All posts tagged: llms

The Download: a new Christian phone network, and debugging LLMs

The Download: a new Christian phone network, and debugging LLMs

Published by skeptic

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. A new US phone network for Christians aims to block porn and gender-related content A new US-wide cell phone network marketed to Christians is set to launch next week. It blocks porn using network-level controls that can’t be turned off—even by adult account owners. It’s also rolling out a filter on sexual content aimed at blocking material related to gender and trans issues, optional but turned on by default across all plans. The trouble is, many websites don’t fit neatly into one category. That leaves its maverick founder with broad, subjective control over what is allowed or banned. Read the full story. —James O’Donnell This startup’s new mechanistic interpretability tool lets you debug LLMs The San Francisco–based startup Goodfire has released a new tool, Silico, that lets researchers peer inside an AI model and adjust its parameters during training. It could give users more control over how this technology is built …

This startup’s new mechanistic interpretability tool lets you debug LLMs

This startup’s new mechanistic interpretability tool lets you debug LLMs

Published by skeptic

Mapping models Silico lets you zoom in on specific parts of a trained model, such as individual neurons or groups of neurons, and run experiments to see what those neurons do. (Assuming you have access to the model’s inner workings. Most people won’t be able to use Silico to poke around inside ChatGPT or Gemini, but you can use it to look at the parameters inside many open-source models.) You can then check what inputs make different neurons fire, and trace pathways upstream and downstream of a neuron to see how other neurons affect it and how it affects other neurons in turn. For example, Goodfire found one neuron inside the open-source model Qwen 3 that was associated with the so-called trolley problem. Activating this neuron changed the model’s responses, making it frame its outputs as explicit moral dilemmas. “When this neuron’s active, all sorts of weird things happen,” says Ho. Pinpointing the source of odd behavior like this is now pretty standard practice. But Goodfire wants to make it easier to adjust that behavior. …

LLMs hallucinate the most when you ask them to do this

LLMs hallucinate the most when you ask them to do this

Published by skeptic

Large Language Models are great at faking confidence. You can ask ChatGPT, Gemini, or Claude just about everything in the sun, and in most cases, you’ll get a well-structured, confident-sounding answer right away. However, just because your model sounds confident, it doesn’t necessarily mean it’s right. We’re all too familiar with LLM hallucinations — the model casually invents a quote, cites sources that don’t exist, or gets dates wrong. You might think that AI hallucinations are a thing of the past and that modern models don’t hallucinate as much, but that’s not the full truth. Certain kinds of requests can send your LLM over the edge, and they show up more often in your daily prompts than you might think. Related I built one ChatGPT prompt that works for absolutely any scenario This simple ChatGPT prompting structure works for any goal, big or small. Ask it to do math, and things fall apart Your favorite AI isn’t as good at math as you think it is LLMs are built to analyze and generate text, not …

Why We Should Be Reading Paul Churchland Right Now: Neurophilosophy and AI

Why We Should Be Reading Paul Churchland Right Now: Neurophilosophy and AI

Published by Jenni Sidey

The more I get into philosophical and philosophy-adjacent discussions of current-generation “artificial intelligence” (large language models and the like), the more dismayed I am not to see any discussion of the large body of relevant work by Paul Churchland. (Full disclosure: he was my dissertation chair.) Paul is not ordinarily thought of as a philosopher of AI, but rather as a philosopher of mind and of neuroscience. However, for reasons I hope to make clear in this post, Paul was one of the first philosophers to engage in detail with the predecessors of the technology behind systems like ChatGPT, and he provides quite extensive conceptual resources for beginning to address many of the ontological and epistemological questions about this type of AI. Paul Churchland is a naturalistic philosopher who has written widely on philosophy of science, philosophy of mind, epistemology, and philosophy of language. For most of his career, the primary naturalistic lens through which he has pursued a variety of philosophical concerns is that of the neurosciences, and from roughly 1986 on, that primarily …

There’s Something Fundamentally Wrong With LLMs

There’s Something Fundamentally Wrong With LLMs

Published by José Betancourt

Sign up to see the future, today Can’t-miss innovations from the bleeding edge of science and tech Once you familiarize yourself with the characteristics of ChatGPT’s writing, it becomes impossible to miss. The internet has been flooded with AI-generated text that often features distinctive language patterns, from liberal use of em dashes and repetitive sentence structures to specific turns of phrase and tone. The trend has become so ubiquitous that experts are now warning that it could even influence the way we speak in real life. As historian Ada Palmer and cryptographer and author Bruce Schneier argue in an opinion piece for The Guardian, it’s a very real risk that could also entrench a fundamental flaw plaguing large language models today. While these models were trained on vast quantities of written text, social media posts, movies, TV shows, and other recordings, this data often comes up short in “unscripted conversations we have face-to-face or voice-to-voice” — which represents the “vast majority of speech, and a vital component of human culture.” It’s a massive blind spot …

