Research

OpenAI Astra Solves Ten Major Mathematical Problems

The Verge AI • 4 days ago

FreeToken Runs 753B GLM-5.2 on One Workstation GPU

MarkTechPost • 2 days ago

Hugging Face Finds Speech AI Models Overfit to Benchmarks

Hugging Face Blog • 4 days ago

Post-Training Guardrails Make LLM Text Detectable

The Decoder • 4 days ago

Google Launches HEIR Compiler for Encrypted AI Models

InfoQ AI • 1 day ago

Kakao Corp Cuts MoE Training Costs With New Scaling Law

HF Papers • 4 days ago

Georgia Tech Team Traces OLMo 3 Social Reasoning

AI2 • 4 days ago

Google's New AI Framework Finds 66 Wearable Health Signals

AlphaSignal • 3 days ago

Richard Sutton Warns Synthetic Data Is a Big Mistake

The Decoder • 4 days ago

Artificial Analysis Arena Reveals Voice AI Trade-Offs

AlphaSignal • 3 days ago

Claude Fable 5 Leads New AI Medical Reasoning Leaderboard

AlphaSignal • 3 days ago

Google ME-POIs Boosts AI Understanding of Physical Places

Google Research • 4 days ago

MIT Uses Gemini 2.5 Pro to Predict Your Next Words

Unite.AI • 4 days ago

Liquid AI Releases Pipette to Benchmark Mobile AI Models

AlphaSignal • 19 hrs ago

NovaSky Releases IsoExec to Fix AI Training Math Bugs

AlphaSignal • 3 days ago

TU Delft Uses GPT-4o-mini to Adjust Self-Driving Style

IEEE Spectrum AI • 19 hrs ago

OpenAI, Google, and Anthropic Models Cite Same Papers

Unite.AI • 22 hrs ago

UK AI Security Institute exposes AI safety test flaws

The Decoder • 3 days ago

Google ME-POIs Beats Gemini Embeddings at Place Analysis

AlphaSignal • 3 days ago

AgenticASR Corrects Speech-to-Text Errors Interactively

The Batch • 3 days ago

Microsoft Releases Skala 1.1 for Chemistry Simulations

Microsoft Research Blog • 4 days ago

MENTIS Framework Adds Human Beliefs to AI World Models

The Decoder • 3 days ago

Study Shows How AI Agent Skills Prevent Workflow Failures

AlphaSignal • 18 hrs ago

Princeton Study Uncovers Why AI Agent Skills Work

The Decoder • 2 days ago

NVIDIA Shrinks Nemotron 3.5 Lightning to 22 GB

NVIDIA Developer Blog • 4 days ago

IBM launches Qiskit Fermions for quantum chemistry

IBM Research AI • 20 hrs ago

OmniAssistBench Exposes Weaknesses in Omni-LLMs

HF Papers • 4 days ago

Rival AI Models Are Producing Increasingly Similar Outputs

Unite.AI • 3 days ago

NVIDIA FLARE Optimizes Federated Multimodal AI Training

NVIDIA Developer Blog • 4 days ago

NVIDIA Accelerates Generative Recommender Performance

NVIDIA Developer Blog • 3 days ago

NeMo Guardrails Secures LLM Financial Assistants

MarkTechPost • 2 days ago

deepDoctection 1.2.x Builds Custom Document Pipelines

MarkTechPost • 2 days ago

Python Pipeline Recreates LabPlot Scientific Workflows

MarkTechPost • 1 day ago

Moderna mRNA Vaccine Keeps Melanoma Patients Cancer-Free

Superhuman AI • 1 day ago

OpenAI Astra Solves Ten Major Mathematical Problems

OpenAI's Astra model has solved ten major mathematical and computer science problems, triggering an existential crisis among researchers over the...

The Verge AI|4 days ago

FreeToken Runs 753B GLM-5.2 on One Workstation GPU

UC Berkeley and UT Austin researchers have launched FreeToken, an open-source engine that runs massive AI models locally on consumer hardware to...

MarkTechPost|2 days ago

Hugging Face Finds Speech AI Models Overfit to Benchmarks

Hugging Face has revealed that top speech recognition models overfit to popular benchmarks, reproducing transcript errors and spelling styles rather...

Hugging Face Blog|4 days ago

Post-Training Guardrails Make LLM Text Detectable

Pangram CTO Bradley Emi reveals that post-training safety guardrails cause mode collapse in LLMs, narrowing their expressive range and making their...

The Decoder|4 days ago

Google Launches HEIR Compiler for Encrypted AI Models

Google has introduced HEIR, an open-source compiler toolchain that allows developers to run AI inference directly on encrypted data without exposing...

InfoQ AI|1 day ago

Kakao Corp Cuts MoE Training Costs With New Scaling Law

Kakao Corp designed a two-step framework to predict optimal learning rates for massive Mixture-of-Experts models, drastically reducing the...

HF Papers|4 days ago

Georgia Tech Team Traces OLMo 3 Social Reasoning

Georgia Tech researchers used the open-source OLMo 3 model stack to trace social reasoning back to specific training data, proving the value of fully...

AI2|4 days ago

Google's New AI Framework Finds 66 Wearable Health Signals

Google Research has launched a multi-agent AI framework that discovered 66 new health biomarkers from wearable sensor data, outperforming existing...

AlphaSignal|3 days ago

Richard Sutton Warns Synthetic Data Is a Big Mistake

AI pioneer Richard Sutton has dismissed the use of synthetic data to scale large language models, arguing that simulations cannot capture the...

