Five Things on a Friday: Who Made Who?

Let's start with a banger...

1. This Week's AFF Army Football Show: AD Tom Theodorakis (As For Football)

After a tough loss last week, I was really happy with how this week's AFF Army Football Show went. I felt like we did a decent job of not wallowing in the loss, and more to the point, we turned the page pretty quickly afterwards to give fans a reason to stay engaged. Ahead of the show, I wasn't sure how well that would work, but afterward, I have to admit I felt like we did a good job.

2. The AI Inference Revolution Is Here (IEEE Spectrum)

Friends, I'm not gonna tell you how to feel about AI, but as a matter of reality, it's not going anywhere. In which case, it behooves us all as informed adults to at least understand the basic underlying technology. 

The bottom half of the article gets into a lot of chip design stuff, which is less useful to regular people, but the top-half is really a must-read for those who want to live in a world they understand rather than in a world that is strange, scary, and mysterious.

An untrained LLM is like a jumble of Scrabble tiles on a table. Instead of single letters, though, the tiles show fragments of words, called tokens. Everything you’d need to write almost anything is present, but nothing makes sense.

Training a model organizes this jumble using a guessing game played at scale. The model is shown real text with the next token hidden and asked to predict what comes next. After each guess, the correct token is revealed and then compared to the prediction, and the difference is used to calculate the model’s accuracy. The game is played not with a single sentence but over billions of passages.

While a real game of Scrabble can be played over a bag of chips and a few drinks, AI training is computationally intense. The model updates its parameters through backpropagation, a process that repeatedly calculates how each of a model’s billions or trillions of parameters should shift to make the next prediction better. This is why tech giants are building larger data centers than ever before.

It takes a lot of memory to do the next step, called "inference," which ties the statistical predictive components of the model in with what's already come before. The basic algorithm will makes sense at a micro level just because the model is trained. However, the models are only useful if they can remember what's come before. That's where this next step becomes important.

3. Ride On by Monica Valli

I'd never heard of Valli before she popped into my Instagram feed this week, but this song is a good one.

If you're wondering, they're calling Valli a blues singer.

4. Gun Girl Gracie

I copy-edited The Return of Dr. Necropolis this week and then started working on the rewrite of Gun Girl Gracie. I'll be looking for a few test-readers soon, if you're interested. In the meantime, here's a working sketch of the cover concept, based on a picture that a friend of mine took at a Panamanian beach.

5. Microsoft exec called AI scraping the “largest theft of labor in human history” (Ars Technica)

Microsoft Director of Applied Science Brent Hecht repeatedly warned in documents that scraping news for AI training was “an astonishing theft of unprecedented proportions,” calling it perhaps the “largest theft of labor in human history,” news orgs said. In another document, Hecht contradicted Microsoft and OpenAI’s argument that training AI on news content is fair use, suggesting that the plan to widely scrape news made “a complete mockery of the idea of ‘fair use.’”

Over at OpenAI, ChatGPT head Nick Turley wrote in an internal message that publishers would face an “existential threat” from commercial products trained on news content that can be used to substitute news providers. One Microsoft document even described a “doom loop,” news orgs said, “that will hurt the performance of our models and the entire web at the same time.”

I'm including this to show that these guys 100% understand the difference between the basic technology  itself and its unethical real-world use. We made this same point on the AFF show we did about AI, but it feels like the nuance -- which is only nuanced in an insane world! -- gets lost a lot of time.

Anyway. We are where we are. But this is gonna blow up on these guys sooner or later. It's the cigarette lawsuits all over again.

* * *

That's all I've got this week, friends. If you're interested in test-reading Gun Girl Gracie, let me know. Or just sign up for the Dr. Necropolis mailing list. I'll probably send out a new email late next week.

Enjoy the weekend!