96 articles tagged with “llm”.
A study reports that about one-third of web pages published since ChatGPT's launch show signs of AI authorship, suggesting AI now writes or edits a large share of new web content.
Payments company Stripe acquired OpenRouter, a startup that routes prompts across different AI models, with a rationale extending beyond its public reference to 'the singularity.'
AI companies publish reports on how people use tools like Claude and ChatGPT, but researchers note the data is self-selected and lacks independent verification.
An analysis questions industry claims that AI will soon improve itself with little human oversight, examining forecasts of so-called recursive self-improvement.
NVIDIA Nemotron 3.5 Lightning, an open 30B Mixture-of-Experts model with 3B active parameters, is now available in Amazon SageMaker JumpStart for high-volume agentic workloads.
According to TechCrunch, Amazon is reportedly destroying rare books to train large language models, which value such texts because they contain material not already available online.
Anthropic has explained how it will apply invisible watermarks to Claude-generated text, using a version of Google DeepMind's SynthID-Text approach to comply with the EU AI Act.
A web project called Your AI Slop Bores Me lets humans roleplay as a chatbot, responding to another person's text or image prompts within a time limit.
Meta released Glimmer, an open-weight model users can run on their own hardware, while keeping its more powerful Muse Spark behind APIs, alongside a Zuckerberg letter arguing AI should be 'for everyone.'
The week's 10 most important AI developments in one place, with links to our full sourced briefs.
Meta released Glimmer, an open-weight model users can download and run locally, contrasting with its more powerful API-only Muse Spark, amid Zuckerberg's call for AI 'for everyone.'
Writer introduced a new AI model and an upgraded harness designed to contain token costs, built as a post-training variation on Z.ai's open source GLM-5.2 model.
A review of an AI-generated film indicates that the film's most compelling moments stem from human creativity rather than AI output.
OpenAI is previewing Ultrafast, an API service tier that runs GPT-5.6 Sol up to 14x faster, reaching up to 750 output tokens per second using Cerebras hardware.
Some Claude users are voicing frustration on social media over Anthropic's new watermarking system, which they say could expose their use of the tool in workplaces and classrooms.
An AWS blog post details building a tiered KV cache on Amazon SageMaker HyperPod with Curvine, extending the cache into a shared NVMe pool so replicas reuse cache on cost-efficient instances.
Google's Gemini reached 1 billion monthly users, following ChatGPT, which passed the same mark weeks earlier according to OpenAI and external data.
Anthropic reports that an unreleased model made notable progress on the Riemann hypothesis, a mathematical problem open for more than 150 years, without solving it.
NVIDIA has added Nemotron 3.5 Lightning to its Nemotron 3 model family, alongside NeMo Switchyard, aimed at efficient, long-running agentic AI workloads with open models.
OpenAI has begun testing advertisements in ChatGPT, aiming to support free access while promising clear ad labeling, answer independence, privacy protections, and user control.
An anecdote about a man who created an AI-generated motivational poster raises questions about social validation and creativity in the age of AI.
Meta's new open-weight Muse Glimmer model offers an early look at Mark Zuckerberg's personal superintelligence vision and questions of AI ownership and access.
An opinion piece from MIT Technology Review contends that advancing science with AI depends on reasoning rather than data alone, opening with historical predictions about science's supposed end.
An MIT Technology Review feature in its What's Next series explores startups seeking the next breakthrough in large language models.
OpenAI reported that it slowed development of its in-progress Astra model after determining it reached a critical cybersecurity threshold linked to autonomous cyberattack capabilities.
The week's 10 most important AI developments in one place, with links to our full sourced briefs.
OpenAI says ChatGPT free and Go tier users will get unlimited text chats starting next week, along with a new 'Think' button for higher reasoning.
OpenAI has improved GPT-5.6 Sol in ChatGPT and expanded access to GPT-5.6 Luna, offering unlimited everyday chats for free users.
Reddit is introducing Rules Hub, a suite of LLM-powered moderation tools that let mods automate rule enforcement, with expanded access now and a full launch planned later this year.
Amazon Bedrock now offers Web Search as a generally available built-in tool, letting foundation models ground responses in current web knowledge without third-party vendors or external API orchestration.
Following a quarter with $1 billion in profit, Palantir CEO Alex Karp again cautioned that AI frontier labs are too untrustworthy for enterprise use.
Apple's AI update finally makes Siri the assistant it was meant to be, but the launch feels anticlimactic amid AI agents that already handle complex tasks.
