A Pragmatic View of ChatGPT in a Web3 World
ChatGPT has stolen the headlines as one of the most impressive technology releases of 2022. Initially unveiled to highlight the capabilities of the upcoming GPT-4, ChatGPT quickly transcended the boundaries of the artificial intelligence (AI) space to become a pop culture phenomenon. For the first time in history, we are seeing AI actively discussed all the way from mainstream media outlets to dinner conversations. The fact that an obscure area of AI was going to cause this revolution was unfathomable just a year ago.
Jesus Rodriguez is the CEO of IntoTheBlock.
ChatGPT is the latest in a family of large generative language models (LLM) that are clearly transforming the nature of content creation, application development and user experience. The speculations of how ChatGPT and the upcoming GPT-4 type technologies might impact different industries have run high and wild and Web3 is not the exception.
The hype about the possibilities of ChatGPT have certainly found its way to the Web3 space. In recent weeks, crypto media outlets have openly discussed the role that ChatGPT can play in ending Google’s dominance, completely automating smart contract development and making Web3 the dominant architecture paradigm. A different perspective becomes apparent once you look into the specific capabilities and limitations of ChatGPT and match it to the current state of Web3 technologies.
For all the hype around ChatGPT, there has been very little discussion about the specific capabilities and differentiators of the model. ChatGPT represents a major evolution in OpenAI’s GPT family of models but that evolution has materialized in a very specific dimension. The main difference between ChatGPT and its predecessors is that the former is key at following instructions. Models like GPT-3 were able to perform a handful of language tasks such as summarization, question answering or text completion based on carefully curated prompts.
Read more: Here’s Why Artificial Intelligence Focused Cryptocurrencies Are Vastly Outperforming Bitcoin
However, GPT-3 exhibited major limitations following more generic instructions. In early 2022, OpenAI quietly unveiled InstructGPT, a fine-tuned version of GPT-3 that uses a technique called reinforcement learning with human feedback (RLHF) to better follow human intents. OpenAI had published the research behind RLHF back in 2017 and went mostly unnoticed until InstructGPT was created. InstructGPT is the crown jewel at the center of ChatGPT and one of the key differentiators with its predecessors. These days, when you are using the OpenAI API by default it uses InstructGPT.
The AI community recently started referring to models such as ChatGPT with the term “foundation models.” Quoted by Stanford University, this term refers to the unique characteristic of these models to be fine-tuned for specific scenarios. For instance, OpenAI created Codex, a new fine-tuned version of GPT-3 to generate programming language code that is powering programs such as GitHub CoPilot. Codex is also integrated into ChatGPT.
Now that we understand the core capabilities and differentiators of ChatGPT, we can extrapolate those to our Web3 universe and start dreaming about the potential.
ChatGPT in a Web3 world
The foundation model revolution with platforms such as ChatGPT is going to deeply influence how software is created and experienced across the entire technology market. Web3 also represents a new paradigm for distributed computing, so the combination with foundation models like ChatGPT offers a blank canvas full of opportunities. Some of those opportunities are already possible with today’s technologies.
Explorers are the search experience in Web3 and the core building block for human interactions with blockchains. However, the user experience of blockchain explorers is designed for domain experts. Imagine an explorer powered by a fine-tuned version of ChatGPT for blockchain activity. In that experience, a normal user could ask questions such as “Any large institutions transferring funds to Binance?,” “When was the last time something similar happened?” or “Are there any interesting patterns in the recent transaction activity?” Search is one of the experiences that might be reimagined with technologies like ChatGPT, and explorers could be a perfect candidate.
Smart contract development assistants
Programming smart contracts remains a highly sophisticated task for developers. ChatGPT components such as Codex are able to generate Solidity code from language descriptions. Imagine a smart contract assistant in which a developer can type something like “What’s the solidity code to request a flashloan in Aave?” and it will generate the corresponding smart contract code snippet.
Read more: Jesus Rodriquez – The Coming Convergence of NFTs and Artificial Intelligence
Smart contract security testing
Smart contract audits are slow, expensive and tedious processes that are nonetheless necessary. A large majority of the auditing process relies on executing tests that are quite often not obvious to smart contract developers. Imagine having a fine-tuned version of ChatGPT for smart contract audits that could take a language input such as “” and run a battery of tests in a given smart contract.
Arguably one of the most obvious applications of models such as ChatGPT is to enable a new generation of non-fungible tokens (NFT) that incorporate conversational intelligence. Imagine a version of your favorite NFT collection that allows you to ask questions about the creator’s inspiration or specific artistic details.
Wallets are the main entry point for interactions with decentralized applications (dapp) in the Web3 world. Just as the user experience in Web2 applications is being reimagined with foundational models such as ChatGPT as a fundamental construct, we can think of a similar trend for crypto wallets. Imagine a wallet experience in which a user can simply express its intentions to perform a transaction, request information or execute specific tasks simply using natural language. Conversation is going to be one of the most interesting trends in Web3 user experience in the next few years.
Web3 in a ChatGPT world
Foundational models such as ChatGPT will, undoubtedly, enable a new generation of capabilities in decentralized applications, but Web3 can play an interesting role in the infrastructure powering these models. Auditability is one of the key concerns around the emergence of models like ChatGPT. Understanding the causes of harmful, fake, unbiased or unfair content has been at the center of the debate for the mainstream adoption of ChatGPT and similar models. Distributed ledgers are the perfect technology to enable trustless transparency and auditability for models such as ChatGPT.
Pre-training and fine-tuning is another aspect in which Web3 platforms can contribute to models like ChatGPT. The computational requirements for pre-training or fine-tuning foundation models results are prohibited for most organizations. Decentralized computation networks such as blockchains can enable scalable computation economies that can facilitate the pre-training or fine-tuning of models like ChatGPT.
Not just ChatGPT
ChatGPT and the upcoming GPT-4 release represent some of the most overhyped technologies of the last few decades. While it might be easy to get caught up in the hype, the transformational impact of these models is real and it definitely applies to Web3. One thing to understand is that ChatGPT won’t stand alone in this area much longer.
Companies such as Google with its model LaMDA, DeepMind with Sparrow, Anthropic with Claude and Stability AI with an open-source version of ChatGPT are likely to become relevant players in this market in a next few months. In the Web3 world, these models will power new experiences for how to author and interact with smart contracts, dapps, Wallets, decentralized finance (DeFi) protocols, NFTs and pretty much every area of the ecosystem. The era of language models in Web3 is here and ChatGPT is just the beginning.