Neural networks have learned to hack the brain: December apocalypse or a new era?

If you thought AI was developing fast, you haven't seen anything yet. The first week of December fired news like a machine gun.

DeepSeek updated its line to version V3.2, Mistral rolled out a new flagship, and the generative Gen-4.5 model from Runway managed to bypass the top-end Veo 3, Sora 2 Pro and Kling. All this in seven days.

Pavel Durov launched Cocoon — and this is perhaps the most unexpected turn of the week. The creator of Telegram decided that messengers were not enough for him, and stepped into the world of neural networks. Intriguing, right?

Meanwhile, Anthropic is setting its sights on an IPO. The company that created Claude is preparing to go public. A sign that the AI industry is moving from the stage of “budding” to the status of “serious business.”

But there is also a fly in the ointment:

· RAM is becoming more expensive at a breakneck pace due to the demand of NVIDIA data centers

· Prices for components are growing faster than the neural networks themselves are getting smarter

· The shortage of microcircuits has not gone away

On Xrust.ru, experts note a paradox: the smarter the AI becomes, the more expensive it is to “feed” it. It turns out that this is a digital version of the law of conservation of energy — nothing is given for free.

Xrust Neural networks have learned to hack the brain: December apocalypse or a new era?