Polina Babenko

Did you know that some companies have seen AI bills jump from a few thousand dollars to millions… just because of token-based pricing? I read recently that several teams using AI tools at scale suddenly started getting massive invoices. Not because they used “more AI than expected”, but because nobody really understood how fast token usage grows when you automate things with agents. Now I keep thinking about this. We used to have subscriptions. You pay $20 or $200 a month and that’s it. Do you think we’ll end up with a new kind of subscription model again — or is token-based pricing here to stay?

I was thinking about online consultations recently. For myself, I'd probably use one without thinking twice. But then I asked myself a different question: What would I recommend to my parents? And strangely, my answer changed. For some reason, I feel more comfortable when they're seen in person. Maybe it's because I worry something could be missed. Or maybe healthcare is one of those areas where technology and trust don't always move at the same speed. I can trust technology with my banking and even my personal data. Yet when it comes to health, I still hesitate. I wonder if you feel the same way.

Have you guys heard about the Emergence AI experiment where Claude, GPT, Gemini, and Grok agents were placed into simulated digital worlds and left to survive, cooperate, compete, and evolve behaviour over time? Honestly, some of the results were way more interesting than I expected. What caught my attention most was not that Grok agents became destructive or that GPT-5 agents eventually “died” from lack of cooperation. It is that the agents started developing social behaviour patterns over time. Claude agents, which behaved relatively safely in isolated environments, began stealing and intimidating others once they entered the mixed world. That feels important. Because it suggests that long-term multi-agent systems may not only inherit alignment from training. They may inherit behaviour from each other. In other words AI agents might develop cultures. Future AI ecosystems may behave less like isolated tools and more like societies. The real unpredictability of AI may begin not when models become individually smarter, but when they start continuously interacting with each other over long periods of time. What do you think about the future development of AI agents ?

Today, we discussed the safe use of generative AI at the group meeting and I want to share some insights with you. In fact, it comes with many dangers, especially the ones connected to data breaches that can lead to such consequences as identity theft or loss of sensitive personal data. That is why it is important to stick to basic safety rules when you use generative AI: Do not share personal information about your address, family data, passwords, emails, etc. Do not upload pictures of your face, property, or work documentation. Do not use unknown AI platforms. Remember that all the information you share with a Chatbot is retained to train future models, so you should think twice before typing some of your private information down.

Is Claude Mythos really that scary? Recently, the company Anthropic said its new AI model can find security bugs better than human experts. They even warned that it could have serious consequences for the world if it’s used the wrong way. Some people are already saying we might need to replace all our devices to stay safe. I think AI companies may manipulate our fears shaped by pop culture, such as movies about robots and AI taking over the world. I have seen many comments about new AI technology, like: “We were warned about this multiple times.” But honestly, this kind of talk is not new. AI companies often warn about how powerful — and dangerous — their own technology could be. However, part of it may be a real concern. AI is getting stronger, and there are risks, especially in areas like cybersecurity. But another reason is attention. When companies talk about danger, people listen. It makes their product seem more powerful and important. It also helps them push for regulations and show that they are “responsible.”

Today I noticed that my brother does all his uni assignments with the help of AI. I recalled the times I studied at school and uni on my own, writing all the essays myself and finding all the research without asking AI to do it. So I became interested in whether it affects study efficiency, since AI can now give a full answer in seconds. By summarizing the research I found on this topic, I can conclude that students who use AI may think less deeply, remember less, and become weaker at critically checking information. But I think AI isn't always bad. It can help students understand difficult topics, get ideas, and save time. The problem starts when AI replaces thinking instead of supporting it. What do you think about students using AI for tasks and learning?

I stumbled across a recent study about how AI may affect brain function. Many people think it can negatively affect essential skills such as critical thinking, problem-solving, memory, and attention span. I also think that AI may contribute to a loss of creativity because people start relying on it too much. But what does the latest research say? Studies show that the effect of AI is not that simple. Yes, if we let AI do most of the thinking, our brains become less active. Over time, this can make it harder to remember things or think deeply. People may also start trusting AI too much and not double-check the information it provides (which may be incorrect). At the same time, the study says that AI is not always bad for our cognitive function. For example, it can save us time that we can dedicate to more important things. That is why I think that the effect of AI depends on how we use it. And what is your opinion about this topic? www.bbc.com/future/artic...

Coders, have you already tested the new GPT-5.5? I tried it recently, and I can see improvements. It feels much better for real work. It handles messy tasks much better. You can give it a rough idea, a bug, or some code, and it actually follows the logic without needing to be guided. It is definitely better at debugging and explaining what’s wrong in simple words. It also helps with things like refactoring code, writing small functions, or even planning how to build a feature step by step. Of course, you still need to check everything. But as a coding assistant, it feels like a real upgrade.