If those leaders believe in the integrity of the scientific institutions, their flock will follow. If they're anti-vax, their flock will follow. If they believe in some medical quackery, their flock will follow. If they believe in eugenics, their flock will follow. It's happened before.

What was fringe yesterday can become mainstream today, with the right leaders.

There is no scientific breakthrough that has occurred sans politics. Politics choose the winners and the losers, and the realm is science is no exception.

All science is political, because the scientific institutions are made up of people, who are political. Your research project lives and dies by politics, as does your dissertation, who gets published, who receives awards, etc.

So when it comes to research of limited utility that has a nasty cadre waiting in the wings to pounce upon it, the wise person would think twice.

The Farpoint is a case in point. I have an older one where the mini-backpack actually zipped onto the bigger one, and had a proper lifting handle integrated for lifting when it's lying flat (among other niceties). I accidentally left butter in it, and the smell was so bad that no amount of cleaning would expunge it. I just bought another one.

Absolute SHIT quality compared to my old one, the mini backpack now buckled rather pathetically to the big one, and no more solid handles (except the top one). Structural integrity is WEAK.

Needless to say I spent a LOT of time and effort cleaning up my old Farpoint, which I'm now using again. The "new" one? Sitting in storage along with the rest of my buyer's remorse.

That's a lot less work than putting an extra validation step into the publishing pipeline. And with sane defaults it lets the user make an informed decision when special circumstances arise.

Opus is already quite capable of zero-day exploits, and so it was inevitable that the next generation of LLMs would be eating them for breakfast.

This is a defensive move because Anthropic was first across the line. As soon as hostile states gain access to this tech, there will be a veritable tsunami of pwnings across the globe, as everyone wielding such power battles for the compute power and allegiance of every processing device on the planet.

It'll be a race of exploit, secret patch, exploit, secret patch as these AIs battle each other, with territory marked by the machines under their control.

And you, Joe Average, won't even see it happening, because your computers and phones and cars and TVs will tell you everything's fine while they do battle in secret.

https://www.biblegateway.com/passage/?search=1%20Samuel%2010...

https://www.biblegateway.com/passage/?search=Proverbs%2016%3...

https://www.biblegateway.com/passage/?search=Leviticus%2016%...

https://en.wikipedia.org/wiki/Cleromancy

https://en.wikipedia.org/wiki/Tyche

That's a stretch. Most early "gambling" was a way of putting the choice to the gods.

The above example doesn't specify workdir mounting mode, so it would be copy, not overlay.

This does the same thing as in the blog post, except that there are a LOT of gotchas and minutiae and some yak shaving involved if you want to keep doing it manually.

I've gone through the whole path the author has, and finally had to admit that it's too much fiddling around to do it manually. Easier to just have a cmdline tool that does it for you. That's why I built it in the first place.

    # Create a new sandbox copying . as workdir (default container, but you can choose vm)
    yoloai new mybugfix . --isolation vm

    # attach to it (it has tmux already)
    yoloai attach mybugfix

    # Chat with the bot inside...

    # Happy with its work? Diff it to be sure
    yoloai diff mybugfix

    # Happy with the changes? Apply them to your workdir
    yoloai apply mybugfix

    # All done? Destroy the sandbox
    yoloai destroy mybugfix

https://github.com/kstenerud/yoloai

https://github.com/kstenerud/yoloai