Catalyzing Education, Inspiration, and Iteration
Web3.com Ventures Original Research Analysis
0xFishylosopher
Introduction
Like many others, my mind was absolutely blown away by ChatGPT’s capabilities. Though I had played with GPT-3’s playground API before, and have written about newer image models such as DALLE 2 before, I was not prepared for the breadth, depth, and fluidity of GPT-3’s responses. From sonnets to Solidity, philosophy to physics, the quality of the content was simply stunning.
Still in its infancy, ChatGPT is undoubtedly a potentially disruptive force to every industry, including Web3. In this essay, I will endeavor to outline some of the ways I believe ChatGPT will accelerate the industry’s evolution, catalyzing the processes of education, inspiration, and iteration in Web3.
Education
First, AI systems like ChatGPT are incredibly useful in educating and helping users navigate the vast world of Web 3. As an industry that melds together cutting-edge technology with opaque financial models, Web 3 has a particularly pronounced education problem. To make a single transaction on a decentralized exchange (DEX) such as Uniswap v3, a user has to know about wallets, seed phrases, concentrated liquidity, liquidity pools etc. All this makes onboarding hellish for a new user, and is currently a major bottleneck for Web 3’s mass adoption [1].
Of course, one can argue that there are countless resources online — blogs, videos, courses, games — that aim at onboarding users into the industry. But there is one massive bootstrapping problem to all this: where do you start? Googling “What is Web3” gives you a bunch of ads, followed by a Wikipedia page containing jargon [2]:
Web3 (also known as Web 3.0) is an idea for a new iteration of the World Wide Web which incorporates concepts such as decentralization, blockchain technologies, and token-based economics. — Wikipedia
What are “blockchain technologies,” “token-based economics,” and why do we need a “new iteration of the World Wide Web”? For a complete beginner, none of this makes any sense. You need to make a bunch of additional queries and click on a bunch of different links before you find something that you understand. Fundamentally, this is because a search engine like Google is an aggregator of information and an indexer of content, rather than a producer of this content. Simply put, Google is basically a digital librarian. It doesn’t know how to exactly answer your question, and can only point you to resources it thinks will help. If the content that Google curates doesn’t match your needs, then tough luck. Even all the personalized information that Google gathers about you doesn’t help with this: all it does is try to point you to better resources that it thinks may help you more [3].
ChatGPT, on the other hand, is paradigmatically different: it is an original producer of content, rather than a simple indexer of content. Rather than act as a librarian that points you to resources, ChatGPT is more like a mentor or a teacher that actually tries to teach you what everything means, presented in the way that you tell it to: through giving you analogies, through directing you to appropriate blogs or resources, or through writing a sonnet about it [4]. As an original content creator, ChatGPT does not rely on a fixed corpus of already-made responses, but rather makes a new one for your specific circumstances. This is why ChatGPT can act as an effective one-stop shop for absolute beginners.
Of course, education goes beyond just for absolute beginners; it is a useful tool for anyone wanting to understand a novel corner of Web3, be it zk-SNARKs or inter-blockchain communication protocols. As Web3 is both a diverse and technically challenging industry, there are always new facets, ideas, and concepts to wrap your head around. And that’s why a sophisticated educational tool, such as ChatGPT, is sorely needed in Web3.
Inspiration
ChatGPT not only can teach you the principles of the technology; it can also give you ideas on how to apply the technology in practice. Essentially, ChatGPT can also act as a “fresh set of eyes” that can snowball innovation. Here’s an example:
Q: What are some examples of dApps that ChatGPT can enable or enhance
A: Some examples of dApps that ChatGPT could enable or enhance include:
Social networking dApps — ChatGPT could be used to generate natural language responses to user posts, creating a more engaging and lifelike experience for users.
Gaming dApps — ChatGPT could be used to generate human-like responses from in-game characters, creating a more immersive gaming experience.
Supply chain management dApps — ChatGPT could be used to automatically generate smart contracts for tracking and managing the flow of goods and services within a supply chain.
Voting dApps — ChatGPT could be used to automatically generate natural language summaries of voting results and other data from the blockchain, making it easier for users to understand and track the results of elections and other voting processes.
Predictive analytics dApps — ChatGPT could be used to generate natural language predictions and insights based on data from the blockchain, providing users with more accessible and intuitive ways to analyze and understand complex data. — OpenAI ChatGPT
Some of the dApp ideas presented here are ones that I have thought about before, namely (1) social media chatbots and (3) smart-contract generation. Though (2) obviously seems to make sense in hindsight, I actually did not think of this right off the top of my head. (4) and (5) are further applications of ChatGPT in the realm of analytics that I completely missed.
