Inside Today's Meteor
- Disrupt: Introducing the Personal AI Teleprompter
- Create: Objkt marketplace is a great place to find Tezos artists
- Compress: Sotheby's cancels "Natively Digital: Glitch-ism" NFT auction
- Cool Tools: What if LinkedIn Had an AI Search Bar?
Introducing the Personal AI Teleprompter
We just saw a demo of an AI-powered teleprompter feeding lines to a faux job applicant over a face mounted augmented reality viewer. Has the Cyrano singularity just arrived?
It's just a hackathon project for now, concocted by a team of Stanford undergrads and cobbled together, but as we've noted before, this path points the way to the future of computer UI/UX, and it has nothing to do with a keyboard or a mouse. Our bet: AR speech-to-text interfaces are going to revolutionize how we interact with computers in the next few years.
Large Language Models currently operate primarily in text environments, but they're quickly jumping over to voice interactions, for example, via telephone agents that can re-negotiate a bill, or record a spam caller and file a legal claim on your behalf using text to speech AI.
We've had Siri and Google Voice and other conversational agents for years, but they haven't really worked very well before now.
Almost immediately after ChatGPT launched, a lawyer announced plans to send it to court through ear pods to guide a defendant through a traffic ticket appearance (he called it off when the local Bar Association threatened to revoke his law license if he followed through on the stunt).
Stanford University undergrads Bryan Hau-Ping Chiang, Varun Shenoy, and Alix Cui have taken that a leap forward with rizzGPT, a "Charisma-as-a-Service" (CAAS) conversational helper that can be put to use in face-to-face interactions. It flashes AI-generated text onto an AR "Monocle" attachment that lets the wearer read them to land that perfect job interview, or get a second date.
Monocle is a real hardware device created by Brilliant Labs, a nonprofit that develops open source solutions for making the Web more accessible, among other things. rizzGPT uses Bluetooth to wirelessly connect to a mobile phone to deliver the conversational prompts using OpenAI's GPT-3 language model via an API to the Monocle device. The on-board battery can only run for about one hour, so it's early days.
Brilliant Labs is currently working on an upgrade, dubbed Frames, that houses all of the electronics in what look like a normal pair of eyeglasses.
There are probably dozens of use cases for heads up conversational prompts – give your spouse the perfect compliment, beat someone in an argument, negotiate a raise, talk the police so you don't get arrested. Of course, this is the Internet, so sex is going to figure into this somehow.
We've already seen a "conversational companion" AI go off the rails into sexting and ensuing broken hearts when the company behind it, Replika, cracked down and ended the steamy TOS breaking sessions.
In the case of rizzGPT, we thought of something related but a bit different: the tale of Cyrano de Bergerac, from the play of the same name (and the movie with Gerard Depardieu). In that story, a brilliant poet writes letters and whispers couplets from the bushes to help the handsome but tongue-tied Christian make his advances to the beautiful Roxanne.
Could ChatGPT make a real Roxanne fall in love with it? Maybe we're about to find out.
We caught up with Bryan Hau-Ping Chiang to learn more about what the team is making, the core challenges, and how they expect to see it applied in the real world.
"This was actually for a hackathon project, " he said. "I'd been building much bulkier 3D headsets for quite a while. And my friend kind of came up with this gadget called Monocle, very portable. We took it to a hackathon. There was a place in Healdsburg called AGI house and we just spent a few hours gluing everything together and decided to film the demo."
"But I think the goal though, if you kind of extrapolate out, you really have this sort of vision where, you know, this is the future, right? Like why do we need a mouse in keyboard and then a phone?
"LLMs really make possible these sort of interfaces where you just talk to your devices. And you're not gonna be tapping on buttons like a Neanderthal. You're just gonna tell your agents what to do, and they're gonna do it for you. And I think that's sort of the future, whether we make it or whether Apple makes it or whatever."
Update: After we went to publication, Varun Shenoy sent us answers to some emailed questions.
What inspired you to come up with this? I walked into the hackathon on the day of and saw Alix and Bryan playing with a Monocle. They're fun people and they already had the idea in mind so I was more than happy to work with them.
How do you see it being used? I don't think this is the actual use case for a tool like this. What would actually be immensely helpful is to have information about the person you're speaking with shown up on a heads-up display.
For example, if I'm having an interview with someone from Microsoft, having information about the recent keynote, the team that they're on, and people they're connected to would be much more useful than a device that is telling me exactly what to say. Humans are great synthesizers, but not nearly as great at remembering details about people. In a way, remembering those details is something charismatic people are extremely good at.
How close to a real product is this? Not too far. We would want to definitely shrink down the Monocle and get it into a more compact form-factor. It's almost comical to wear in public (but also kind of cyberpunk and cool). I also think if some big tech company built an impressive pair of consumer-grade AR glasses, this would be an awesome application that could run on it.
Any downsides? A few things: the aforementioned size, the latency between API calls, and privacy. To clarify the latter two, when using this hack, your conversations (in the current state) are being streamed to an OpenAI server twice, first for transcription, and then again for generating a response. Ideally, this will all be done locally in the future. Lots of strides are currently being done in scaling down large language models. Wearables plus on-device LLMs could make for a very fun future.
Usually, we showcase artists that publish their NFTs on the Ethereum blockchain. Today, we're mixing it up by featuring artists that publish on the Tezos chain. For those jumping into NFT art, the Objkt marketplace is a great place to find Tezos artists, which tend to be less established and less expensive. But that doesn't mean the quality is any less. As you'll see below.
Last month, FAIC, an AI fashion collective curated dozens of artists as part of AI Fashion Week. These were some of our favorites from their collection (starting from top left, listed clockwise).
"The Liberation of Consciousness" by Junk Culture on Objkt.
"Perpetuity" by Max Drekker on Objkt.
More Than A Glitch
Auction house Sotheby's canceled its "Natively Digital: Glitch-ism" NFT auction on Sunday, after one of the artists in the show called out the lack of female-identifying creators.
Don't Bank on the Banks
Another day, another wrinkle in the banking mess we've been following now since the collapse of SVB: Investor legend Tim Draper now recommends businesses diversify cash sufficient to meet payroll for 6 months each across two banks, one local and one global, and hold two months worth in Bitcoin or other crypto currencies.
Draper is a Bitcoin whale, he got 30,000 BTC in the US Marshall's auction of Silk Road assets in 2014, and he predicted a $250,000 BTC price by end of 2022, which obviously did not happen. Unlike a fantastical bet set up to highlight the situation and encourage a flight to crypto as a safe haven from hyperinflation of the greenback, he says it's just a prudent to do.
The Jobs, They Are A-Changing
Generative imaging AI has transformed the work of 3D creation overnight, as a Redditor laments below, from making artwork to writing text prompts for MidJourney, Stable Diffusion and other tools that do all the rendering. That was fast.
Thrown Away With the Wrapper
An (un)lucky CryptoPunks collector wiped out about $135,000 by sending his #685 Punk to a dead end address where it cannot be recovered. The owner was attempting to "wrap" the NFT to bring it up to date with the Ethereum ERC-721 standard, but wound up sending it to a dead address instead of his own wallet by mistake.
Be careful out there.
What if LinkedIn Had an AI Search Bar?
Noon.ai claims they have a better way to find talent.