Kurzweil signs on with the brain-computer startup you inhale
Two very different bets on where the interface goes next: Ray Kurzweil has joined a startup trying to skip the skull entirely, and Lenovo just shipped a laptop that keeps frontier-scale models off the cloud.
Ray Kurzweil is joining Subsense, a Silicon Valley startup developing a non-surgical brain-computer interface you take like a nasal spray, according to Fortune. Subsense's approach introduces charged nanoparticles into the brain intranasally — the company likens the delivery to inhaling allergy medicine — and then reads or stimulates neurons from outside the skull using a cap fitted with magnetic coils. It has raised $27 million and is still at the preclinical stage, running experiments in mice and showing off a nonfunctional mock-up of the headwear at an event this week. Kurzweil, who still holds a research role at Google, framed the appeal in familiar terms: external devices like phones are a barrier between you and the AI you're trying to use.
The pitch is a direct rebuke to Neuralink's model — no drilling, no electrodes, no surgical theatre — but "non-surgical" is doing careful work in that sentence. Getting foreign nanoparticles across the blood-brain barrier and back out again is a much harder regulatory problem than installing a chip, and a consumer version is years away at best, if it survives the lab at all. What it does signal is that BCI funding is drifting toward anything that removes the operating room: the bottleneck was never the electrodes, it was the craniotomy.
Lenovo used IFA in Berlin to put a 120-billion-parameter model inside a 1.65-kilogram laptop. The Yoga Pro 9n carries Nvidia's RTX Spark superchip — a 20-core Grace CPU fused to a Blackwell GPU with 6,144 CUDA cores over NVLink-C2C, rated at 1 PFLOPS of FP4 AI performance — with up to 128GB of unified memory that both the CPU and the GPU can address at once. That pool is the point: on a conventional laptop, a model's weights get shuffled between system RAM and VRAM, and VRAM size sets the ceiling on what you can run. Nvidia's own reference figure is a 120B-parameter model at 1M tokens of context, locally, on a 15.3-inch machine 16.7mm thick. A 2-in-1 sibling, the Yoga 9n, tops out at 64GB.
This is the cloud-cost argument showing up in hardware form. Lenovo also extended its Qira assistant to Motorola phones, tablets and a new LTE watch, and showed a 1.6-litre ThinkCentre that can be clustered four-deep for bigger local models — the same "why is this token bill so high" logic that's been pushing enterprises toward device-side inference. No pricing on the Yoga Pro 9n has been disclosed yet, which, with memory prices where they are, is its own kind of answer.
South Korea's AI stock rally has started rewriting the country's dating pool and its university intake. CNBC reports that matchmaking firm Gayeon is seeing sharply higher demand for partners at SK Hynix and Samsung Electronics, where bonuses tied to the memory boom have run to roughly 700 million won per employee at SK Hynix — more than 15 times the average Korean salary. A YouTube dating show matching Samsung and SK Hynix staff aired in August and pulled about 650,000 views; Netflix released a retail-trading variety show last month. Meanwhile nine in ten undeclared engineering students at Seoul National University chose electrical and computer engineering this year, up from two in ten in 2023.
If a founder offered you a nasal-spray interface and a surgeon's drill, which one would you actually say yes to — and does "non-surgical" make it sound safer than it is? Tell us in the comments.
