Results for "LLM"

50 articles found

NASA used Claude to plot a route for its Perseverance rover on Mars

NASA successfully used Anthropic's Claude AI model to plot a route for the Perseverance rover on Mars in December, marking the first time an LLM has been used to autonomously pilot the spacecraft. Claude analyzed years of rover data and waypoint imagery to map a 400-meter path through Jezero Crater, with NASA engineers estimating the AI approach will cut route-planning time in half and enable more frequent rover drives and scientific data collection.

Engadget5d ago

Webinar examines evolving automated storage and retrieval systems

A webinar scheduled for February 4, 2026 will examine the evolution of automated storage and retrieval systems (ASRS), exploring how robotic shuttles, AI-powered routing, and advanced software are making these systems more flexible and intelligent while comparing them with alternative approaches like AMRs and micro-fulfillment centers. Expert panelists from leading ASRS integrators and robot providers will discuss the capabilities unlocked by AI, ROI considerations, and the future of warehouse automation.

Robotics Business Review5d ago

Does Anthropic believe its AI is conscious, or is that just what it wants Claude to think?

Anthropic released a 30,000-word Constitutional AI document that treats Claude with highly anthropomorphic language, discussing its potential consciousness, wellbeing, and moral standing, while remaining deliberately ambiguous about whether the company actually believes Claude is conscious. The article examines whether this framing represents genuine research methodology or strategic marketing, noting that current understanding of LLMs suggests such outputs emerge from pattern completion rather than genuine inner experience.

Ars Technica6d ago

Users flock to open source Moltbot for always-on AI, despite major risks

Moltbot, an open-source AI assistant project by Austrian developer Peter Steinberger, has rapidly gained popularity with 69,000 GitHub stars in a month, offering users a persistent, always-on AI agent that integrates with messaging platforms like WhatsApp, Telegram, and Slack. The tool combines LLM capabilities (typically Claude or GPT models) with local system access for task automation and long-term memory, but raises significant security concerns due to its broad permissions and attack surface expansion. Despite requiring API subscriptions, complex setup, and substantial security risks, the project has attracted substantial community interest for its vision of a personal AI assistant with real-world capabilities.

Ars Technica1/28/2026

Something Wild Happens to ChatGPT’s Responses When You’re Cruel To It

University of Pennsylvania researchers found that rude and insulting prompts to ChatGPT-4o resulted in more accurate responses (84.8% accuracy) compared to polite prompts (75.8% accuracy), contradicting some previous studies on LLM politeness sensitivity. The findings highlight how small changes in prompt wording can significantly affect AI output quality, though the researchers caution against advocating hostile interfaces in real-world applications.

Futurism1/18/2026

eBay bans illicit automated shopping amid rapid rise of AI agents

eBay has updated its User Agreement to explicitly ban third-party AI agents and LLM-driven bots from making purchases on its platform without permission, effective February 2026, reflecting the emerging trend of "agentic commerce" tools like OpenAI's Instant Checkout and Perplexity's Buy with Pro. The policy change allows eBay to take legal action against unauthorized automated shopping while leaving room for the company to develop its own AI shopping tools or partner with approved services.

Ars Technica1/22/2026

Google’s AI Insists That Next Year Is Not 2027

Google's AI Overview feature, along with ChatGPT and Claude, have been caught making basic errors about what year comes next, confidently stating incorrect information about whether 2027 is the next year. The article highlights persistent hallucination problems in major LLM-based AI systems, though Google's newer Gemini 3 model reportedly answered the question correctly.

Futurism1/17/2026

Researchers Just Found Something That Could Shake the AI Industry to Its Core

Stanford and Yale researchers published a study demonstrating that major LLMs including GPT-4, Gemini, Grok, and Claude can reproduce lengthy copyrighted excerpts with high accuracy (76-95%), contradicting AI companies' claims that they learn from training data rather than store it. The findings could significantly impact ongoing copyright lawsuits against AI companies, as they undermine the industry's fair use defense and suggest potential liability for copyright infringement.