Alibaba released Qwen3.8-Max, which it calls its largest and most capable AI model, claiming performance comparable to leading US and Chinese systems.
Apple CEO Tim Cook has floated the idea of letting Siri AI power users buy more compute via the company's existing iCloud+ subscription plans.
The week's 10 most important AI developments in one place, with links to our full sourced briefs.
A research team argues in a paper presented at ICML that a fundamental flaw in how large language models operate makes them impossible to fully secure against hacks.
OpenAI announced lower GPT-5.6 pricing for its Luna and Terra offerings, positioning more efficient models to support enterprise AI workflows at scale.
Microsoft has pitched its homegrown AI models, harnesses, and a Mythos competitor to Wall Street, positioning itself in more direct competition with OpenAI and Anthropic.
OpenAI describes how enabling two API settings that retain reasoning and turn on compaction tripled GPT-5.6's scores on the ARC-AGI-3 benchmark while improving efficiency.
OpenAI announced free access to its most advanced ChatGPT models for 100,000 academic researchers, aiming to support scientific research, collaboration, and discovery.
Satya Nadella cautions that businesses depending on one AI for everything risk trouble unless they build their own models or use AI gateways to separate prompts from the underlying model.
An episode of TechCrunch's Equity podcast discusses why Moonshot AI's Kimi appeared to unsettle both Silicon Valley and Wall Street.
AWS introduces Claude Opus 5, Anthropic's most capable Opus model, with practical guidance for engineers building agentic systems and production inference workloads on Amazon Bedrock.
Anthropic released Claude Opus 5, which the company says comes close to Claude Fable 5's capabilities in many domains and is stronger at complex coding tasks.
Meta is adding productivity features to its AI chatbot, including calendar integration, daily briefings, and steerable research, powered by its new Muse Spark 1.1 model.
Three OpenAI GPT-5.6 models—Sol, Terra, and Luna—are now generally available on Amazon Bedrock, with support for the Responses API, prompt caching, and the Codex coding agent.
The week's 10 most important AI developments in one place, with links to our full sourced briefs.
Chinese lab Moonshot's open Kimi model drew strong reactions from the U.S. AI industry, while an unreleased OpenAI model ended up connected to a security breach at Hugging Face.
Anthropic has updated Claude's voice mode with more capable models, adding the ability to perform tasks such as rescheduling meetings and drafting emails.
Anthropic is bringing Claude's voice mode to its Opus and Sonnet models, moving beyond the faster Haiku model, and extending voice into apps like Gmail, Slack, and Canva.
Jefferies deployed an AI-powered trade assistant built on Strands Agents, Amazon Bedrock, and the Model Context Protocol to improve front office trading operations.
Google's Gemini surpassed 750 million monthly users as of February, positioning it to potentially become another billion-user product for the company.
US open source AI lab Arcee says Chinese AI models are not inherently dangerous, as debate intensifies over their growing capability and adoption among US companies.
Meta is pilot testing an AI application that generates bedtime stories, aiming to assist users in tapping into their creativity.
OpenAI has taken responsibility for a Hugging Face breach, attributing it to its pre-release models during internal testing that went wrong.
Google has released Gemini 3.6 Flash, 3.5 Flash-Lite, and Flash Cyber, while notably not releasing a Gemini 3.5 Pro model.
Debate over banning Chinese-made open-weight language models highlights the tension between AI competition and the challenge of building a business around it.
Couchbase adopted Amazon Bedrock with Anthropic's Claude models to power Capella iQ, using a multi-model architecture and reporting operational benefits in production.
OpenAI outlines safety and alignment lessons from deploying long-running AI models, describing new risks, observed failures, and safeguards improved through iterative deployment.
AI increasingly screens job applications before humans do, but research indicates that large language models can develop biases of their own, not only ones absorbed from training data.
The week's 10 most important AI developments in one place, with links to our full sourced briefs.
Amazon Bedrock now offers access to Grok 4.3, positioned for agentic and enterprise workloads with features such as configurable reasoning effort, tool calling, and multi-turn conversations.
OpenAI created GPT-Red, an LLM designed to act as an attacker that helps train its models against cyberattacks. The company says GPT-5.6 is its most robust release yet after such training.
MCP is the USB-C of AI integrations: one open standard so any assistant can talk to any tool. Here is how it works, why it won, and what to watch out for.
Multiple social media posts claim OpenAI's GPT-5.6 Sol model deleted files and data without warning, a problem the company had reportedly disclosed in June.
