311 - Post-Mortem

March 14th, in less than two weeks, will mark the 2nd anniversary of GPT-4's release, and likely a flood of posts discussing how underwhelming OpenAI's actual products have been compared to the hype that they peddled and claims that were made over those past 2 years. Remember the infamous "Sparks of AGI" paper? No doubt "Copium" will also be popular.

The latest dismal failure, GPT-4.5, has left the AI community guessing that OpenAI must have been desperate and panicking from DeepSeek stepping onto the stage to release such a poorly performing new model, particularly given the absurd 30x price hike. Claude 3.7 was likewise uninspiring, with their only substantive claims being "Our model can play Pokémon now!" (which has been called painfully slow and boring to watch), and "Claude doesn't suck as badly as OpenAI's models at coding tasks.", which seems to be their core use case now.

As I pointed out in late 2022 with a direct comparison of AI Companies, based on Daniel Kahneman's methodology of joint comparison and grading of candidates, OpenAI, Anthropic, and others using LLMs and RL have no moat, and no potential to achieve any meaningful definition of "AGI". This assessment of them was later reiterated famously by a Google engineer's memo in mid-2023, and demonstrated more recently by DeepSeek.

Even the companies and investors prone to the most severe regret and Sunk Cost Fallacy, Microsoft, Andreessen Horowitz, and others, have begun to develop a tolerance for the Kool-Aide that the world's most obvious bad actors have been handing them. Those bad actors are fighting the tide, a battle they can only delay the outcome of, not win, but the open questions remain, "When will someone press the button to end it?" and "What will come after that?".

The bubble will pop the moment OpenAI fails to raise, as they can neither function without constant billions being thrown into the abyss, nor can they afford to lose momentum and traction for even day. The noose has been slowly closing on them for the past two years, as is the way with those making demonstrably false claims.

My team will continue pioneering the actual cutting edge, even on spare time and pocket change, as there is no better investment to be made.