...The debate should not be about whether there will be autonomous weapons, but about how the use of AI will be secured. The problem is that AI is a black box, and AI creators don't invest enough in understanding how it works. This means that to secure AI, complex systems need to be built to check inputs and outputs, adding a lot of overhead, and that a possible scenario would be to be killed by a highly educated trainee, but who can sometimes go off the rails.

Reminds me of my demomakers days on the Amiga, when we were hacking the hardware to create fancy FX.

From the great "Dark Territory" book, by Fred Kaplan, after Regan viewed the movie: Reagan turned to General John Vessey, the chairman of the Joint Chiefs, the U.S. military’s top officer, and asked, “Could something like this really happen?” Could someone break into our most sensitive computers? Vessey, who’d grown accustomed to such queries, said he would look into it. One week later, the general came back to the White House with his answer. WarGames, it turned out, wasn’t at all far-fetched. “Mr. President,” he said, “the problem is much worse than you think.” Reagan’s question set off a string of interagency memos, working groups, studies, and meetings, which culminated, fifteen months later, in a confidential national security decision directive, NSDD-145, signed September 17, 1984, titled “National Policy on Telecommunications and Automated Information Systems Security.“

Since the hype around the leak of LLaMA has happened because some great guys made it run without even requiring a GPU, it will be interesting to see if Meta learned something from this, first of all that 4-bit quantization seems the produce good results...

What a surprise...

Maybe the news here is the number of dimensions. Word embedding implies vectors with a much lower number of dimensions (in his book, François Chollet tells us about 256-, 512- or 1024-dimensional vectors), which is far less than hot encoding requires. Also, considering the huge number of dimensions, perhap's the maths in this space would be more related to semantics ? In their article about word2vec, the authors write that it was a surprise and it may now be intented ? : "Somewhat surprisingly, many of these patterns can be represented as linear translations. For example, the result of a vector calculation vec(“Madrid”) - vec(“Spain”) + vec(“France”) is closer to vec(“Paris”) than to any other word vector" (in "Distributed Representations of Words and Phrases and their Compositionality").

This is epistemology, not philosophy, to be accurate. This is when scientists ponder about science. Useful, as long as you take great care not to fall into the bottomless pit of deconstrustivism, or the hell of relativism.

It stil is is a good computer :) The documentary shows the famous cover for Deluxe Paint 2 by the great Avril Harrison, and we just released a "cracktro" that features a new version of King Tut as a tribute to such great graphic artists : https://www.pouet.net/prod.php... Coding the hardware of the Amiga in assembly language is still fun in 2018, because of the various coprocessors that work in parallel with the CPU. It is a very interesting machine. The documentary is nice. Those who are interested in the adventure of the Amiga may read the great book from Brian Bagnall ("Commodore: The Amiga Years").