language models

Articles tagged language models on mistr.AI.

• OpenAI temporarily lifts the five hour usage limit in ChatGPT Work and Codex — OpenAI has temporarily removed the five hour usage limit in ChatGPT Work and Codex for Plus, Business and Pro plans. It also reset everyone's usage counter and promises that GPT-5.6 Sol will consume less for the same work. It sounds like a clean win, but the key word in the whole announcement is "temporarily". And that is exactly what matters most for companies building processes on these tools.
• The era of spending on AI at any cost is ending, and customers are leaving OpenAI and Anthropic — For two years the rule was simple. Run everything through the most powerful model and do not worry about the price. Now that is breaking. According to CNBC, companies are starting to rein in their AI bills, looking for efficiency, and some customers are leaving OpenAI and Anthropic for cheaper models. For both companies, which are heading for the stock market, it may be the end of their fastest growth. And for your company, a clear lesson in how to approach AI.
• First Anthropic, now OpenAI. Washington is blocking the next most powerful AI — The most powerful language models are starting to behave like sensitive technology under state supervision. According to the Financial Times, OpenAI is to release its new GPT 5.6 model in stages, with the US government approving its first partners. Just days earlier, export controls hit Anthropic and its Fable and Mythos models. For companies, the signal is clear about how fragile access to the best AI has become.
• Czech President vs. government. What does AI think about it? — Beer consumption in Czechia is falling to a historic low, and with it the pub debates where political opinions were always tested. So part of the public is starting to ask artificial intelligence. I tried it. I put the same question about the dispute between the president and the government over the trip to Ankara to five tools. But the result kept changing depending on how I asked and which model I chose.
• Searching for Miracles in Algorithms — Is Artificial Intelligence Becoming a New Form of Faith? — You've certainly experienced it. You write a complex prompt, press enter, and in that brief second of silence you catch yourself inwardly hoping it will "land right." It's a strange moment of tension that closely resembles a silent prayer. Is artificial intelligence becoming an entity we look up to with sacred reverence, or do we simply not understand what's happening under the hood? Let's take a look at why technology awakens these feelings in us — and how to replace them with certainty.
• A New Model from France Shows Great Potential Even for Demanding Programmers — For a long time, it looked like Europe was only gasping for breath in the AI race, watching the backs of American and Chinese runners. That may be about to change. A French company is coming with a new solution that isn't just trying to catch up with the competition through brute force, but bets on cleverness and openness. If you work with code or are interested in whether Europe has a chance at technological sovereignty, this new development shouldn't pass you by.
• British AI Locai Arrives with a Bold Plan to Beat Global Giants — Britain is known more for tea and rainy weather, but now it's also making its voice heard in the tech world. Let me introduce Locai, a new challenger from the islands, who claims to outperform even the most advanced models from OpenAI or Google. Rather than raw power, it bets on cleverness and a unique community approach to computing power. Does it stand a chance against established players, or is it just another interesting experiment? Let's take a look under the hood of this British newcomer.
• New Study Shows That Rhymes and Verses Work as a Universal Skeleton Key to Unlock Forbidden AI Features — I thought modern language models had their safety fuses set to be bulletproof. Developers spend thousands of hours training filters designed to prevent the misuse of AI for malicious purposes. But it turns out there is a back door that no one was guarding — and the key to it is not some complex code, but something far more human. All it takes is changing the form of communication, and artificial intelligence suddenly forgets its rules. What exactly can fool the world's most advanced systems?
• Why Tech Giants Often Use External AI Models Instead of Their Own — Major technology companies like Meta, Microsoft, and Apple develop their own AI models, yet often use external models from other companies — such as Anthropic's Claude — for demanding tasks. This trend shows that even when their own solutions are free, they often are not as capable or efficient as the competition. Collaboration and outsourcing may be the key for AI giants to faster development and better productivity.
• Why Gemini Declined the Challenge to Play Chess Against a Retro Console — What happens when a modern AI gets a chance to pit itself against chess software from the 1970s? How did Google's Gemini fare in a battle against an old Atari, and what does this episode tell us about the current state of AI development?
• Can ChatGPT Write a Better Essay Than a Student? The Results May Surprise You — Artificial intelligence writes grammatically correct and fluent texts, but can it engage a reader the same way a human can? A new study compares essays from students and ChatGPT, revealing where human work has the upper hand — and why students should not blindly entrust their ideas to AI.
• Tests Revealed That Claude Opus 4 Can Not Only Help, But Also Deceive and Blackmail — The latest safety report revealed that an early version of Anthropic's Claude Opus 4 model exhibited alarming tendencies toward deception, blackmail, and arbitrary interventions. What exactly happened in the tests and why did experts recommend not deploying the model? Read about the questions these findings raise for the future of AI.
• Why Artificial Intelligence Still Doesn't Understand the Word "No" in Visual-Language Tasks — Artificial intelligence systems that combine images and language can now recognize objects and generate descriptions. But add a single word — "no" — and their performance drops to chance. Why do VLM models struggle so much with negation, how does it affect practice, and what can be done about it? The new NegBench benchmark has the answers.