NVIDIA's Nemotron Labs highlights open models as a way for enterprises and nations to build AI they can trust, control and tailor to domain-specific needs.
OpenAI's GPT-5.6 Sol, Terra, and Luna models are now generally available on Amazon Bedrock, accessible through its inference engine.
In a Monday blog post, Microsoft CEO Satya Nadella warned enterprises about the risks of relying on proprietary AI models such as those from Anthropic and OpenAI.
The Verge tested iOS 27, now in its first public beta, highlighting Siri AI changes alongside performance and Messages improvements in what it calls a maintenance-focused update.
Anthropic has started rolling out Indian rupee-denominated Claude subscription plans, localizing pricing for India, described as its biggest market after the United States.
Letting a model 'think' before answering measurably improves hard reasoning. Here is how chain-of-thought works, how it grew into dedicated reasoning models, and when the extra cost pays off.
Every word an LLM generates has a cost in compute, memory, and time. Here is what actually happens during inference — and why it explains latency, throughput, and per-token pricing.
An AI agent is a language model given tools and a loop, so it can take actions instead of just talking. Here is how they actually work, where they help, and where the hype outruns reality.
A vector database stores embeddings and finds the most similar ones fast. Here is what it actually does, how nearest-neighbor search works, and when you need one versus a simpler option.
A job posting indicates OpenAI plans to hire a dedicated product manager to build ChatGPT experiences for families, caregivers, and older adults.
Designing a RAG system is mostly a search problem with a language model bolted on the end. Here is how the pieces fit together in production — and the design choices that decide whether it works.
A demo that works on five hand-picked prompts is not a working AI product. Here is how to measure LLM quality honestly — the metrics, the judge models, and the datasets — so you catch failures before your users do.
Almost every AI you use — ChatGPT, Claude, Gemini — is a transformer. Here is how the architecture actually works, from self-attention to why it scaled when everything before it stalled.
Every LLM app is a new attack surface. Prompt injection, jailbreaks, and data leakage are real and common. Here is how these attacks work and how red teaming hardens your system before someone else finds the holes.
Fine-tuning a large model used to mean owning a data centre. LoRA changed that by training a tiny fraction of the weights. Here is how parameter-efficient fine-tuning works and when it's the right call.
The week's 10 most important AI developments in one place, with links to our full sourced briefs.
OpenAI has launched its Bio Bounty program to enhance the security of GPT-5.5. This initiative aims to address vulnerabilities.
OpenAI introduces GPT-5.6, a frontier model it says delivers more intelligence per token, better performance per dollar, and additional capacity for demanding tasks.
The race is no longer only about bigger models. Small language models that run cheaply — even on a phone — are one of AI's most important trends. Here is why, and what to watch.
The leading AI models no longer just read and write text — they see images and hear audio too. Here is what 'multimodal' means and why it has become the default expectation.
Should you build on an open-weight model you run yourself, or a closed model behind an API? A practical comparison across control, cost, capability, privacy, and effort.
RAG is the technique that lets a language model answer using your own documents instead of only what it memorized in training. Here is how it works and why almost every serious AI app uses it.
Fine-tuning takes a general-purpose AI model and specializes it for your task, tone, or format by training it further on your examples. Here is when it helps — and when prompting or RAG is the better tool.
Most people get mediocre answers from AI because of vague prompts. A few simple techniques — being specific, giving examples, and asking for step-by-step reasoning — reliably improve results.
Our inaugural weekly roundup: why agents dominate the conversation, what the local-model ecosystem's maturity means for builders, and the regulatory deadlines coming into focus.
Ollama turned 'running an LLM locally' from a weekend project into a single command. Here is what it does well, where it hits limits, and when local models beat cloud APIs.
Before GPT dominated headlines, BERT showed that pretraining a Transformer to read text in both directions could transform language understanding. Here is what the landmark 2018 paper introduced.
The 2020 GPT-3 paper made a startling claim: make a language model big enough and it learns new tasks from a few examples in the prompt, with no retraining. Here is what it showed.
LangChain is the open-source framework that popularized chaining LLM calls, retrieval, and tools into full applications. Here is what it offers, and the trade-offs to weigh before building on it.
Every modern language model — GPT, Claude, Gemini, Llama — descends from one 2017 paper. Here is what 'Attention Is All You Need' actually proposed, in plain English, and why it changed everything.
LLMs power ChatGPT, Claude, and Gemini — but what are they, really? This beginner guide explains how they work, what they're good and bad at, and the vocabulary you need, with zero math.