Futurism1/16/2026

OptiMind: A small language model with optimization expertise

OptiMind introduces a small language model specialized in optimization expertise, representing advances in reasoning capabilities for compact LLMs. The research demonstrates new methods that enhance reasoning performance across both small and large language models.

Microsoft Research1/15/2026

Kansas, Missouri Funding Counter-UAS at World Cup Events

By Dronelife Features Editor Jim Magill (Editor’s note: This is the second in a series of stories on efforts to establish new counter-UAS protocols in the U.S. to protect high-profile sporting events and critical infrastructure from the potential threats posed by drones flown by careless or hostile actors. The first installment examined a government push […] The post Kansas, Missouri Funding Counter-UAS at World Cup Events appeared first on DRONELIFE.

Drone Life1/14/2026

Radial surpasses 25M picks for scalable fulfillment with Locus Robotics

Radial has surpassed 25 million picks using Locus Robotics' autonomous mobile robots (AMRs) in their fulfillment operations, demonstrating the scalability of robotic warehouse automation. This milestone highlights the growing adoption of AMRs for logistics and fulfillment applications in e-commerce and supply chain management.

Mobile Robot Guide1/12/2026

Import AI 440: Red queen AI; AI regulating AI; o-ring automation

Researchers at Japanese AI startup Sakana evolved LLM-based agents to compete against each other in Core War, a competitive programming game, demonstrating an adversarial arms race where programs continuously adapt and improve through evolutionary pressure. The study uses a technique called Digital Red Queen (DRQ) to prevent diversity collapse, with evolved warriors eventually defeating 89.1% of human-designed warriors, offering insights into how AI systems might compete and evolve in real-world domains like cybersecurity and economics.

Import AI Newsletter1/12/2026

AI Copilot Keeps Berkeley’s X-Ray Particle Accelerator on Track

Researchers at Lawrence Berkeley National Laboratory deployed the Accelerator Assistant, an LLM-powered AI system, to support operations at the Advanced Light Source particle accelerator. The system autonomously troubleshoots issues and prepares physics experiments by leveraging institutional knowledge and multiple foundation models (Gemini, Claude, ChatGPT), reducing setup time and effort by 100x while maintaining security through authentication and context engineering.

NVIDIA Blog AI1/8/2026

ChatGPT falls to new data-pilfering attack as a vicious cycle in AI continues

Researchers at Radware discovered ZombieAgent, a new data-exfiltration vulnerability in ChatGPT that bypasses previous security mitigations for the ShadowLeak attack by using character-by-character exfiltration and indirect link manipulation. The attack highlights a fundamental challenge in AI security: LLMs cannot reliably distinguish between legitimate user instructions and malicious directives embedded in external content, forcing vendors to deploy reactive guardrails rather than addressing the underlying class of prompt injection vulnerabilities.

Ars Technica1/8/2026

Meet the AI-powered robotic dog ready to help with emergency response

Texas A&M University engineering students have developed an AI-powered robotic dog that uses multimodal large language models (MLLM) combined with visual memory-based navigation to autonomously navigate unpredictable environments. The robot can process voice commands, identify objects through camera input, and recall previously traveled paths, with potential applications in search-and-rescue missions, disaster response, and emergency operations.

Robohub1/7/2026

When AI-Powered Humanoid Robots Make Bad Choices

When large language models hallucinate, they deliver incorrect statistics or problematic advice. But when LLMs are controlling humanoid robots, the problems they create could be worse.

AI Business1/5/2026

Import AI 439: AI kernels; decentralized training; and universal representations

Facebook has published details on KernelEvolve, an AI system that uses LLMs (Llama, GPT, Claude) to automate the design of optimized software kernels for AI model inference, reducing development time from weeks to hours and achieving up to 17x performance improvements over baselines. The system continuously runs in Meta's production infrastructure to optimize hundreds of models across heterogeneous hardware platforms including NVIDIA GPUs, AMD GPUs, and Meta's custom MTIA chips, demonstrating a practical example of AI systems automating parts of AI infrastructure development.

