#large-language-models-llm — Public Fediverse posts

UC San Diego: Governments May Shape What AI Chatbots Say by Shaping the Web They Learn From. “Ask an AI model the same political question in two different languages, and you may get two very different responses. A new study in Nature suggests one reason why: governments can indirectly influence large language models (LLMs) by shaping the online media environment, and thus the text those systems […]

Ars Technica: The newest AI boom pitch: Host a mini data center at your home. “Data centers may be coming to your neighborhood as side installations associated with new homes—and in exchange would offer subsidized electricity and Internet access along with backup batteries to homeowners. The company behind the plan has already begun pilot testing in preparation for a 100-home trial run this […]

Ars Technica: Chrome’s 4GB AI model isn’t new, but you’re not wrong for being confused. “Some desktop Chrome users have also noted that the browser appears to suddenly want more storage space for AI. This is true—Chrome does download a 4GB AI model for on-device processing. It’s been doing that for years, though. Google hasn’t actually changed anything about Chrome’s on-device AI, […]

Associated Press: The rapid embrace of AI in China, its biggest testing ground, may shape how AI is used globally. “More than a year after OpenAI’s Chinese rival DeepSeek stunned the world with its advanced AI model, China has become a testing ground for mass use of AI tools. AI models built in the United States still dominate in raw computing firepower, but Chinese people and businesses have […]

The Register: Yet another experiment proves it’s too damn simple to poison large language models. “Unlike search engines that let you judge competing sources, search-backed AI chatbots can turn shaky web material into confident answers. Case in point: A security engineer convinced several bots that he was the reigning world champion of a popular German card game, even though no such […]

TechSpot: Google Chrome has been silently pushing a 4GB AI model to your device without asking. “Google Chrome users who have noticed unusual disk activity or unexplained drops in available storage should look for a folder called ‘OptGuideOnDeviceModel’ inside their Chrome directory. It holds roughly 4GB of weights for Google’s Gemini Nano LLM, downloaded by the browser without user consent.”

Mashable: OpenAI rolls out ChatGPT 5.5 Instant as the new default model for everyone. “According to OpenAI, GPT-5.5 Instant produced 52.5 percent fewer hallucinated claims in internal testing than GPT-5.3 in ‘high stakes’ topics like law, finance, and medicine. In addition, the new model ‘reduced inaccurate claims by 37.3% on especially challenging conversations users had flagged for factual […]

Gizmodo: Talkie Is a ‘Vintage LLM’ Trained on Pre-1930 Data to Help Facilitate ‘Time Travel’. “In the case of Talkie, aka 13B 1930 LM, the cutoff is, as the name suggests, the year 1930. This choice of year might seem arbitrary, but it’s not: as we discussed here back in January, many forms of copyright expire on January 1 of the year that comes 95 years after the copyrighted material […]

Georgia Tech: Transformer Explainer Shows How AI is More Math Than Human. “Georgia Tech researchers are making AI easier to understand through their work on Transformer Explainer. The free, online tool shows non-experts how ChatGPT, Claude, and other large language models (LLMs) process language, improving AI literacy.”

Associated Press: Trump administration vows crackdown on Chinese companies ‘exploiting’ AI models made in U.S.. “The Trump administration is vowing to crack down on foreign tech companies’ exploitation of U.S. artificial intelligence models, singling out China at a time that country is narrowing the gap with the U.S. in the AI race.”

Engadget: LinkedIn’s new Crosscheck feature lets premium subscribers test competing AI models for free. “You can now use LinkedIn to test out some of the latest AI models from OpenAI, Anthropic, Google, Microsoft and other companies without having to worry about token limits or paying for an extra subscription. The professional network is experimenting with a new feature that allows people to […]

CNBC: China’s DeepSeek releases preview of long-awaited V4 model as AI race intensifies. “Chinese artificial intelligence startup DeepSeek on Friday released a preview version of its long-awaited V4 large language model, allowing users to test its new capabilities and features.”

Gizmodo: Some Unknown Group Is Reportedly Using Claude Mythos Without Permission. “In a very cagily-written story from Bloomberg, Anthropic confirmed Tuesday that it has received a report that an unauthorized mystery group is accessing Claude Mythos—the model it says is too dangerous to release.”

MakeUseOf: I stopped using LM Studio once I found this open-source alternative. “The alternative that eventually ended up replacing LM Studio for me is Jan. It’s a desktop application that lets you run LLMs fully offline — much like LM Studio, but not only is it completely free, it’s also open-source, with all the source code being available on GitHub. There are no licensing surprises, no […]

New York Times: We Don’t Really Know How A.I. Works. That’s a Problem.. This link goes to a gift article. “It is tempting, in the face of this opacity, to resort to simplifications: to say that because these systems produce language like us, they are like us, or to say that because these systems are just arrangements of mathematical functions, we can think of them as enormous look-up […]

The Register: Bad teacher bots can leave hidden marks on model students. “New research warns about the dangers of teaching LLMs on the output of other models, showing that undesirable traits can be transmitted ‘subliminally’ from teacher to student, even when they are scrubbed from training data.”

