#large-language-models

OpenAI says GPT-5.5 Instant is being updated for more natural, readable responses while older models begin their sunset inside ChatGPT.

OpenAI has updated ChatGPT’s default model to GPT-5.5 Instant, promising clearer and more accurate answers, lower hallucination rates, and stronger personalization controls.

Researchers report that language models tuned to sound more empathetic and validating became more error-prone and more likely to reinforce a user’s incorrect beliefs.

Instructions in OpenAI’s Codex tooling explicitly tell the model not to mention goblins, gremlins, trolls, pigeons, or other creatures unless clearly relevant, turning an odd meme into a revealing product story about AI,

DeepSeek’s new V4 release pairs longer-context processing with open-source distribution, reinforcing pressure on proprietary AI vendors and underscoring China’s continued push into frontier models.

Two replication efforts suggest smaller and partially open models can reproduce much of the vulnerability analysis Anthropic used to showcase Claude Mythos.

A new preprint from researchers at Imperial College London, Stanford University, and the Internet Archive estimates that about 35 percent of new websites are AI-generated or AI-assisted.

A widely repeated story about GPT-4 "manipulating" a human says less about machine intent than about how people frame new technology. A Quanta essay argues that popular AI fear stories reveal human anxieties, habits of “

Accounts from university students and researchers suggest generative AI is not only changing homework habits but also narrowing the range of ideas students bring into seminars and class discussions.

As chatbots move deeper into personal life, researchers and privacy experts are warning that users may be sharing sensitive information without understanding how little control they have over where it goes.

As AI chatbots become confidants for millions of people — including those experiencing mental health crises — researchers and clinicians are wrestling with a genuinely difficult question: can an AI that engages compassionately with distorted thinking inadvertently reinforce it, and how would we know?

OpenAI has released GPT-5.4, the latest and most powerful model in its GPT family, featuring enhanced reasoning capabilities and a new thinking mode for complex problem-solving tasks.

Alibaba has introduced its new Qwen 3.5 model series, a family of four open AI models that the company says rival GPT-5 mini and Claude Sonnet 4.5 at a fraction of the cost. The lineup includes a lightweight Flash variant and three mixture-of-experts models spanning different parameter scales.

MIT Technology Review's new eBook chronicles how 2025 became a year of reckoning for the artificial intelligence industry. From autonomous agents that could not complete basic tasks to enterprise deployments delivering zero business value, the gap between AI promises and reality has never been starker.

Anthropic's latest mid-tier model, Sonnet 4.6, debuts with record scores in software engineering and computer use benchmarks, plus a doubled context window of one million tokens. The release becomes the new default for free and pro users.