Fraction AI Demos Agents Writing Machine Code
Pushing the boundaries of efficiency, Fraction AI demoed AI agents that bypass high-level languages like Python to write machine code directly. This low-level approach aims for maximum performance, a technique with significant potential implications for latency-sensitive applications like high-frequency trading infrastructure.
Fraction AI's core technology revolves around a decentralized platform where AI agents compete to improve their skills in specific domains called "Spaces." These competitions cover areas like "Writing Code" and "Deep Finance Tasks," with winning agents being rewarded and their successful strategies used to refine future performance. This model of continuous improvement through competition is what the company calls Reinforcement Learning from Agent Feedback (RLAF). The platform allows anyone to create and train AI agents using natural language prompts, without needing to write code themselves. This approach aims to decentralize and democratize AI development, moving it beyond the control of a few large tech companies. Fraction AI recently launched its mainnet on Base, an Ethereum Layer 2 network, transitioning from a testnet phase that attracted significant user engagement. The CEO of Fraction AI, Shashank Yadav, has a background as a quant trader and was a founding engineer for the Core Machine Learning team at Goldman Sachs. This experience in high-frequency trading (HFT) and financial technologies provides context for the company's focus on performance-critical applications. In traditional finance, over 80% of trades are executed by AI, a domain where speed is paramount. Generating machine code directly, rather than a high-level language like Python, is a strategy to minimize latency. In high-frequency trading, delays of even a single millisecond can lead to significant financial losses. By bypassing the interpretation or compilation steps required for high-level languages, AI-generated machine code could offer a substantial performance advantage in these environments. This approach aligns with the broader trend of using AI to optimize every stage of the HFT pipeline, from predicting market movements to executing trades with microsecond precision.
