On June 19, Ethereum block explorer and analytics platform EtherScan launched a new tool called “Code Reader” that uses artificial intelligence to retrieve and interpret the source code of a given contract address. After prompting the user, the code reader generates a response via her OpenAI Large Language Model (LLM), providing insight into the contract’s source code files. Etherscan developers write:
“Use of this tool requires a valid OpenAI API key and sufficient OpenAI usage limits. This tool does not store API keys.”
Code Reader use cases include gaining deeper insight into a contract’s code through AI-generated explanations, retrieving a comprehensive list of smart contract features related to Ethereum data, distributed underlying contracts This includes understanding how they interact with type applications (dApps). “Once the contract file is retrieved, he can select a specific source code file to read through. In addition, he can also modify the source code directly within the UI before sharing it with the AI,” says Dev. person writes.
Amid the AI boom, some experts warn about the feasibility of current AI models. According to a recent report released by Singaporean venture capital firm Foresight Ventures, “computing power resources will be the next big battleground in the next decade.” That said, despite the growing demand for training large-scale AI models on decentralized distributed computing power networks, researchers believe that current prototypes are complex data synchronization, network optimization, data It says it faces significant constraints, including privacy and security concerns.
As an example, Foresight researchers noted that training a large model with 175 billion parameters in single-precision floating-point representation would require about 700 gigabytes. However, distributed training requires sending and updating these parameters frequently between compute nodes. If you have 100 compute nodes and each node needs to update all parameters at each unit step, your model will need to send 70 terabytes of data per second, which is the capacity of most networks. much more than The researchers summarized:
“In most scenarios, small AI models are still a more viable option, and the FOMO trend towards large models should not be overlooked too early.”
