Update 2: GPT-3, DALL-E-2, Product DAOs, Data NFTs & More
I’ve initially kept the format of this newsletter as a review on the past few weeks. At the moment it’s a longer-form status update to my friends or whoever might be interested and a chance to dump all the thoughts I’ve been thinking into a semi-structured format.
In this way, it’s been really helpful to look back at the last little bit and see what I’ve done, and also look forward and think about what I actually want to spend time on. This reflection adds some structured to what’s been generally a fairly unstructured, explorative time.
I expect that in the coming weeks I may shift from this format to more full-featured posts on a single topic at a time. But would love to hear from y’all if you like getting these little more scattered updates too :)
Cold Water Immersion & Sauna.
I’m still in Aarhus, Denmark, and for the past 10 days I’ve started my days off jumping into the ice cold ocean every morning followed by a meditation in the sauna. At first it was 10 seconds in the water and an immediate “NOPE” out of there. Now I’m hanging out for 5 - 10 min, and I can tell it’s going to be a practice I really miss heading back to New York. The water is also filled with Jellyfish that thankfully haven’t been the stingy kind. Here’s a photo of the dock I’ve been swimming in, edited by Dall-E 2 to add “A psychedlic jellyfish in the sky” and a “Humanoid-Jellyfish Cyborg about to jump in”
Teaching my Nephew to Code. I got to spend some time over the past two weekends teaching my nephew to code. We coded a tic-tac-toe game together in Python, and it was really fun to see how much joy he took out of “debugging” and fixing errors we hit.
Back to NYC. Gearing up for the summer in New York, in part a bit sad to be leaving Denmark, but excited to be home.
Building & Writing
There’s been two big areas of focus over the last couple weeks: web3 & building on top of OpenAI’s GPT-3 & Dall-E 2
GPT-3 & Dall-E 2 - AI Everywhere
I went to an OpenAI event the other week, and based on the demos and the upcoming capabilities, I got a sense that language and visual models like GPT-3 & Dall-E 2 are going to become widely embedded into products, and will be quite transformative. I
tweeted a bit tweeted a lot about how Dall-E 2 will shift industries and our experience of media, as we become deluged in AI assistants and AI generated or edited media. I came across a nice directory, GPT-3 Demos, that showcases GPT-3 powered tools
One of the areas that I got excited excited about during the OpenAI event is realizing that you could likely use GPT-3 to extract structured data from text in a really easy way. I built and launched a service that lets you scrape any website and pull out structured data just by describing the data you want - check out the video below.
I’m hoping to spend some time to open up access more broadly and also put up a few examples powered by this service.
I found one other tool called Elicit (elicit.org), that’s using a similar GPT-3 based data-extraction approach, to build a GPT-3 Research assistant.
It’s built by a non-profit called Ought, it’s pretty cool to see it pull out additional structured data from the papers like a summary of the abstract, what methods were used, the size of the population in the study, characteristics of the population etc. - all of which become filterable, searchable and sortable facets of the data. I think it’s a good example of the promise of this sort of technique.
More on GPT-3 & Dall-E 2
Dall-E 2, the OpenAI image generator, is addictive. You just give it any description and it’ll generate an image - I do think it opens up a lot of possibilities, and essentially agree with this sentiment:
I find myself wanting to play with different ideas each day. Using it to add fun images to my journal, or to attach a generated picture to a joke I’m making to a friend. Earlier this week I used it to generate currencies from imagined worlds where Jeff Bezos or Elon Musk are on the currency. I also find myself using it to create images for any blog-post or tweet I’m making, like the edited photo with the jellyfish above, or the header image for the blog post I included below.
The API is not available yet for exploration on the Dall-E 2 side, but the GPT-3 one is widely available. One thing that gets me excited is thinking about what ambient, embeddable versions of these technologies inside products will look like.
I think there’s also whole slew of low-code tools to be built around training and using GPT-3. For example, the OpenAI playground right now is quite limited, and as more people look to tap into this functionality, I could imagine there being drag & drop tools that make it easy to upload sets of documents to train models on and more.
Web3 “What If” — Disruptive Innovation & Flippenings .
I took some time to publish a piece that I drafted in January trying to project out the trends of web3 and see what might surprise us in the long-term:
Web3 "What If" -- Disruptive Innovation & Flippenings
Re-Imagining Block Explorers & Visualizing On-Chain Data.
As I discussed in my first newsletter post, I think that the state of understanding and looking at on-chain data leaves a lot to be desired. Right now, the standard Block Explorer is Etherescan, and people also use tools like Dune Analytics & Flipside Crypto to get insights and try to do more sophisticated queries. I think there’s room for a much more well-crafted, and intuitive block-explorer that brings in the sort of more sophisticated SQL style querying and visualization functionality that Dune offers as well.
I’ve been mainly using this theme as a chance to get a much better sense of the EVM internals, and working with on-chain data.
I’ve learned a fair bit about how Etherscan and Dune are built, and what it’s like currently interacting with on-chain data. Interestingly, one reason on-chain data is tricky to work with is because it’s encrypted, and you need the “ABIs” of contracts in order to decrypt the data - Etherscan is the main (only?) place where people put these ABIs up for others to access.