The more important point is that ChatGPT has the ability to inspire: it gives you ideas on what type of projects you can create using a specific technology in a specific field. Because it is trained on essentially the entire Internet, it can draw parallels and inferences between disparate fields to create a combinatoric explosion in the possibility of ideas. Just imagine the power that ChatGPT can bring, to a hackathon team ready to get their hands dirty after learning a specific tech stack, or to the founder that has only the vaguest of ideas of what she wants to solve.
While ChatGPT’s power of inspiration is game-changing in any field or industry, it is particularly powerful for potential builders in Web3. Fundamentally, Web3 is still a nascent field, one that requires a torrent of new ideas, new models, and new concepts in order to push it toward maturity and mass adoption. At this early stage, because there is no set formula for success, every experiment and idea counts in this industry. Because ChatGPT is able to give you original content, it is also a far better source of inspiration than “a list of 10 project ideas that you can do to improve your Python” online, which are often cookie-cutter and mundane. Web3 is also unique in that the field offers enormous financial incentives, such as grants and hackathon rewards, for such experimental innovation [5]. Thus, ChatGPT can potentially be an enormous catalyst in allowing people to discover and explore new project directions, providing a guiding source of inspiration that kickstarts the flywheel of disruptive innovation.
Iteration
Nonetheless, arguably one of ChatGPT’s most impressive capabilities is its ability to be involved in the iterative process of creation itself. Basically, ChatGPT not only educates and inspires, but allows you to concretely instantiate your idea in code in two important ways: as an original creator of code, and as a sanity check for security vulnerabilities.
Here’s an example of the first case:
Q: Write a smart contract that creates a ERC-20 token called GPToken with symbol GPT using the OpenZeppelin with a disable_mint function that is triggered when 1 million tokens are minted
A:
While this smart contract has some style and optimization potential, such as removing the use of decimals (since the number of tokens will always be an integer), and adding in an if statement to check if the number of tokens equals 1 million and disable_mint if so, it gets the basic gist of the smart contract code, and acts as a great boilerplate starting point. Given that a large portion of development time is dedicated to building these boilerplate templates, ChatGPT can drastically increase the pace of development. Its capabilities are especially impressive when we consider that its output can be customized into various use cases (such as the 1 million token disable-mint here) that traditional online how-to’s and coding guides cannot give. In the field of creation, ChatGPT therefore greatly reduces the developer’s equivalent of “writer’s bloc,” where you have a vague idea of how the code might look but don’t know where to start from.
The second use-case of ChatGPT with actual smart contract code is in detecting vulnerabilities in smart contract code. Even as ChatGPT claims that it does not specifically audit smart contracts, when you feed in a vulnerable smart contract, it is able to provide decent feedback on the contract’s potential vulnerabilities.
As with ChatGPT’s creation of code, the system’s raw output may not be perfect, and may mis-detect some errors in the process. Therefore, while it is not reliable as a complete replacement for a human auditor, it represents a passable first-checker that looks for obvious vulnerabilities [6].
Conclusion
To what extent can ChatGPT truly automate the development process for Web3? I believe that Vitalik summarizes this fairly well:
At this point, AI is quite far from being a substitute for human programmers… That said, it did introduce me to some coding patterns I had not seen before, and it wrote the base converter faster than I would have on my own… That said, AI is improving quickly and I expect it to keep improving further and ironing out bugs like this over time. [7]
The bottom line for an advanced AI like ChatGPT is that it accelerates the learning, thinking, and coding phases of Web3 innovation by reducing the amount of “dead time” spent thinking about how to go from zero to one and repeatedly writing boilerplate code. While in the future, especially with domain-specific fine-tuning, we may be able to see much more sophisticated accurate code automaters, there’s still no way to take humans out of the process of creation.
🐦 @0xfishylosopher
📅 28 December 2022
References
[1] For more information on the onboarding problem, see: https://moralis.io/how-to-boost-web3-user-onboarding-success-rates/
[2] https://en.wikipedia.org/wiki/Web3
[3] https://www.nytimes.com/2022/12/21/technology/ai-chatgpt-google-search.html
[4] https://www.cnbc.com/2022/12/15/google-vs-chatgpt-what-happened-when-i-swapped-services-for-a-day.html
[5] https://www.decentralised.co/p/grants-and-web3
[6] https://beosin.com/resources/can-chatgpt-the-"most-powerful-ai"-detect-vulnerabilities
[7] https://vitalik.ca/general/2022/12/06/gpt3.html
