Uhlík vs Křemík · Czech podcast about artificial intelligence

Why AI sounds convincing even when it is wrong, and what a free chat really costs

Episode 1 · 23 January 2026 · 39:56

In the pilot episode of the podcast Uhlík vs Křemík, host David Salomon, who calls himself an AI llama, asks his co-host Adam Hanka what today's artificial intelligence can really do. Adam Hanka, a mathematics and physics graduate who leads data and artificial intelligence at Creative Dock, explains how to check what AI tells you, how to get started with it and why you should watch what you upload into it.

A new animal we are only getting to know

By his own account, Adam Hanka has been working with artificial intelligence for the last 10, maybe 12 years, and for most of that time nobody cared. He describes today's wave of interest with a comparison from his university days. A programmer is supposed to know what software he is writing and how it will behave. With today's AI, however, we are more like biologists who have caught a new animal and are trying to find out what it is good for.

On top of that, the world keeps changing under our hands, because a new model comes out every so often. Hanka admits that over the past three years AI has actually created more work for him rather than less. Still, he says newer models are easier to control than two years ago, when they were already usable and yet he had to wrestle with them a lot.

Consistent does not mean true

A key part of the conversation is about hallucinations. David Salomon asks whether a wrong answer can be recognised somehow, for example by a different font. Hanka answers plainly: "No, you won't recognise it, because it is consistent. It just isn't true." In his view, this is a deeper problem, because people were not made to recognise the truth, but to recognise what is consistent.

As a defence, he recommends working with sources. Either with vetted lists of articles and websites, or with your own data, or by asking the AI for a source for every claim and clicking through the links, because a model can make up a source too. For important things, he has a report prepared in an OpenAI model and then has it checked by Google's Gemini model. Into his own custom GPTs he puts a list of links the model should draw from.

He describes artificial intelligence itself as a tool that has learned the structure of human language plus information from the internet and from people. Put simply, "it is a machine that has learned the outward signs of human intelligence".

How to start and what you do not need

Asked what someone who has never tried AI should do, Hanka has simple advice: switch off the podcast and start playing with a chat model. People first ask about unimportant things, he says, and then perhaps take a photo of their open fridge and ask what to cook today. You do not need technical knowledge to begin. What helps is curiosity and self-reflection, meaning the ability to notice where AI could help you. He would not recommend ChatGPT for psychological help, as there are specialised apps for that.

The conversation also turns to children. Hanka says he is no expert, but if social networks are meant to be for ages 13 and up, he would treat artificial intelligence at least the same way. If children try it out at school in a safe environment, he thinks that is fine.

Data as currency

According to Hanka, artificial intelligence has been running our lives since roughly 2016 at the latest, through social media algorithms that choose what content we see. David Salomon points out that people confide quite private things to chatbots, and from what he has heard, some even upload their psychological profile from psychological tests. Hanka replies that this way we hand over a lot of sensitive data about ourselves. That data is always stored on servers, and with the free version it may be used to train further models. "So on that free account, you pay with your data." That is why he recommends having at least a basic paid licence, where the data is not used for training, and not uploading things like your personal identification number.

At the end, Hanka talks about his training sessions. Czechs, he says, like to use new technologies, and some companies test products in the Czech Republic earlier than elsewhere in Europe. In larger companies, however, he often finds that people first need to properly master Excel and working with data. So he also covers what Markdown or the CSV format is.

You can listen to the whole conversation in the pilot episode of the podcast Uhlík vs Křemík.

The episode is in Czech. Czech page with chapters, quotes and the full transcript