From LLMs to hallucinations, here’s a simple guide to common AI terms

From LLMs to hallucinations, here’s a simple guide to common AI terms

Published by skeptic

Artificial intelligence is a deep and convoluted world. The scientists who work in this field often rely on jargon and lingo to explain what they’re working on. As a result, we frequently have to use those technical terms in our coverage of the artificial intelligence industry. That’s why we thought it would be helpful to put together a glossary with definitions of some of the most important words and phrases that we use in our articles. We will regularly update this glossary to add new entries as researchers continually uncover novel methods to push the frontier of artificial intelligence while identifying emerging safety risks. Artificial general intelligence, or AGI, is a nebulous term. But it generally refers to AI that’s more capable than the average human at many, if not most, tasks. OpenAI CEO Sam Altman recently described AGI as the “equivalent of a median human that you could hire as a co-worker.” Meanwhile, OpenAI’s charter defines AGI as “highly autonomous systems that outperform humans at most economically valuable work.” Google DeepMind’s understanding differs slightly from these two definitions; …

Meta’s new structured prompting technique makes LLMs significantly better at code review — boosting accuracy to 93% in some cases

Meta’s new structured prompting technique makes LLMs significantly better at code review — boosting accuracy to 93% in some cases

Published by skeptic

Deploying AI agents for repository-scale tasks like bug detection, patch verification, and code review requires overcoming significant technical hurdles. One major bottleneck: the need to set up dynamic execution sandboxes for every repository, which are expensive and computationally heavy.  Using large language model (LLM) reasoning instead of executing the code is rising in popularity to bypass this overhead, yet it frequently leads to unsupported guesses and hallucinations.  To improve execution-free reasoning, researchers at Meta introduce “semi-formal reasoning,” a structured prompting technique. This method requires the AI agent to fill out a logical certificate by explicitly stating premises, tracing concrete execution paths, and deriving formal conclusions before providing an answer.  The structured format forces the agent to systematically gather evidence and follow function calls before drawing conclusions. This increases the accuracy of LLMs in coding tasks and significantly reduces errors in fault localization and codebase question-answering.  For developers using LLMs in code review tasks, semi-formal reasoning enables highly reliable, execution-free semantic code analysis while drastically reducing the infrastructure costs of AI coding systems. Agentic code reasoning …

Mistral bets on ‘build-your-own AI’ as it takes on OpenAI, Anthropic in the enterprise

Mistral bets on ‘build-your-own AI’ as it takes on OpenAI, Anthropic in the enterprise

Published by skeptic

Most enterprise AI projects fail not because companies lack the technology, but because the models they’re using don’t understand their business. The models are often trained on the internet, rather than decades of internal documents, workflows, and institutional knowledge.  That gap is where Mistral, the French AI startup, sees opportunity. On Tuesday, the company announced Mistral Forge, a platform that lets enterprises build custom models trained on their own data. Mistral announced the platform at Nvidia GTC, Nvidia’s annual technology conference, which this year is focused heavily on AI and agentic models for enterprise. It’s a pointed move for Mistral, a company that has built its business on corporate clients while rivals OpenAI and Anthropic have soared ahead in terms of consumer adoption. CEO Arthur Mensch says Mistral’s laser focus on the enterprise is working: The company is on track to surpass $1 billion in annual recurring revenue this year. A big part of doubling down on enterprise is giving companies more control over their data and their AI systems, Mistral says.  “What Forge does is …

Spanish ‘soonicorn’ Multiverse Computing releases free compressed AI model

Spanish ‘soonicorn’ Multiverse Computing releases free compressed AI model

Published by skeptic

Large language models have a problem: they are large. Multiverse Computing, a Spanish startup, is addressing this issue with compressed models that aim to close the gap between what frontier models can do and what companies can actually afford to deploy.  The secret sauce is CompactifAI, a compression technology inspired by quantum computing that the Basque company has applied to models released by OpenAI. As of today, developers can access a newer version of Multiverse’s HyperNova 60B model for free on Hugging Face. The company also plans to open-source more compressed models in 2026 to support a wider range of use cases. According to Multiverse, its models are smaller, but nearly as potent and accurate. At 32GB, HyperNova 60B is roughly half the size of the model it derives from — OpenAI’s gpt-oss-120B — while boasting lower memory usage and lower latency. The updated version, called HyperNova 60B 2602, now also better supports ​​tool calling and agentic coding, where inference costs can be high. One of the competitors Multiverse claims to have beaten with HyperNova …

MIT’s new fine-tuning method lets LLMs learn new skills without losing old ones

MIT’s new fine-tuning method lets LLMs learn new skills without losing old ones

Published by skeptic

When enterprises fine-tune LLMs for new tasks, they risk breaking everything the models already know. This forces companies to maintain separate models for every skill. Researchers at MIT, the Improbable AI Lab and ETH Zurich have developed a new technique that enables large language models to learn new skills and knowledge without forgetting their past capabilities. Their technique, called self-distillation fine-tuning (SDFT), allows models to learn directly from demonstrations and their own experiments by leveraging the inherent in-context learning abilities of modern LLMs. Experiments show that SDFT consistently outperforms traditional supervised fine-tuning (SFT) while addressing the limitations of reinforcement learning algorithms. For enterprise applications, the method enables a single model to accumulate multiple skills over time without suffering from performance regression on earlier tasks. This offers a potential pathway for building AI agents that can adapt to dynamic business environments, gathering new proprietary knowledge and skills as needed without requiring expensive retraining cycles or losing their general reasoning abilities. The challenge of continual learning Once an LLM is trained and deployed, it remains static. It …