Both Dune & Flipside crypto’s main offering is loading the data into Postgres, decrypting it based on ABIs that Etherscan has, and then using an open-source tool called Redash that allows for SQL querying & visualization platform to give easy access to it. Because on-chain data is so hard to parse, they also do the work to create higher-level decoded views of the data.
I’ve got a barebones prototype block explorer working now. My goal would be to put up a simple version to start getting feedback from folks like you!
Alongside this prototype I’ll put up a post on designing & building this type of tool, and where I see the space going from the design side in the next while.
One fun idea that came out of my work on ScrapeMagic is thinking about what it would look like to be able to run a side-project like ScrapeMagic as a DAO. I’ve been referring to this in my head as a “Product DAO”
For many software products, to make them get adoption there’s often a whole slew of things that needs to get done include marketing, sales, onboarding, customer support, and more. For a project like ScrapeMagic that I don’t intend to run full-time, it’s almost like: why even bother to put it up if you’re not able to do all the other things needed to get it to “operate”?
One of the issues with companies is that it’s hard to easily allow for freelancers, or other contributors to jump in and collaborate, and be able to incentivize them with a “share” of the company. In their currently amorphous legal status, DAOs with their corresponding tokens allow for a mechanism to give contributors compensation for chipping in around specific tasks. For example, I’d love to give some tokens to someone who wants to go around marketing, selling and onboarding folks to ScrapeMagic since I might not want to spend my time doing that. Similarly I’d love to have other engineers and designers chip in around specific tasks, in exchange for a share of any future revenues we collectively generate.
This idea’s been tried before in the pre-DAO era. I remember in 2014 Union Square Ventures backed a company called Assembly: https://www.usv.com/writing/2014/11/assembly/ that was “a platform that pairs developers, designers, and other creatives with great ideas to collaborate on the development and launch of software businesses and applications. […] Contributors are stakeholders who own their fair share of the products, and the revenues they generate.”.
It didn’t work out in the pre-DAO world - you can read this HN tear down: https://news.ycombinator.com/item?id=10555710. One question we can ask is: do DAOs and the advances in collaborative, remote tooling change what sort of organizations can thrive and build great software?
Perhaps a different version of this question though could be - will most organizations also have a DAO component that lets them more fluidly incentivize and allow collaboration from external or more passive contributors?
I took on a fun design project with Spring Discovery, a new ML & technology focused pharmaceutical drug discovery platform to help them visualize their data. It’s been really fun working with their team to think about how to communicate narratives from their data in intuitive, and interactive ways. Here’s a little screenshot of some of my Figma board for the project'
Trying to draw and animate some of the more advanced interactions in Figma led to fun out-of-memory errors I hadn’t seen before:
Things I found interesting
The DAO Tooling space continues to grow. Two products that I’m excited about in this realm that just announced their funding are Dework and Utopia Labs
Dework is an open, Linear style work management platform designed for DAOs to post issues for people to work on, take applications, and award bounties to contributors.
Utopia Labs is the Gusto for DAOs, enabling them to pay their contributors and be compliant amidst the shifting regulations.
Mirror’s Writing NFT launch was dope. Now every piece on Mirror is a collectible NFT, which I translate to as: now every single piece of writing has a tip-jar that issues a collectible, and shows a readers support for the authors writing. Similar to how Substack enabled many more people to write and create small businesses through their writing alone, I think this aspect of Mirror will allow more people to find funding for their writing.
Things to Dive into Next Few Weeks
There’s a few areas I’m hoping to continue spending time on these next few weeks:
Deep dive into DAOs & DAO Tooling
I’ve chatted with a few folks who are very skeptical about DAOs, and DAO tooling projects in general. With the market turning “bear-ish” it’ll be interesting to see how the space evolves.
For me, DAOs are one of the areas in web3 that I’m most excited about the possibilities of. I want to take some time to share some structured thoughts around why I believe that DAOs are not a scam, and where the exciting opportunities for DAOs lie.
Open Data DAO, Collaborative Databases & Data NFTs.
I’ve been thinking a fair bit about what new primitives for collaboratively creating data sets might look like. Along the lines of my interest in DAOs, one area I’m thinking about is building a DAO that is focused on assembling open data sets, that would be available as public goods for developers to build applications on top of.
What sorts of data sets do people pay for and want access to right now? Here’s just a few quick examples: Company Data, Place / Location Data, Nutritional Information Databases, Bio Sequencing Data.
Right now these data sets are typically assembled by centralized players, who make the data available as APIs, typically charging based on the amount of usage. Developers looking to build applications need to pay for access at prices that are often prohibitive for many ideas they might want to try out. I think turning these data sets into public-goods would open up the ability to more easily build new applications in these categories.
One of the mechanisms that I’ve started thinking about to enable this is what I’m calling a Data NFT, a way to incentivize people to contribute / generate data, in exchange for future rewards from people looking to access that data down the road.
On-chain Websites & Link NFTs
I’m hoping to spend this week to rebuild my personal website as a smart contract. Along the lines of the Web3 “What Ifs” I started imagining what would be the benefits if we were to store everything on-chain, and if every website, and every link on a website was actually an NFT, stored on-chain. I think a lot of interesting stuff comes out of this - but more on that soon :)
Thanks for reading & sending y’all wishes for a good week!