Ethereum’s analytics and block explorer platform, Etherscan, unveiled a novel application called “Code Reader” on June 19. The application is an AI-empowered tool that fetches and deciphers the source code tied to a particular contract address. Using prompts, users engage with Code Reader, triggering responses from OpenAI’s sophisticated language model. This results in a deeper understanding of the files making up a contract’s source code. As per instructions on the tool’s page:
“Access to a valid OpenAI API Key and adequate OpenAI usage quotas is necessary for using the tool. Your API keys are not stored by this tool.”
This tool presents an array of applications, including obtaining AI-facilitated explanations for a better understanding of a contract’s code, generating comprehensive lists of smart contract functionalities tied to Ethereum data, and decoding the interaction of a base contract with decentralized applications. “Upon retrieving the contract files, a specific source code file can be selected for perusal. Users also have the option to modify the source code within the UI before presenting it to the AI,” the page suggests.
Despite AI’s burgeoning presence, some experts have reservations about the practicality of current AI models. Foresight Ventures, a Singaporean venture capital firm, in a recent report stated that “control over computing power resources will shape the competitive landscape in the next decade.” Even as demand grows for the training of large AI models on decentralized distributed computing power networks, researchers note that existing prototypes face substantial challenges such as complex data synchronization, network optimization, and data privacy and security concerns.
In a scenario provided by the Foresight researchers, the training of an expansive model with 175 billion parameters with single-precision floating-point representation would necessitate roughly 700 gigabytes. With distributed training, the continuous transmission and updating of these parameters among computing nodes becomes a requirement. Given a situation with 100 computing nodes, and each node needing to update all parameters at each increment, the model would need to transmit a massive 70 terabytes of data per second, greatly surpassing the capability of most networks. The researchers concluded:
“Small AI models often represent a more viable choice in many situations, and should not be prematurely dismissed in the rush of FOMO [fear of missing out] on larger models.”
OpenAI is an influential research organization focusing on ensuring the benefits of artificial general intelligence (AGI) are shared by everyone. They have developed several versions of a highly capable AI language model, GPT-3 being the most recent until September 2021.
AI is gradually permeating the blockchain and cryptocurrency sector. It is deployed for various purposes including data analytics, predicting market tendencies, optimizing trading, and now, with Etherscan’s Code Reader, for enhancing our comprehension of contract codes.
The launch of Etherscan’s AI-driven Code Reader signifies another stride towards combining AI’s abilities with the openness and decentralization offered by blockchain technology. This not only heightens the efficiency of contract code interpretation but also paves the way for new breakthroughs in the blockchain domain.