It’s been awhile since I opined on AI. For those keeping score at home, I first wrote about OpenAI’s innovations back in 2018. At the time I was impressed with the demonstrations I saw, but also saw some real limitations that needed to be overcome. I revisited the topic two years ago where I said there were some valuable aspects of the technology, but there were good reasons to be skeptical about how transformative the technology would be. My hope was that it could be used to help empower human creativity as opposed to acting as a race to the bottom. I think in the two years since then we have enough information to see where things have gone, and unfortunately the race-to-the-bottom goals are winning. AI, more than anything else, is a tool for spam and cheap propaganda. Social media interactions are degrading, the ability to source hgih quality information on the internet is diminishing, and the overall reliability of outputs are decreasing. What we’ve gotten out of the deal is vibe coding, e.g. the ability to rapidly build apps that more or less work for toy projects. This is nice, but I think it is difficult to state how much less productive and well off we are as a society that we now need to contend with exponentially more spam in our web content.

The new dream among the AI booster set now is agentic AIs e.g. you use an AI prompt, and it takes that input, and runs it into some application that is able to actually do something useful for the user. Early results are not auspicious, but I have no doubt there will be some improvements along the way even if you can never fix the accuracy problem since LLMs are at their heart bullshit machines. I suspect at some point it will dawn on some of these developers that once they build a good enough application experience, they don’t actually need the AI at all, and we can get back to the business of building things that solve real-world problems. However in the meantime we do need to contend with the degraded information environment that has been unleashed on the world.