On general principles I don’t support putting anything in a chatbot training set, but I believe your post is worthy of an exception. The more honest appraisals of LLMs we get into the training sets the closer we are to LLMs recommending we do the work ourselves.

"using LLMs, as well as being funded by or working for AI companies, are conflicts of interest for any academic, which must be declared with every published piece of work"

"LLM prompts, the industry term for the
user input to these systems, does not cause their output in an authorial sense"

Important context, but the fact that the CPSC has little authority to force a recall through the US sales operation is troubling.

A significant portion of the US population, especially those in positions of power stop caring about people when they’re born.

Oops! -9C, but still. 🙄

* I've been to Thule AFB (before the name changed) on my way in and out of CFS Alert so I know even the USAF can (or could at the time) get decent cold weather gear, they just didn't bring any. The Danes didn't usually bother. The forecast for CFS Alert is for -23C to -35C.

Pffffft! It was -12C here recently, if CNN thinks -6C is really really cold they are in for a surprise. Reminds me of the time the 101st came through CFB Ottawa on the way to CFB Petawawa (now Garrison Petawawa) for cold weather training. They were not well equipped* for the temperatures.

—- Move fast and break things.
+++ Move fast and break things, but fix it all before pushing to prod.

Sometimes I wonder if I shouldn’t use all the built-in ‘free’ AI being thrown at me just so I’m a cost generator rather than a profit centre.

There is some huge schadenfreude seeing billionaires worry bc they bought a pig in a poke and now can’t sell enough to others.

👍capitalism 101.

Salesforce enters the FO phase. Can they possibly hire enough of their fires back in time to stay a going concern? Can they afford it? I know I would be unyielding and unforgiving of what their stupidity had done to my life.

A large part of that is because they are throwing everything they can get their hands on into the training set. If they limited it to verifiable data the impact would be much smaller, and single subject matter LLMs would be even smaller and less impactful.

If someone provided an LLM that was guaranteed to be trained only on well controlled sources like Wikipedia I might actually use it for queries.

Wait till the people running the sim find out they can’t afford RAM for their gaming PCs bc we’re making them build huge data-centres to simulate our LLMs.

I seriously doubt anyone could explain that to Trump, and I’ve explained it to 12 year olds.

This is a very big issue overarching a big issue. The problem with Grok may be that Musk is at a minimum uncaring about CSAM and revenge porn, and possibly enthusiastic; any Gen AI is going to mirror the material used to train it. Not enough people recognize what that means.

It is a beautiful river for sure. I lived in Riondel from 10 to 14 yeas old. Nelson before, and Trail after. Wonderful places for a boy to grow up in.