Damian Voltes

2025 is ending, and it's clear that AI is behind the expectations we had one year ago. Sure, the new models are better but in the same way that the iPhone 17 is better than the 16. What is going on, and what do I imagine happening in the next year? For one, it's become clear that verbal tasks are a small subset of what we call intelligence. This was not completely obvious when ChatGPT came out. There was the hope that creating larger models would lead to emerging intelligent behavior that would include better reasoning. It's clear now that larger models and forced introspection (verbal reasoning) is not enough. Models keep making mistakes, they keep hallucinating. It's a fundamental problem of how they work: they predict text that needs to be fact-checked later. Many users do not realize this, and end up propagating false information. Secondly, the vast majority of intelligent decisions in life are nonverbal. If you ask Messi why he made a certain play, he will probably say something "I took the touch to draw the defender in, waited for him to commit, and then..." Of course he'll say it in Spanish, but the point is that those words never existed in his head. He was making decisions just like a cat or like a Waymo. Sensory inputs, and quick nonverbal thinking (Kahneman's System 1). This is why we still don't have independent autonomous workers that can be given a task to complete in a week. Life involves too many small decisions that LLMs are not equipped to handle. You somehow realize that something you're doing makes no sense and you change course. You don't have a constant internal monologue that operates while you're at work. So what does this mean for companies like OpenAI and Anthropic? OpenAI has raised something like 60B dollars thus far. One round of funding was 40B at a 300B valuation. This means investors expect it to become a company as large as Nvidia. How does this happen? Not with $20 per month for ChatGPT, but with the automated worker that costs $2000, and replaces a human that makes 2x or 3x as much. But I'm pretty confident we're nowhere close to that, just like we were not close to scalable self-driving cars in 2015. So when they need funding again the investors are going to push back. Meanwhile, companies like Google can keep up without investor help. Google makes 350B of ad revenue per year, and it can patiently wait for OpenAI to run out of reasonable sources of funding. I imagine this is why OpenAI attempted to create a browser and a social network this year: it's worth a shot trying to beat Google at its game, and getting a piece of the ad revenue cake. Do you use ChatGPT Atlas? Do you scroll Sora every day like you do Instagram? I'm pretty sure the answer is no. So I imagine it's likely that OpenAI will one day become part of Microsoft, something similar will happen to Anthropic. Then the incumbents will have all the time to develop AI more slowly and with less hype. Just like with self-driving cars.