Some of the worst people you know are going to pivot to “See, AI is useful for cancer doctors, that was what I’ve been saying the whole time. Sentient chatbots? I haven’t written those specific words, you must be very bad at reading. Now, lets move on to Quantum!”

The ideas are in general good.

I think the long term cost argument could be strengthen by saying something about DeepSeeks claims to run much cheaper. If there is anything to say about that, I have not kept track.

The ML/LLM split argument might benefit from being beefed up. I saw a funny post on Tumblr (so good luck finding that again) about pigeons being taught to identify cancer cells (a thing, according to the post, I haven’t verified) and how while that is a thing you wouldn’t leap to putting a pigeon in charge of checking CVs and recommending hires. The post was funnier, but it got to the critical point of what statistical relationships reasonably can be used for and what it can’t, which becomes obvious when it is a pigeon instead of a machine. Ah well, you can beef it up in a later post or maybe you intended to link an already existing one. There is a value in being consise instead of rambling like I am doing here.

One author (Daniel) correctly predicted chain-of-thought reasoning, inference scaling, and sweeping chip export controls one year BEFORE ChatGPT existed

Ah, this reminds me of an old book I came across years ago. Printed around 1920 it spent the first half with examples of how the future has been foretold correctly many, many times across history. The author had also made several correct foretellings, among them the Great War. Apparently he tried to warn the Kaiser.

The second half was his visions of the future including a great war…

Unfortunately it was France and Russia invading the Nordic countries in the 1930ies. The Franco-Russian alliance almost got beat thanks to new electric weapons, but then God himself intervened and brought the defenders low because the people had been sining and turning away from Christianity.

An early clue to the author being a bit particular was when he argued that he got his ability to predict the future because he was one quarter Sami, but could still be trusted because he was “3/4 solid Nordic stock”. Best combo apparently and a totally normal way to describe yourself.

I started thinking about when Emma Goldman’s partner Alexander Berkman tried to kill a 19th century robber baron who had sent in Pinkerton to murder workers into ending a strike.

One can make an argument about the economic conditions creating the condition’s for what the anarchists back then called “the propaganda of the deed”. But that isn’t where I am going. Instead let’s look at the aftermath.

From an assassination perspective, the quality of the assassination was lacking. Also, Wikipedia (my bold):

Frick was back at work within a week; Berkman was charged and found guilty of attempted murder. Berkman’s actions in planning the assassination clearly indicated a premeditated intent to kill, and he was sentenced to 22 years in prison.[5] Negative publicity from the attempted assassination resulted in the collapse of the strike.[19]

In other words, today’s robber barons gets less sympathy than the O.G. kind. That’s a bit interesting.

From what I have read, it can be a support as long as:

• It is trained on local data, from the machine and procedures normally used.
• The accuracy is regularly tested (because any variation in the indata, whether from equipment or procedures changes the input data).
• It is understood as a tool that gives suggestions for the radiologist, not a replacement.

Of course, it cannot be better than the best radiologists around. So the question is if it is worth it, compared with for example hire more staff.

The Golem and The Golem at large are two excellent little books about how science and technology actually works. History of science, so heavy on examples (as the historical subjects tend to be) and light on theory. Several examples of what today would be pseudo science but was treated seriously at its time, because they didn’t know what we consider basic knowledge (and you can’t get it from first principle…)

Good for anyone interested in science or technology, but perhaps particularly useful for the cultists (if they can be persuaded).