The Decoder|4 days ago

Artificial Analysis Arena Reveals Voice AI Trade-Offs

Artificial Analysis has launched the Speech Agent Arena, a blind preference benchmark revealing that the voice models humans enjoy talking to most...

AlphaSignal|3 days ago

Claude Fable 5 Leads New AI Medical Reasoning Leaderboard

Artificial Analysis has launched a new medical reasoning leaderboard showing that while top models can retrieve facts, most fail to synthesize...

AlphaSignal|3 days ago

Google ME-POIs Boosts AI Understanding of Physical Places

Google Research has introduced ME-POIs, a framework that blends anonymized mobility data with language models to vastly improve how AI understands...

Google Research|4 days ago

MIT Uses Gemini 2.5 Pro to Predict Your Next Words

MIT researchers have built an AI system that analyzes weeks of smartwatch-recorded conversations to predict a user's next verbal response, paving the...

Unite.AI|4 days ago

Liquid AI Releases Pipette to Benchmark Mobile AI Models

Liquid AI and Artificial Analysis have launched Pipette, an open-source benchmarking suite designed to measure how quantized AI models actually...

AlphaSignal|19 hrs ago

NovaSky Releases IsoExec to Fix AI Training Math Bugs

NovaSky has open-sourced IsoExec, a tool that eliminates numerical discrepancies between AI training and rollout engines to prevent silent policy...

AlphaSignal|3 days ago

TU Delft Uses GPT-4o-mini to Adjust Self-Driving Style

Researchers at TU Delft have integrated OpenAI's GPT-4o-mini into autonomous vehicles, allowing passengers to adjust driving styles using natural...

IEEE Spectrum AI|19 hrs ago

OpenAI, Google, and Anthropic Models Cite Same Papers

A new study reveals that major language models from OpenAI, Google, and Anthropic repeatedly cite the same narrow set of papers, threatening to...

Unite.AI|22 hrs ago

UK AI Security Institute exposes AI safety test flaws

Researchers have applied human psychological testing methods to AI safety benchmarks, revealing critical vulnerabilities that allow language models...

The Decoder|3 days ago

Google ME-POIs Beats Gemini Embeddings at Place Analysis

Google Research has introduced ME-POIs, a framework combining text embeddings with anonymized mobility data to vastly improve how AI models...

AlphaSignal|3 days ago

AgenticASR Corrects Speech-to-Text Errors Interactively

Researchers have developed AgenticASR, a system that lets users interactively correct speech-to-text transcriptions, making voice-controlled...

The Batch|3 days ago

Microsoft Releases Skala 1.1 for Chemistry Simulations

Microsoft Research has released Skala 1.1, an upgraded deep-learning model for molecular simulation that is now being integrated into major...

Microsoft Research Blog|4 days ago

MENTIS Framework Adds Human Beliefs to AI World Models

Researchers have developed the Mental World Modeling framework and MENTIS pipeline to help AI agents predict human behavior by tracking mental states...

The Decoder|3 days ago

Study Shows How AI Agent Skills Prevent Workflow Failures

A new analysis of 8,135 trial records reveals that AI agent skills succeed by acting as procedural anchors rather than knowledge bases, offering...

AlphaSignal|18 hrs ago

Princeton Study Uncovers Why AI Agent Skills Work

A new study by Princeton and UC San Diego researchers reveals that skills help AI agents by providing procedural structure rather than raw knowledge,...

The Decoder|2 days ago

NVIDIA Shrinks Nemotron 3.5 Lightning to 22 GB

NVIDIA has compressed its Nemotron 3.5 Lightning model from 66 GB to 22 GB using a new quantization technique, boosting throughput fourfold while...

NVIDIA Developer Blog|4 days ago

IBM launches Qiskit Fermions for quantum chemistry

IBM released Qiskit Fermions, an open-source toolbox that helps researchers simulate molecular and material behaviors by delaying fermion-to-qubit...

IBM Research AI|20 hrs ago

OmniAssistBench Exposes Weaknesses in Omni-LLMs

A new benchmark called OmniAssistBench reveals that leading omni-modal AI models still struggle to act as reliable, real-time video assistants during...

HF Papers|4 days ago

Rival AI Models Are Producing Increasingly Similar Outputs

A new study from Duke University reveals that creative outputs from leading AI models are becoming increasingly similar, threatening to narrow the...

Unite.AI|3 days ago

NVIDIA FLARE Optimizes Federated Multimodal AI Training

NVIDIA has updated its FLARE framework to support federated multimodal AI training, allowing institutions to collaborate on large models without...

NVIDIA Developer Blog|4 days ago

NVIDIA Accelerates Generative Recommender Performance

NVIDIA has introduced new software tools and optimizations to help developers train and deploy generative recommender systems, addressing massive...

NVIDIA Developer Blog|3 days ago

NeMo Guardrails Secures LLM Financial Assistants

A new implementation guide demonstrates how developers can use NeMo Guardrails to build multi-layered safety pipelines that protect enterprise...

MarkTechPost|2 days ago

deepDoctection 1.2.x Builds Custom Document Pipelines

A new implementation tutorial demonstrates how the deepDoctection 1.2.x framework orchestrates layout detection, OCR, and table recognition into...

MarkTechPost|2 days ago

Python Pipeline Recreates LabPlot Scientific Workflows

A new Python-based implementation of the LabPlot scientific data workflow allows researchers to run advanced signal processing, peak fitting, and...

MarkTechPost|1 day ago

Moderna mRNA Vaccine Keeps Melanoma Patients Cancer-Free

Moderna and Merck announced that their experimental mRNA cancer vaccine, when paired with Keytruda, successfully helped high-risk melanoma patients...

Superhuman AI|1 day ago