AI ConversatIons as SortIng Hats

Conversations about AI has been feeling a lot more loaded overtime. I thought maybe this is a decision bias, because I recently moved to a hub where I have a lot more of these converations than before. But no, I do think this is more about now then here. I lived in other hubs and did talk quite a lot about AI before 2023, and it was quite different. 

Maybe it’s because the conversation is attracting new profiles and backgrounds, which is great. But it also seems to attract our tendency to figure out which camp the other person is in as quickly as possible. If you’re excited about AI do you not care about data privacy? If you think we’re in an AI bubble then we won’t get to AGI soon? However, most people working in the field (or adjacent, you know) have quite nuanced opinions and probably won’t fit whatever our mental Sorting Hat concludes.

1. AI capabilities will continue to grow, AND there is also an AI bubble

Most people expecting tremendous capability growth also think not every startup that got thrown a $100M cheque for doing “AI” will survive. Even though I disagree with certain points that Zitron made in his widely-circulated article, I agree with one thing that this article puts out: there are so many dependencies in the market that it is almost impossible to expect a complete “AI takeoff” and there naturally will be some losers. In that nuance, for example, I personally find infrastructure as the most stable part of the ecosystem, and interestingly agree with Bezos’ statements here. But returning to Zitron’s examples, many of the new “AI neoclouds” could still fail, not because the market won’t grow, but because some of these companies pivoted from traditional cloud services into AI workloads without the business fundamentals or strategy to claim meaningful market share.

My timelines are a bit longer than the shortest ones out there, for reasons Demis Hassabis’ expressed in this interview. I don’t know if it is too long that some returns won’t be generated in expected time. I do believe that timelines are short for society to adapt to the changes it would bring as a whole. So saying “this is coming fast!” doesn’t make you a bubble-denier (if that’s a phrase).

2. AI can be great, and AI products can be awful

The latest models have been showing remarkable successes. While chatbots seem like the most scalable deployment of general-purpose LLMs, they are definitely suboptimal. It is true that chatbot companies don’t make it intuitive whether they use your data for training and while I tend to enjoy personalization more than others around me, I do think one should have the transparency to learn what’s going on with their data or product usage. I do really hope that the primary monetization of this technology ends up looking very different from social media. Maybe I’m too optimistic, but even though the companies do not have the right incentives, and that the road is rocky, I believe the adoption will evolve with more human and society-compatible ways. 

AI development itself is very dynamic, and beyond academic papers, the product releases are a window for us to look into what some of the biggest investors in AI R&D (companies like OpenAI, DeepMind etc.) are doing. I can get just as excited about a chatbot’s improved answers after a new model update as I do about a research milestone from a company that hasn’t even released a product yet. So, it’s not a mere fascination of these tools or companies themselves. In the meantime, though, I do think intentional users can gain enormous value from chatbots. Speaking from firsthand experience, this doesn’t make me indifferent to child safety, job loss, or other real risks which could indeed affect me as an employee, or my nieces and nephews as they grow up. That’s why I appreciate voices like Givens who emphasize the importance of using these tools deliberately and remind us in which ways we can, and should, demand stronger consumer protections and clearer standards.

I could come up with more, but I probably made the point. I hope I don’t do what I’m complaining about here, but if I do, I hope you’re a Ravenclaw.