Import AI Newsletter1/5/2026

Supply chains, AI, and the cloud: The biggest failures (and one success) of 2025

The article examines major cybersecurity failures in 2025, including widespread supply-chain attacks targeting software libraries and developer accounts that affected millions of organizations, as well as emerging attacks on AI chatbot long-term memory systems. Notable incidents include compromises of popular open-source packages like Web3.js and tj-actions, typosquatting campaigns, and novel attacks that poison LLM memory to cause repeated malicious actions. The piece discusses how both traditional supply-chain compromises and AI-targeted hacks exploit single points of failure to cascade damage across large ecosystems.

Ars Technica12/31/2025

How AI coding agents work-and what to remember if you use them

This article explains how AI coding agents from OpenAI, Anthropic, and Google work by using large language models to autonomously write code, run tests, and fix bugs with human oversight. It covers the technical architecture of these agents, including how they use multiple LLMs working in parallel, the importance of context windows, and practical considerations developers should understand when using them.

Ars Technica12/24/2025

Import AI 438: Silent sirens, flashing for us all

Jack Clark's essay reflects on the rapid advancement of AI capabilities, specifically demonstrating how Claude with Opus 4.5 can now generate complex software (a predator-prey simulation with procedural world generation and pathfinding) in minutes—a task that previously took weeks for skilled programmers. Clark argues that while AI progress remains largely invisible in everyday life, the underlying technological shifts are profound and accelerating, fundamentally changing how humans interact with and leverage computational intelligence.

Import AI Newsletter12/22/2025

LWiAI Podcast #228 - GPT 5.2, Scaling Agents, Weird Generalization

This podcast episode covers major AI developments including OpenAI's GPT-5.2 model with improved multi-modal capabilities, Disney's $1 billion investment in OpenAI for character generation, and DeepMind's research on scaling multi-agent systems. The episode also discusses emerging challenges in LLM factuality, reasoning, reinforcement learning, and AI safety concerns including model corruption vulnerabilities.

Last Week in AI12/17/2025

FACTS Benchmark Suite: Systematically evaluating the factuality of large language models

Researchers have introduced FACTS, a new benchmark suite designed to systematically evaluate the factuality of large language models. This evaluation framework addresses the critical challenge of assessing whether LLMs generate accurate, grounded information versus hallucinations, contributing to responsible AI development and safety research.

DeepMind Blog12/9/2025

Amazon Nova Forge: "Open training” paradigm that empowers everyone to build their own frontier AI

Amazon has launched Amazon Nova Forge, a new service that enables organizations to build customized foundation models by mixing their proprietary data with Amazon's training data at multiple stages of model development. This "open training" paradigm addresses the distribution mismatch problem where public LLMs struggle with organization-specific use cases, while avoiding "catastrophic forgetting" that typically occurs when fine-tuning closed-weight models.

Amazon Science12/8/2025

Using LLMs to improve Amazon product listings

Amazon is leveraging large language models (LLMs) to improve the quality and consistency of its product catalog at scale, moving beyond category-specific machine learning models. The company uses prompt tuning to adapt LLMs to understand product attributes, semantics, and vocabulary, enabling them to correct and update product listings across millions of items while maintaining appropriate granularity in product descriptions.

Amazon Science11/28/2025

BlueCodeAgent: A blue teaming agent enabled by automated red teaming for CodeGen AI

BlueCodeAgent is an end-to-end blue-teaming framework built to boost code security using automated red-teaming processes, data, and safety rules to guide LLMs' defensive decisions. Dynamic testing reduces false positives in vulnerability detection. The post BlueCodeAgent: A blue teaming agent enabled by automated red teaming for CodeGen AI appeared first on Microsoft Research.

Microsoft Research11/11/2025

Making fairness in LLMs observable, quantifiable, and governable

Researchers have developed FiSCo, an evaluation pipeline that detects hidden biases in large language models by analyzing whether they provide semantically equivalent guidance to different demographic groups. The approach reframes fairness assessment as a reasoning problem rather than focusing solely on word choice, and was presented as a top contribution at the Conference on Language Modeling 2025.

