We all gonna die! But AI will fire us first
The profile media and the industry mood remind me a person with bipolar disorder without medical help (Hello, Kanye). Some crawl in the shroud towards the cemetery, because tomorrow they will be released with black mark. Others tell how they gained family happiness by installing OpenClaw and connecting it to the fridge.
Meanwhile, the AI industry itself is quietly experiencing a fundamental shift. You can compare it to the movement of the continents. Not so long time ago, it was Pangaea (OpenAI) - a single monolith, known and understood by everyone. There were many tectonic plate beneath that were in constant motion, but none of this was visible from the outside. And now a great shift of these plates has begun, and so that Pangaea, if not completely crumbles, will definitely split into some kind of parts. It turned out (suddenly!) that Nvidia not only knows how to make cards (very advances guys will also mention CUDA), or racks of cards for data centers, but also excellent models, and in general, a vertically integrated stacks not only for data centers, but also for businesses of any size. AMD is slowly but surely making ROCm fully compatible with CUDA - that is, they are knocking the ground out of Nvidia’s feet, simply because as Microsoft once showed a long time ago - important the software platform is, not the hardware that serves it. Arm - one who quietly has its own “small profit” from everyone in the world, decided to also bite off a piece of the big pie and released its CPUs, and immediately for Facebook. At the same time, ARMv9 was developed from the very beginning with an eye on AI-loads, which are slowly but surely approaching the performance plateau, overcoming all the shortcomings.
Fine-tuning, prompt-engineering, AI-communication gurus - have sunk into oblivion and as quickly as they took off. Never happened before, and here it is again!
The architectures of the models themselves, without changing radically, have transformed from black box Terminators (which will capture all toasters tomorrow) into controlled and probability deterministic calculators, measured and evaluated from all sides by measuring tests.
The new reality and real benefit lies in the field of small agents that can do something by themselves. It is surprising that none of marketing geniuses has not yet announced the era of IoA (Internet of Agents) - after all, there is IoT (dead in marketing terms, but industry is alive and doing great).
OpenClaw and the hype around it have shown the real opportunity and benefit of a whole group of technologies and trends over last 20 years, which can now form from a number of streams (automation, big data, metadata and semantic models, chat bots and NUI) into a stream that really blows everything away.
And very few people from either side think about what knowledge and skills agents (who write code, make sales, you name it) will require from a person. After all, 99% of tasks and problems will not go anywhere, and 99% of people will also not be able to solve some complex specialized tasks. For example, writing code - sites development is ok, but from the point of view of architecture AI will return us to the 90s - static code, deployed and forgotten. For 3 pages or a landing page use cases where you don’t need a CMS it`s good solution. But something more - when you need to understand architecture, structure, patterns - this is much more difficult and important. Yes, in 5 years AI will be able to keep the context, the new architecture will solve the fundamental shortcomings of transformers (which does not mean a better solution to the problem of writing code at all), and you will be able to press the “make it beautiful!” button (by the way, it exists even now), but abstraction =! control and understanding reasons “why”.
Right now in Amsterdam at ComicCon, KubeCon every first stand is AI ready/enabled/powered XXX - everyone is selling tickets for a train that has long gone. The call “Deploy AI, or you’ll lose” has shown its incompetence and already smells. The cargo cult of AI is being replaced by the smart questions of the right architect “What? Why? How?”.
In the end, for those in the industry, regardless of their personal mood and vision of the situation at the moment, the questions remain the same: Which continent should I be on? Why there? How do I get there?
And only curiosity and openness will allow us to discover the still hidden huge segment of non-hype companies working in the rapidly growing field of AI, which grows incredibly fast.