AllLLMRoboticsStartupsResearchEthicsBusinessHardwareAutonomousMedicineCreative

All news

10783
NYIT builds an LLM assistant for its applied analysis course in MATLAB
LLM·arXiv cs.AI

NYIT builds an LLM assistant for its applied analysis course in MATLAB

Jul 11, 2026·2 min
Infinity-Parser2 beats DeepSeek-OCR-2 at document parsing and releases a 5-million-sample dataset
Creative

Infinity-Parser2 beats DeepSeek-OCR-2 at document parsing and releases a 5-million-sample dataset

Jul 11, 2026·3 min
Hallucination Self-Play: a small language model beats larger LLMs at detecting hallucinations
LLM

Hallucination Self-Play: a small language model beats larger LLMs at detecting hallucinations

Jul 11, 2026·2 min
Gemini 2.5 Flash Replaces Human Raters in Voice Agent Testing
LLM

Gemini 2.5 Flash Replaces Human Raters in Voice Agent Testing

Jul 11, 2026·3 min
Idiobionics: Researchers Warn of Hacking Risks in Smart Bionic Prostheses
Robotics·arXiv cs.AI

Idiobionics: Researchers Warn of Hacking Risks in Smart Bionic Prostheses

Researchers have coined the term "idiobionics" to study privacy threats in smart AI prostheses: the same sensors that make them effective open a path to stealing…

Jul 11, 2026·2 min
TACO: A Method That Removes Credit "Contamination" in RL Training of Language Models
Creative·arXiv cs.CL

TACO: A Method That Removes Credit "Contamination" in RL Training of Language Models

A new method, TACO, tackles a systemic problem in LLM RL training: low-probability "tail" tokens get the same credit as correct ones, gradually degrading the…

Jul 11, 2026·2 min
New MiniLM-based method better spots queries outside a chatbot's knowledge scope
Creative·arXiv cs.CL

New MiniLM-based method better spots queries outside a chatbot's knowledge scope

Researchers propose multi-cluster boundary learning on MiniLM embeddings to detect out-of-scope queries in chatbots. It beat all baselines on three datasets.

Jul 11, 2026·2 min
Debiasing NLP Models Creates New Stereotypes for Other Demographic Groups
Research·arXiv cs.CL

Debiasing NLP Models Creates New Stereotypes for Other Demographic Groups

An arXiv study finds that preprocessing methods reduce bias for targeted groups but inadvertently amplify it for others, and standard benchmarks do not catch it.

Jul 11, 2026·3 min
Top themes
01Researchers propose "alignment plausibility" as a safety standard for LLMs in healthcareJul 11, 2026·3 min02A review of 18 medical LLMs: specialist models diagnose more accurately than general-purpose onesJul 11, 2026·2 min03EspanStereo: A New Dataset Exposes LLM Stereotypes Across Spanish-Speaking CulturesJul 11, 2026·2 min04DeepSearch-Evolve: a 9B-parameter web agent trains without a teacher and scores 61% on GAIAJul 11, 2026·2 min05AI and Disinformation: How LLM Agents Erode TrustJul 11, 2026·2 min
LLMs prove theorems but don't discover mathematics: researchers outline the next step
LLM·arXiv cs.CL

LLMs prove theorems but don't discover mathematics: researchers outline the next step

Jul 11, 2026·2 min
New AI Tool Combines GTAP and APSIM to Assess Shocks in Agricultural Supply Chains
Research·arXiv cs.AI

New AI Tool Combines GTAP and APSIM to Assess Shocks in Agricultural Supply Chains

Jul 11, 2026·3 min
LSTM beats Random Forest and Naive Bayes at tweet sentiment analysis with 0.92 ROC-AUC
Research·arXiv cs.CL

LSTM beats Random Forest and Naive Bayes at tweet sentiment analysis with 0.92 ROC-AUC

Jul 11, 2026·3 min
Context Graph: Researchers Teach Enterprise AI Agents to Work Without Being Asked
LLM·arXiv cs.AI

Context Graph: Researchers Teach Enterprise AI Agents to Work Without Being Asked

Jul 11, 2026·2 min
LiST: A New Method Delivers Accuracy, Robustness and Calibration in Neural Networks at Once
Creative·arXiv cs.LG

LiST: A New Method Delivers Accuracy, Robustness and Calibration in Neural Networks at Once

The LiST method automatically tunes the Lipschitz constant so a neural network stays accurate, robust to attacks and calibrated, with no manual tuning.

Jul 11, 2026·2 min
Hol-PCFG cuts parsing parameters by 99.94% and tops results in six languages
Creative·arXiv cs.CL

Hol-PCFG cuts parsing parameters by 99.94% and tops results in six languages

The new Hol-PCFG model swaps opaque neural networks for holographic embeddings and posts the best unsupervised parsing results across six languages.

Jul 11, 2026·2 min
MetaNCA teaches neural networks to self-organize without backpropagation
Research·arXiv cs.LG

MetaNCA teaches neural networks to self-organize without backpropagation

The new MetaNCA framework generates weights for neural networks of different architectures using local cellular-automata rules, with no classical backpropagation.

Jul 11, 2026·2 min
Jet-Long extends language model context windows to 128K without fine-tuning
Hardware·arXiv cs.LG

Jet-Long extends language model context windows to 128K without fine-tuning

Jul 11, 2026·2 min
Eternis-Forecaster 8B: Activation Probes Beat Chains of Thought and Save Up to 47% of Tokens
LLM·arXiv cs.CL

Eternis-Forecaster 8B: Activation Probes Beat Chains of Thought and Save Up to 47% of Tokens

Jul 11, 2026·2 min
AI for documents: Datalab Lift vs. the competition
Creative·MarkTechPost

AI for documents: Datalab Lift vs. the competition

Jul 11, 2026·2 min
Anthropic launches Reflect, a dashboard for analyzing how you use Claude
LLM·Anthropic Blog

Anthropic launches Reflect, a dashboard for analyzing how you use Claude

Jul 11, 2026·3 min
Meta releases Muse Spark 1.1, a multimodal agent with a 1-million-token context window
Creative·Meta AI Blog

Meta releases Muse Spark 1.1, a multimodal agent with a 1-million-token context window

Jul 11, 2026·2 min
Managing AI Prompts in the Enterprise: Mistral Adds Version Control to Studio
LLM

Managing AI Prompts in the Enterprise: Mistral Adds Version Control to Studio

Jul 11, 2026·2 min
Amazon closes Mechanical Turk to new customers, ending an era of crowdsourced AI
Business

Amazon closes Mechanical Turk to new customers, ending an era of crowdsourced AI

Jul 11, 2026·2 min
AI in education: what Forge Prep and Alpha School are selling
Hardware

AI in education: what Forge Prep and Alpha School are selling

Jul 11, 2026·2 min
Station F opens new intake for F/ai, its AI accelerator for European startups
Startups·TechCrunch

Station F opens new intake for F/ai, its AI accelerator for European startups

Paris startup campus Station F, founded by Xavier Niel, has announced a new intake for F/ai, a specialized accelerator for Europe's most promising AI companies.

Jul 11, 2026·2 min
Why the rise of open AI doesn't yet threaten Anthropic: a lifecycle analysis
LLM·TechCrunch

Why the rise of open AI doesn't yet threaten Anthropic: a lifecycle analysis

TechCrunch explains the paradox: open models are breaking records, yet Anthropic keeps growing, because each player occupies a different phase of the same lifecycle.

Jul 11, 2026·2 min