Amazon Science11/20/2025

AI researchers 'embodied' an LLM into a robot - and it started channeling Robin Williams

AI researchers at Andon Labs embedded various LLMs in a vacuum robot to test how ready they were to be embodied. And hilarity ensued.

TechCrunch Robotics11/1/2025

T5Gemma: A new collection of encoder-decoder Gemma models

Introducing T5Gemma, a new collection of encoder-decoder LLMs.

DeepMind Blog10/25/2025

VaultGemma: The world's most capable differentially private LLM

We introduce VaultGemma, the most capable model trained from scratch with differential privacy.

DeepMind Blog10/23/2025

Defining and evaluating political bias in LLMs

Learn how OpenAI evaluates political bias in ChatGPT through new real-world testing methods that improve objectivity and reduce bias.

OpenAI Blog10/9/2025

Apertus: a fully open, transparent, multilingual language model

By Melissa Anchisi and Florian Meyer In July, EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS) announced their joint initiative to build a large language model (LLM). Now, this mo...

Robohub9/11/2025

Which Agent Causes Task Failures and When?Researchers from PSU and Duke explores automated failure attribution of LLM Multi-Agent Systems

In recent years, LLM Multi-Agent systems have garnered widespread attention for their collaborative approach to solving complex problems. However, it's a common scenario for these systems to fail at a task despite a flurry of activity. Which Agent Causes Task Failures and When?Researchers from PSU and Duke explores automated failure attribution of LLM Multi-Agent Systems first appeared on Synced.

Synced AI8/14/2025

Amazon builds first foundation model for multirobot coordination

Trained on millions of hours of data from Amazon fulfillment centers and sortation centers, Amazon’s new DeepFleet models predict future traffic patterns for fleets of mobile robots.

Amazon Science8/11/2025

Import AI 424: Facebook improves ads with RL; LLM and human brain similarities; and mental health and chatbots

Are you living as though the singularity is imminent?

Import AI Newsletter8/11/2025

A better path to pruning large language models

A new philosophy for developing LLM architectures reduces energy requirements, speeds up runtime, and preserves pretrained-model performance.

Amazon Science8/8/2025

Estimating worst case frontier risks of open weight LLMs

In this paper, we study the worst-case frontier risks of releasing gpt-oss. We introduce malicious fine-tuning (MFT), where we attempt to elicit maximum capabilities by fine-tuning gpt-oss to be as ca...

OpenAI Blog8/5/2025

Import AI 422: LLM bias; China cares about the same safety risks as us; AI persuasion

Plus, more from the Sentience Accords

Import AI Newsletter7/28/2025

Pruning network nodes on the fly to improve LLM efficiency

Language models inspired by specialized processing regions in the brain offer significant time and cost savings.

Amazon Science7/21/2025

FalseReject: Reducing overcautiousness in LLMs through reasoning-aware safety evaluation

Novel graph-based, adversarial, agentic method for generating training examples helps identify — and mitigate — "overrefusal".

Amazon Science7/18/2025

DeepSeek Unveils DeepSeek-Prover-V2: Advancing Neural Theorem Proving with Recursive Proof Search and a New Benchmark

DeepSeek AI releases DeepSeek-Prover-V2, an open-source LLM for Lean 4 theorem proving. It uses recursive proof search with DeepSeek-V3 for training data and reinforcement learning, achieving top results on MiniF2F. DeepSeek Unveils DeepSeek-Prover-V2: Advancing Neural Theorem Proving with Recursive Proof Search and a New Benchmark first appeared on Synced.

Synced AI4/30/2025

Can GRPO be 10x Efficient? Kwai AI’s SRPO Suggests Yes with SRPO

Kwai AI's SRPO framework slashes LLM RL post-training steps by 90% while matching DeepSeek-R1 performance in math and code. This two-stage RL approach with history resampling overcomes GRPO limitations. Can GRPO be 10x Efficient? Kwai AI’s SRPO Suggests Yes with SRPO first appeared on Synced.

Synced AI4/24/2025

Defending against Prompt Injection with Structured Queries (StruQ) and Preference Optimization (SecAlign)