coindesk.com
Spectral Labs, a crypto development firm focused on artificial intelligence, wants to make it easier for non-programmers to build on blockchains. On Tuesday the company will launch Syntax, an AI app that can help anyone – coders and non-coders alike – spin up smart contracts for Ethereum and dozens of other blockchains.
While Syntax represents an impressive technical achievement, Spectral still faces the hurdle of convincing users to entrust AI with their precious digital assets.
Those familiar with Chat GPT and similar chatbots will be at home on Syntax interface, which is powered by a crypto-specific large language model (LLM) and can carry on SMS-like conversations with users.
In addition to answering blockchain-related queries, like "what is the circulating supply of ETH," the web app can turn prompts from users into production-ready Solidity code – the programming language used by Ethereum and other blockchains based on the Ethereum Virtual Machine (EVM) standard.
"Syntax users can compile, debug, and deploy AI-generated solidity code," Spectral Labs said in a statement, adding that NFTs, arbitrage bots and rollups are among the tech's possible use cases.
"Say I want to code an ERC-20 token called 'ABC' with 100 million supply," Spectral Labs CEO Sishir Varghese told CoinDesk in an interview. "Syntax would generate all of the code for the user. The user would be able to compile, test for vulnerabilities, test for gas efficiency and code optimization, and then deploy the app directly on-chain right from the interface."
AI-powered coding assistants are nothing new: Engineers have been using ChatGPT to generate code for the past year, and GitHub Copilot, Microsoft’s AI-powered coding whiz, has become a mainstay of the modern programming arsenal since its release in 2021.
"I would say there hasn't really been an LLM, fine-tuned for Solidity," said Varghese.
In addition to being tuned specifically for Solidity development, Syntax is novel in that it plugs straight into EVM-compatible blockchains, meaning users can press a button to instantly deploy their Syntax-made "agents" to blockchains like Ethereum, Arbitrum or Coinbase's Base.
The app can be configured with private keys and other data to give the agents control over real crypto funds, which means the agents – depending upon what they're programmed to do – can be let loose to buy, sell and trade tokens as if they were actual humans.
When it comes to using LLMs, there's always the tricky matter of "hallucinations" – where AI can make things up or, in the case of coding, create correct-seeming code that is error-prone or behaves unexpectedly. When the code generated by an AI is tasked with handling real money, hallucinations become an obvious problem.
"We tried to put in some safeguards around vulnerability detection and code efficiency," said Varghese. "All those are kind of built-in already. Plus, there will be some pre-built agents that will also help with vulnerability detection."
Syntax was trained on a specially curated set of audited smart contracts, security reports and developer docs, he added.
"Creating a quality data set is pretty crucial because that's what you want the LLM to reproduce in terms of code," said Varghese. "If you just took everything that ever existed, you won’t really get good results."
The Spectral Labs CEO allowed that using an AI tool like Syntax will inevitably carry some risks: "I think skilled developers are still pretty apprehensive about all of the copilot stuff – except in Web2. For Solidity, this is going to be a brand-new experience, to be honest."
Varghese expects that retail traders, rather than experienced programmers, will be Syntax's main users.
"I don't foresee them jumping in and trusting it," he said in reference to skilled developers. "I think the cooler thing is that a retail person who's never deployed a contract or even generated Solidity code before, he's gonna be able to do this quite seamlessly and easily."