The Metaverse's Environmental Cost - Tech Insiders News, November 2022

A curated collection of «signal over noise» links about interactive technologies, AI, web3, sense-making, entrepreneurship, and psychology.
Today’s image generated by Midjourney, prompt: «Alignment is very difficult for humans, and it's difficult for AI as well!» (Inspired by the links below.) Full image at the bottom of the article.
- It’s easier to adjust the course of something before it happens, right? This is the hopeful message of Amelia Kallman in her TEDx talk «The Metaverse & the Environment».
In short: the increased usage of Virtual Worlds is creating more and more data, and the move to streaming will also increase data center usage. We will go that way anyway; it’s now up to us to address this issue… Will we be able to? Remember, coordination is complex, except when it’s for profit or power. - Meta has showcased CICERO, an AI that can achieve human-level performance in Diplomacy, a famous board game. It’s impressive… Here’s my take:
One way of reacting to that is «Oh my god AI now knows how to do politics and it’s very bad please stop it or it’s going to kill us». I partially agree; AI has indeed a very high risk of turning badly…
…but that’s not it yet. When presented with an AI solving a game, focus on the game mechanics, not its theme. CICERO has solved the specific problems of the game Diplomacy, not (yet?) all the complex political games. - In another AI news, Meta released Galactica, a generative AI that was trained on millions of research papers; in the hope of «organizing science»… and shut it down in 2 days.
The main cause is that it was mostly spitting out wrong information. Once again, this shows that it’s very difficult to align AI to our needs. - It’s thanksgiving season, so a couple links for you:
Did you know that you can actually have a great impact by donating to charities? I’ve myself taken the Giving What We Can pledge to donate 10% of my income until I retire.
And Underrated reasons to be thankful, a very funny read by .
And today’s image, thanks to Midjourney: «Alignment is very difficult for humans, and it's difficult for AI as well!».
It’s looking very nice. It made me realize why AI Art generation acceptance is easier: even if the result is not exactly what I had in mind (How could it be?), it’s still nice and I like it. For models that needs to be exact (like Galactica), it’s much harder…