Friday Links: Édith Piaf, AI Pins and the Deep Learning Wall

It’s still Friday somewhere in the world, so at least the title isn’t off yet. This week, we have a hardware and software combo.

• Humane launched its AI Pin. The intro video is a great peek forward into some of the modalities that I think will become commonplace in interacting with AI. The palm display is really unique. Will the company succeed - it’s hard to stay, but they will likely have to stick it out a long time since I’m not sure how many people are willing to dive head-first into usage now + at this price point. Kudos for bringing new things to the table, though. The post-cellphone future is in sight.
• Stability AI’s Head of Audio resigns over copyright concerns. The copyright for the training issue will be in various forms with us for a long time. You have to respect an individual’s choices here. However, I do think we’ll need to come to rest as a society on whether AI systems can learn from the human cultural canon. Having AIs NOT train on your work means you are excluded; having AIs train on it may mean you earn less. My hope is that there’s a solid middle ground where AI can train on everything, but if specific artists are used in prompts, there’d be compensation that goes back to them. There’s already a tremendous amount of precedent for this in the music industry. Related to this is Google DeepMind’s release of Lyria. It does look genuinely amazing. The promo includes a heartfelt statement on creator rights AND a fully auto-generated tune. A bit of dissonance there.
• Édith Piaf’s voice was re-created using AI so she can narrate her own biopic. “When creating the film, we kept asking ourselves, ‘If Édith were still with us, what messages would she want to convey to the younger generations?’”. The singer’s estate unsurprisingly backs this particular production.. perhaps we’ll hear Édith Piaf singing Set Fire to the Rain, You Know I’m no Good, or Who Wants to Live for Forever. Authorizing AI rights seems like a new bonanza for artist estates, so we’ll need to get used to blasts from the past. Walter Cronkite might end up being America’s newscaster for eternity. If you’re in the mood for an Édith Piaf original, try La Foule.
• Deep Learning and the Wall. Gary Marcus calls out Sam Altman and Yanm LeCun for making statements that the current generation of tech has limits. I get that Gary wants to highlight that he “called it” 18 months ago, but this isn’t news. Every computational technique has limits(*). However, I think the article ignores that the advances we’ve seen have taken us to a place where the set of new applications is almost limitless. So much more can be done today than 18 months ago. It’ll keep us busy for a good long while. That we need new architectures for the future is true (people are already working on them), but the application impact of this generation of tech has still to play out.
• EU’s AI Act hits the buffers over foundation models. No doubt some progress was made again this week, but it looks like it’ll be a hard nut to crack. I agree with the idea that Foundation models should not be singled out, and I’m generally in favor of regulating applications of AI, not AI tech. However, it sure looks like when nations have skin in the game, they change their tune.

Wishing you a peaceful and recuperative weekend.