Google, IBM, and Code.org
President Trump signs Executive Order to promote AI education among America's youth, partnering with leading organizations like Google, IBM, and Code.org to provide free resources.