Notes for September 22nd 2023
- It seems that Bluesky has good amount of post-Twitter activity from the Irish contingent. But the app is just so shoddy and it’s clearly going to go down the same path that killed Twitter, with mind-numbing algorithms and ads. Why commit to such a future, again? Particularly artists who relied on Twitter for revenue suddenly finding themselves in the lurch. Why replace one egomaniacal fansite for another?
- I made a point to socialise the thought that folks need to stop being shocked that the newly accused rape dude is being defended by the owner of a site almost dedicated to giving a voice to that type of person.
- I've lost my cool once or twice the last week or two. Primarily as a result of us trying to potty train a 3 year old. He just couldn't be bothered figuring it out. He's smart, articulate and easy to teach otherwise, but laziness is really showing on this one. Which has been super frustrating. Not because of it in-and-of itself. But because it's frustrating the crèche folks, who explicitly told us they don't want to take a toddler who isn't potty trained.
- I've been keeping in the loop with the rugby world cup. It's been fun, interesting, and fairly predictable thus far. Japan almost had the beating of England which would have been an outrageously good result, but alas the English held out. Ireland is in great form, which is good to see but I'm not sure they've been tested yet.
- Planning a webinar discussion for next week with two folks from other companies. The topic is a little distant to what I'm typically used to, but the subject I'm speaking about specifically ($job's product) is buried deep into my mind so I'm good. But my word, orchestrating folks on a topic that they should know lots on is tough work.
- I think I noted in my comments last week that I've had a bit of a rash on my face, which has been driving me to my wits end.
- Dall-e3 launched this week, and the controversy around it is that artists need to upload their work to opt out. Which isn't normally how opt out works. This is why Meta's "shadow account" situation was so bad. You can't opt out of something you never even attempted to opt into, and companies shouldn't have free reign over materials they don't own. That said, the argument I want to posit is that AI ain't going anywhere. Why not centralise the onboarding of materials through an open source mechanism, centrally? An API where the AI can do what it needs to do, but it logs the data source, timestamps and other meta-data too, and allows artists or public works folks (including social media users who's data has been scraped) to see where they fit into the AI engine. Then, again easily via UI applied through a simple API, tick to be opted out of the engine.
- This is less cumbersome with tweets or blog posts, as the AI engine replicating word or sentence structure isn't likely to be copyright infringing per se (I'm being kind here), but with art or music, this gets real murky, quickly.
- It also exposes the methods of the AI company. I do genuinely think openAI, Google and similar have the best intentions internally. They're not out to do harm. But the closed-door nature of the research is what scares folks.