Quote of the Day
The novel superconducting neural chip can process signals at extremely high speeds. It uses superconducting circuitry arranged to mimic the connections found in the human brain.
Georgina Jedikovska
October 19, 2026
US superconducting smart chip analyzes fusion plasma at record speeds
I found this both incredibly awesome and frightening. Reading this a few paragraphs later did not soothe the concerns:
At present, the team is developing a new chip with a neural network 10 times larger than the current design. They plan to present the hardware at BNL in early 2027.
I like living in the future, but I do not recall a time, even during the depths of the cold-war era, where I could see potential imminent existential issues in multiple areas. Maybe it is all the sci-fi books, movies, and shows which require stress to be interesting that biases my view toward the fear side of things. But maybe that same bias will help the people creating these things to be more careful.
Prepare appropriately.
Any surprises I get will be happy. That is the great advantage of being a pessimist. I suppose I miss some opportunities.
“. . . that biases my view toward the fear side of things. But maybe that same bias will help the people creating these things to be more careful.”
I hope that is the case, but it seems that the people building the new thing are more enthusiastic about the building and do not see the downside the same way that people not in the thick of building the new thing see it.
Superconducting LSI is hardly a new thing. There’s a company in Marlboro, MA that has been selling it since the previous century. And cryogenic (though not superconducting) chips are even older, going back to the ETA supercomputers from around 1980.
It was the neural net aspect that caught my eye. This dramatically increases the speed of certain human like capabilities. I was thinking of target detection, identification, acquisition, and weapon use in the submillisecond domain. Existing LLMs are very slow for this type of task.
*Claims* like this are decades old. They’ve never been true in any significance before.
Maybe one of these days, the claims will be true. Maybe it’s even this time…
But I’m not going to hold my breath.
It’s really interesting technology but I’m skeptical of the extrapolation of claims because they seem to carefully avoid any mention of model parameter counts or sample channels or depth. It’s analog inference so a little difficult to compare apples to apples anyway. Their website claims 300x boost in efficiency (per token cost) and an astonishing 60M fps video processing, but again no equivalent model sizes or video-resolutions or accuracy. They claim 10x larger parts next year but don’t say what today’s size is. It smells like a heaping helping of marketing hype around some genuinely innovative tech.