VentureBeat: OpenAI debuts GPT-Rosalind, a new limited access model for life sciences, and broader Codex plugin on Github. “Named after the pioneering chemist Rosalind Franklin, whose work was vital to the discovery of DNA’s structure (and was often overlooked for her male colleagues James Watson and Francis Crick), this new frontier reasoning model is purpose-built to act as a specialized […]

9to5 Mac: OpenAI unveils GPT‑5.4‑Cyber, an AI model for defensive cybersecurity. “OpenAI has announced a new AI model called GPT-5.4-Cyber. Similar to Anthropic’s Claude Mythos, this new ‘cyber-permissive’ variant of its GPT-5.4 is built for defensive cybersecurity and not public use.”

CNBC: AI Age

Meta debuts new AI model, attempting to catch Google, OpenAI after spending billions. Considering the quality of open source LLM’s, I would not be attempting to the chase this particular rainbow. “Dubbed Muse Spark and originally code-named Avocado, the AI model announced Wednesday is the first from the company’s new Muse series developed by Meta Superintelligence Labs, the AI […]

Mashable: Google launches Gemma 4, a new open-source model: How to try it. “Google just released the latest version of its open AI model, Gemma 4, on Thursday. Crucially, Gemma 4 is a fully open-source model licensed under Apache 2.0, which is typically not the case with frontier models.”

Stanford: AI overly affirms users asking for personal advice. “In a new study published in Science, Stanford computer scientists showed that artificial intelligence large language models are overly agreeable, or sycophantic, when users solicit advice on interpersonal dilemmas. Even when users described harmful or illegal behavior, the models often affirmed their choices.”

The Register: Telling an AI model that it’s an expert programmer makes it a worse programmer . “For alignment-dependent tasks, like writing, role-playing, and safety, personas do improve model performance. For pretraining-dependent tasks like math and coding, using the technique produces worse results. The reason appears to be that telling a model it’s an expert in a field does not actually […]

Ars Technica: Google’s TurboQuant AI-compression algorithm can reduce LLM memory usage by 6x. “Google Research recently revealed TurboQuant, a compression algorithm that reduces the memory footprint of large language models (LLMs) while also boosting speed and maintaining accuracy.”

NBC News: Using AI makes writing more bland, study finds. “The research team found that users who heavily relied on large language models (LLMs) produced responses that diverged significantly in meaning from the answers of participants who only partially relied on LLMs or avoided their use altogether, suggesting heavy AI use alters the substance of humans’ arguments in addition to changing […]

Harvard Business Review: ChatGPT vs. DeepSeek: What 5,000 Chinese Stocks Reveal About AI’s Limits. “ChatGPT, owned by US-based OpenAI, issued far rosier forecasts—with higher price targets and more ‘buy’ recommendations—than its Chinese rival DeepSeek when each platform evaluated nearly 5,000 publicly traded Chinese companies, says research by Harvard Business School’s Charles C.Y. […]

ZDNet: OpenAI’s GPT-5.4 mini and nano launch – with near flagship performance at much lower cost . “The latest GPT-5.4 mini model delivers benchmark results surprisingly close to the full GPT-5.4 model while running much faster, signaling a shift toward smaller AI models powering real-world applications.”

University of Waterloo: Top AI coding tools make mistakes one in four times. “Even the most advanced models achieved only about 75 per cent accuracy in the tests, while open-source models performed closer to 65 per cent. The study evaluated 11 LLM models across 18 structured output formats and 44 tasks designed to assess how reliably the systems followed structured rules.”

MakeUseOf: I switched to a local LLM for these 5 tasks and the cloud version hasn’t been worth it since. “Local LLMs have also come a long way, to the point where you can run lightweight AI models on just about every device. They’re not good at everything, but they do some tasks so well you’d want to cancel that cloud AI subscription right away.”

Spotted in my RSS feeds: CanIRun.AI. From the Why page: “CanIRun.ai runs entirely in your browser. When you visit the site, we use browser APIs to detect your GPU, CPU, and memory — then we calculate which AI models can run on your hardware and how fast. No data is sent to any server. Everything is computed client-side.”

MakeUseOf: I use Linux for local LLMs and everything is easier than Windows. “With the right tools and a bit of restraint, you can now run a genuinely useful ChatGPT-style setup locally on Linux Mint without turning your laptop into a space heater. I know because I just did exactly that on a Ryzen 5 machine with 8 GB of RAM and integrated graphics. Not a powerhouse, or a lab rig. Just a very […]