large language models

Articles tagged large language models on mistr.AI.

• In China, You Can No Longer Fire an Employee Over AI — Two Chinese courts in Hangzhou and Beijing ruled that a company cannot dismiss an employee simply because AI has taken over their work. According to the courts, deploying AI is a strategic decision, not an unforeseen change in circumstances. Meanwhile, the global tech industry laid off over 78,000 people in just the first four months of 2026 — nearly half of them because of AI. The US and EU provide no comparable protection.
• Copilot is for entertainment purposes only, says Microsoft — Microsoft has spent years investing billions in its AI assistant Copilot and integrating it into Windows and Office applications. Yet in its terms of use, it states in black and white that Copilot is for entertainment purposes only and that users should not rely on it for important decisions. How should we make sense of this contradiction?
• Seven AI breakthroughs in March 2026 that are pushing the boundaries of artificial intelligence — March 2026 brought a fundamental shift in artificial intelligence. AI systems are no longer just text generators — they are transforming into autonomous agents that plan, decide, and act. The cost of running models is falling, robots are learning to operate in the real world, and language models are beginning to understand code and security at the level of experienced developers. Here is an overview of the seven most important breakthroughs of the month.
• Why Local AI Alone Doesn't Guarantee Your Security — Many companies think that running AI locally is enough to take care of data security. But the reality is more complex. Local operation does mean that data doesn't leave your network, but without proper design, clear rules, and ongoing oversight, even a local model can become a security risk. Let's look at what local AI can actually do, where its limits are, and how to approach it with a clear head.
• From a One-Hour Prototype Came 1.5 Million AI Agents and an Offer from OpenAI — All it took was one hour of work to create a prototype that became a platform with 1.5 million AI agents. Austrian developer Peter Steinberger found himself courted by the heads of Microsoft, Meta, and OpenAI. Why did Sam Altman win, what can AI agents actually do, and where are their biggest weaknesses so far?
• Librarians Are Exhausted Searching for Books That Artificial Intelligence Made Up — Artificial intelligence is a great helper, but sometimes it has an overly vivid imagination. Librarians around the world have recently been reporting a troubling trend, as people flood them with requests for titles that simply don't exist. Why does this happen, and why do we tend to trust a chatbot more than a real person at the counter? We'll look at how AI "hallucinates" and why even seemingly magic phrases in prompts don't easily solve this problem.
• Open MCP Standard Will Break Down Barriers Between Enterprise Applications and Language Models — Remember how complicated it was to connect a printer to a computer before USB became the standard? Something very similar is happening in the world of artificial intelligence. The Model Context Protocol (MCP) technology, which teaches applications to effectively "talk" with language models, is becoming an open global standard under the Linux Foundation. Google is going all in on this innovation, and Czech developers are among its founders. What does this crucial step mean for the future of enterprise data, and why should you care? Find out in this article.
• Most Financial Institutions Are Tackling AI Ethics While Neglecting Fraud Prevention — Artificial intelligence has long ceased to be just a trendy marketing label — today it actually works in banks and insurance companies. Yet it turns out that only a small fraction of institutions have a truly bulletproof plan for using it fully and safely. While automation of routine tasks runs at full speed, the technology is surprisingly lagging in the fight against fraud. Let's look at what's holding bankers back and why they prefer developing tools themselves behind closed doors.
• Just 250 Manipulated Documents Are Enough to Make a Large Language Model Vulnerable — Imagine someone being able to sabotage a chatbot with just a few hundred manipulated texts. Anthropic, in collaboration with British security institutes, has found that as few as 250 malicious documents are sufficient to introduce a backdoor into a large language model. The size of the model or the volume of training data makes no difference. What does this mean for AI security?
• We Found Out Why AI Makes Things Up — and How to Learn to Prevent It — Every few weeks, headlines appear about AI lying, conspiring, or even seducing users. It sounds alarming, but the reality is far more mundane. The most extensive study of its kind showed that AI assistants make errors in nearly half of all cases. ChatGPT is no Bond villain. It's a statistical tool with design flaws that could have been prevented. And what's more — part of the problem lies with users themselves, who don't understand how to work with these tools properly.
• Samsung Proved That Successful AI Doesn't Need Billions of Parameters — What if it turned out that size isn't what matters in artificial intelligence? Samsung has introduced the TRM model, which has only 7 million parameters, yet in complex reasoning tasks it has literally outperformed models that are ten thousand times larger. This breakthrough proves that efficient architecture can matter more than raw computing power. What approach did Samsung use, and what does it mean for the future of AI?
• YouTube Is Testing Artificial Intelligence as a Music Guide — YouTube is launching a new initiative called Labs, which brings experimental features based on artificial intelligence. The first to debut are AI hosts in the YouTube Music app, acting as virtual DJs to enrich music listening with interesting facts and commentary. For now it's a limited test for selected users in the US, but it signals the direction the popular platform will be heading.
• Why Businesses Need Artificial Intelligence Orchestration for Real Success — Investments in artificial intelligence are growing every year, but most companies run into the same problem: isolated AI tools that don't communicate with each other. The result is inconsistent outputs, frustrated employees, and unmet expectations. The solution is AI orchestration, which can connect different models and data sources into a functioning whole. Find out why orchestration is becoming the key to successful AI deployment.
• OpenAI Launched a Browser with Security Problems It Warns About Itself — OpenAI introduced its new ChatGPT Atlas browser to the public a week ago. Within 72 hours, seven security companies discovered critical vulnerabilities. Atlas fails 94.2% of tests against phishing attacks, while Chrome stops 47%. OpenAI itself warns people not to use Atlas with sensitive data.
• South Korea Invests Billions in Homegrown Artificial Intelligence — South Korea has allocated nearly $400 million for the development of its own large language models. Five selected companies have been tasked with creating AI that will compete with OpenAI or Google. The government does not plan to fund all of them for the same duration. Every six months, results will be evaluated and the most successful will move forward. In the end, only two winners will remain. What is behind this strategy and how are the local players approaching it?
• Woke AI and Its Impact on Truth and Freedom of Speech in the USA — The term "woke AI" has become a new topic in artificial intelligence, stirred up by US President Donald Trump. According to him and some conservatives, "woke AI" poses a threat to the accuracy of information and independent thinking. At the same time, his strategy for combating this phenomenon raises questions about freedom of speech and possible infringement of rights under the First Amendment of the US Constitution.
• How to Use AI Chatbots in Business Correctly Without Unnecessary Risks — Generative AI chatbots promise a revolution in business writing, but the reality is considerably more complex. In this article I'll reveal when they truly make sense, why they are definitely not "all-powerful," and how to avoid critical mistakes when deploying them.
• AI Memory Under the Microscope: A New Look at How It Works — Memory in artificial intelligence is often wrapped in myths and misunderstandings. A new scientific framework brings a clearer division of memory types and defines six fundamental operations that determine how AI stores, updates, and forgets information. What does this mean for the future of AI?