LLM

I remember back in November 2022, when ChatGPT was released and everyone was going crazy about how smart and humanlike it was. I sat down and started talking to it in Czech. After reading so many raving reviews and hyperbolic predictions, I was shocked by how quickly I ran into the model’s limits in Czech. Now don’t get me wrong: I know my native tongue is not a top-tier language, given it only has about 10 million native speakers. I don’t expect AI models to be as proficient in Czech as in English or other major languages. But it still made me think: if I am getting a much worse experience from large language models than English speakers are, what is the experience like for people who speak languages with even less digital support? ...

It started with a broken hinge in our kitchen cabinet. I am no handyman, but I like to do as much as I can around the house and I dare say I’ve gotten pretty okay at it. The one thing that can set me back, though, is not knowing the correct terminology. What type of hinge do I need? Concealed? Inset? Overlay? Clip-On? Soft-close? Sprung? An avalanche of terms that I only barely understand paralyzes me. Luckily, there’s a new helper in town: AI! ...

Recently, I have been testing how well the newest generations of large language models (such as GPT-5 or Claude 4.5) handle natural language, specifically counting characters, manipulating characters in a sentences, or solving encoding and ciphers. Surprisingly, the newest models were able to solve these kinds of tasks, unlike previous generations of LLMs. Character manipulation LLMs handle individual characters poorly. This is due to all text being encoded as tokens via the LLM tokenizer and its vocabulary. Individual tokens typically represent clusters of characters, sometimes even full words (especially in English and other common languages in the training dataset). This makes any considerations on a more granular level than tokens fairly difficult, although LLMs have been capable of certain simple tasks (such as spelling out individual characters in a word) for a while. ...

Our old friend Pygmalion Humans have always anthropomorphized1 non-human entities, from ancient gods and natural forces to modern cars and ships. But our tendency to see machines as human-like has dramatically intensified with the advent of conversational AI. This isn’t entirely new - the word “robot,” coined and popularized by Čapek brothers in 1920, originally described artificial beings indistinguishable from humans. Our drive to find and create humanness in artificial entities runs deep, even back to the ancient times with myths like Pygmalion and Galatea. ...

Recently, I have been re-reading one of the classic books of Czech literature, Karel Čapek’s War with the Newts, and I have been intrigued by some of the parallels between Čapek’s newts and LLMs. Mind you, I am probably highly primed to associate and connect random things to LLMs given how much I read and talk about AI these days (a case of apophenia?). But hear me out! I think we’re on to something. ...

Many moons ago (in fact long ago so that contemporary sources are beginning to be difficult to find), iPhone users were told they’re holding their phones wrong. This was one of the more memeworthy communication blunders of Apple and something we’re unlikely to see from them, but the reason I mention it here is that asking ourselves whether we’re holding the thing wrong is actually a good strategy for exploring novel use cases. ...