Bingo. The secret sauce is never a sustainable long term moat. Things leak, competitors copy, or they make even better things, employees switch jobs. AI looks less vulnerable, since it’s academically difficult and expertise is still limited. But time and time again, new secret sauce ingredients last for months at a maximum, before they are exceeded oftentimes by hobbyists or other small actors.

I remember at Google they said well if source code leaks nobody is actually worried about stealing tech, the vast majority of code is open to all employees, with some exceptions like spam-, ranking, etc. It’s still protected, but not considered moat.

The moat comes from other things, such as datacenter & global networking infrastructure, marketing new products by pushing them through existing products (put Gemini in search, add chrome to Android etc). Most importantly you can use data you already have to bootstrap new products, say Gmail and calendar integrated with personalized assistants.

If you play your cards right, yes there’s some first-mover advantages, but they are more superficial than your average Twitter hype thread makes you think. It can give you the ability to set unofficial standards and APIs, like Kubernetes, S3 – (maybe OpenAI APIs?). And you can set certain agendas and market your name for recognition and trust. But all that can slip through your fingers if a behemoth picks up where you left off. They have so many advantages, except for being the fastest.

In fairness, the AGI definition predicted by doomsday safety experts who don’t have time for such unimportant concerns as copyright or misinformation via this tech, everyone who is most certainly not merely hyping to get investor cash and the utterly serious and scientifically grounded MIRI, is essentially, one company will achieve AGI, shortly after a singularity and that’s that. No one else could create a second because that scary AGI is so powerful.

Believe we are far away from that with LLMs and that the attention those working in model safety give to this over current day concerns like Meta just torrenting for model data, are not very serious people, but have accepted that this isn’t a popular opinion amongst industry professionals. Not least because letting laws prevent them from model training is bad, either for the same weird logic Musk uses to justify testing FSD Betas on an unwilling public due to the potential to prevent future deaths, or because they genuinely took RB seriously. No idea what’s worse for serious adults…

The funny thing is that nobody stops to ask if that version of AGI is even possible. For example, it would have to run a simulation of reality with enough accuracy faster than reality happens, and also multiple simulations in parallel. For physical world interaction, this is doable, but when it comes to predicting human behavior with enough accuracy, its probably not.

So even if we manage to come up with an algorithm that self learns, self expands, and so on, it likely won't be without major flaws.