Scarry myhtic stories

I am a myth! I’m a terror flying on AI wings! I am Mythos!

I wonder why about AI everyone is so interested in millions of lines of code, whether the LLM was able to handle Stroustrup during training, and how many million eyes looked into the any kind of bug and couldn`t see at all until the the mythical and AI-thical beast appeared…

After all, two things are interesting: the future first true automation (a boring topic) and the intellectual property underlying the neural networks (here even Cola can rest, but the recipe will be cracked too).

If everything is clear with the first topic - the agents do and will do any thing possible, and turn into the robots from movies that take over all the work, and the human only orchestrates, directs and manages towards the desired result, then with the second topic everything is much more interesting and “silenced”.

By the way, if you think about it, in the case of agents, management skills (providing guidance) will be very important, as will the skills of big and broad vision and creativity. So one hand nothing will really change - bad managers will also get an “AI” result regardless of the number and quality of agents, while good ones will be able to win quicker and faster.

But in general, there is little to change in this area simply because there are many of the first, few of the second, and the system comes to balance anyway.

But the most important and interesting thing in the case of Mythos is not Mythos itself, not what kind of bugs were introduced in the dinosaurs era it finds or will find, but rather the intellectual property, the code and the algorithms that taught it to do it.

The models themselves have no special value and will not have, but will become commodities. Although large models may remain very expensive and affordable to create by few, but not all of us drive a Maseratti, and somehow the world did not fall apart because of it. The future of GPT models is generally not predictable at the moment — the fog of future architectures allows us to predict a future with the quality of Wanga or Nostradamus.

The value is not in the model, but in the “sauce” that creates and trains the model. On this path, we have already queitly passed several stages - the first was the reasoning of the models, and now the orchestration of these models is clearly coming to the edge. Simply put, the reasoning`a orchestration — here the model found a strange/suspicious place in the code, and what to do with it? How not to lose context (even if it is 1M characters), link pieces of the desired code, in the right sequence, etc.

And it is this sauce that scares AI pessimists. It is not the models themselves that are numerical thrashing machineы, but that orchestration of models and automation that will allow automating a bunch of work and processes.

At the same time, it is somehow overlooked that it is, although a huge amount of work can be dropped on AI and a huge number of people thrown out of work (including even me). But no less people will still be needed in order to build processes, production lines, tracking agents’ work and everything connected and tied around AI based on agents.

Yes, our generation will clearly work more and longer than the next one, and next gen will have less work, but only because the chain of knowledge is and will be in a wild imbalance: young people do not have knowledge, AI knows everything (until the cutoff date, lol), and our generation has at least outdated knowledge in practice. And in the next 20-30 years, that balance will even out, and there will only be isolated cases like the US IRS and some “floppy airports”. And although the examples given are not the best, since with the help of Mythos even Cobol will become a myth, we are building a new generation of legacy right now.