It's Friday, 27 October, 2023 in Portland, Oregon and I am tired and have a headache and need to look at something other than this monitor.
\nIf you're interested in how technology and the internet has affected work and what work might look like in the future, then you should probably register for Hallway Track 004.
\nI'll be chatting with two of my favorite authors and writers, Joanne McNeil and Tim Maughan and hosting around 25 of you to pretend we're mobbing them after an exceptionally interesting panel.
\nNext Thursday, 2 November, 2023, free and on Zoom at 9am Pacific, 12pm Eastern and 5pm London. Come join us!
\nOK, I'm calling it. Via Metafilter[1], Boston Dynamics glued together:
\nto create a Spot-based robot tour guide[2].
\nThis is all fine and good but what caught my attention is not the gluing-together of things, but exactly how well the things behave when they're glued together. The qualitative glue-result, as it were.
\nFor that, there's a video on YouTube[3].
\nAssuming this isn't an elaborate prank, what really caught my attention is the quality of the generated text and speech output.
\nThe video shows people (most of them Boston Dynamics employees) demoing interactions with these tour guides:
\nThe speech synthesis is off the charts to the extent that I'm a bit suspicious, but on the other hand, ElevenLabs'[12] demos sound super good.
\nListening to the demos and the different personalities is amazing. I think it's the speech synthesis that tips the whole thing out of the uncanny valley and into something different.
\nOne personality stuck out in particular -- "Josh", if only because the Josh examples feel like they have the same sensibility and attitude as TARS and CASE from Nolan's Interstellar. These all feel like, well, if not genuine people personalities[13], then I'm going to call them generative people personalities. The thing about Adams' GPPs was that they were more a commentary on how those with the power to create and deploy those personalities would use them. I am relatively convinced that, somewhere out there, there's a team trying to figure out how to shoehorn in the above types of technologies into making sure smart home windows and doors can deliver surprise and delight, a sort of horrific torment-nexus manifestation of Share and Enjoy.
\nAnd, these systems don't necessarily require the processing power the size of a planet, but seriously: inference on GPT-4 takes a lot, and I have no idea what sort of hardware ElevenLabs' speech synthesis is running on.
\nNow that I'm thinking about it, there's one other thing that feels surprising, or at least was interesting to see, provided there's no sneaky editing in the film: there's totally enough room in the human experience needed to fool people into thinking they're talking with something to cover round-trip latency. Although that said, there's no interrupting in any of the videos. That's still hard.
\nAnyway, it's nice to see that Boston Dynamics had fun with this, if only in the creation of different kinds of personality and that at least one of them is ornery.
\nThis is how my brain works: Taylor Swift's album 1989 (Taylor's Version) was released today, which I think for most people means that the streaming content became available at some timezone's local midnight.
\nApparently Taylor Swift is a big deal!
\nSo I wonder: what's the infrastructure required to support the release of a new Taylor Swift album?
\nHow many copies of 1989 exist at the edge? How many were seeded to the edge ahead of time?
\nSome random Twitter accounts says 1989 (Taylor's Version) has hit over 700 million streams on Spotify. The version I've got downloaded over Apple Music is ~380MB, so naively, if that were streamed and not cached at the client, that's 266 petabytes of Swift, just on Spotify, in the last (as of time of writing) 16 hours.
\nMeanwhile, the Mirror breathlessly and tabloidly reports that Spotify and Apple music were DOWN[14] after the release of the album.
\nI would love to see an internet traffic report, something like "on 27 October, x% of the internet's traffic was Taylor Swift-related".
\nOK, that's it for today, and that's it for this week.
\nHow have you been?
\nBest,
\nDan
\nThings That Caught My Attention is a free newsletter, and if you like it and find it useful, please [consider becoming a paid supporter]({{ upgrade_url }}).
\nDo you have an expense account or a training/research materials budget? Let your boss pay, at $25/month, or $270/year, $35/month, or $380/year, or $50/month, or $500/year.
\nPaid supporters get a free copy of Things That Caught My Attention, Volume 1, collecting the best essays from the first 50 episodes, and free subscribers get a 20% discount.
\nYou have 20 seconds to comply, old sport | MetaFilter (archive.is), Rhaomi, 27 October, 2023, Metafilter ↩︎
\nRobots That Can Chat | Boston Dynamics (archive.is), Boston Dynamics ↩︎
\nMaking Chat (ro)Bots - YouTube (archive.is), Boston Dynamics, 26 October, 2023, YouTube ↩︎
\nFancy Butler @ 0:00 ↩︎
\nPrecious Metal Cowgirl @ 1:39 ↩︎
\nExcited Tour Guide Robot @ 2:36 ↩︎
\n1920s Archaeologist @ 3:50 ↩︎
\nShakespearean Time Traveler @ 4:19 ↩︎
\nTeenage Robot @ 6:02 ↩︎
\nNature Documentary @ 6:42 ↩︎
\nElevenLabs - Generative AI Text to Speech & Voice Cloning (archive.is) ↩︎
\nGenuine People Personalities | Hitchhikers | Fandom (archive.is) ↩︎
\nSpotify and Apple Music DOWN after Taylor Swift's 1989 rerecord release - Mirror Online (archive.is), Zoe Forsey, 27 October, 2023, The Mirror ↩︎
\n