Today I’m going to walk you through a fairly unique position: that OpenAI is just another boring AI startup lacking any meaningful product roadmap or strategy, using the press as a tool to pump its bags while very rarely delivering on what it’s promised. It is a company with massive amounts of cash, industrial backing, and brand recognition, and otherwise is, much like its customers, desperately trying to work out how to make money selling products built on top of Large Language Models.
OpenAI lives and dies on its mythology as the center of innovation in the world of AI, yet reality is so much more mediocre. Its revenue growth is slowing, its products are commoditized, its models are hardly state-of-the-art, the overall generative AI industry has lost its sheen, and its killer app is a mythology that has converted a handful of very rich people and very few others.
It’s such a pity too, because surely there’s actual money to be made from consulting with industry on optimization of all sorts of processes and products or grant money to obtain by helping looking through large scientific datasets etc. You know, useful stuff.
Instead, these twats wasting ridiculous amounts of electricity and hardware on building better echo-chambers.
They are clearly targeting very broad audiences instead of any specific processes or operations any one company may do. It’s part of their view that they are working on general intelligence. Not specific intelligence.
I agree that it may be a less productive approach. Especially since pure scaling did hit a wall. Even if everyone and their sycophantic singularity cheerleaders say otherwise.
That’s why I believe OpenAI will eventually go down the shitter, because they don’t have a good usecase. Instead, they’re trying to do a “one fits all” approach and burn a ton of money in the process.
Companies like Claude Code might be here to stay, because they’re running (and improving) their own models and they focus on a single usecase. Claude is already much better than ChatGPT when it comes to generating useful code.
But we’ll see what happens when the bubble inevitably bursts, because they all seem to be hitting a ceiling already and the improvements within each iteration get more and more marginal.
The “undifferentiated service for all the world” business seems to eventually pivot to advertising. Facebook, Google, etc.
So eventually ChatGPT won’t shut up about the new Ford Maverick.
It would have still had the flaws and issues inherit with the process, but imagine current level LLMs that were trained on data that was curated and responsibly collected. It would also have helped to have the training emphasis on being accurate and search for verification and clarify, rather than be pleasing and agreeable to a fault. In the end it’s not AGI and can’t be alone, but the methods chosen was the direction that would be profitable and that’s where and why we are here now.