Some of the major companies in the artificial intelligence software industry have encountered challenges with ensuring that the technology behaves as intended by its creators.
Prominent corporations such as Google, OpenAI, and Alphabet Inc. are actively dedicating their workforce, financial resources, and computational power to address this issue. In a similar vein, Anthropic, a competitor of OpenAI, is pioneering the development of Claude, an AI robot touted for its enhanced safety features.
A newcomer on the scene, Synth Labs, is also stepping up to tackle this problem starting this month. Founded by established figures in the AI field, the company is stepping out of the shadows this week with backing from M12, M12, and Eric Schmidt’s Innovation Endeavors. Synth Labs aims to provide software solutions, including open-source options, to assist various enterprises in ensuring the smooth functioning of their AI systems. The company prides itself on transparency and collaboration.
The challenge of alignment, a common issue faced by AI applications like chatbots that rely on extensive language models trained on vast datasets, is further compounded by the variability in people’s ethical standards and opinions regarding AI boundaries. Synth Labs’ offerings will play a pivotal role in guiding and customizing large language models, particularly those that are open source.
Key figures behind Synth Labs, including founders Louis Castricato and Nathan Lile, along with consultant Stella Biderman from EleutherAI and co-founder Francis deSouza, formerly of Illumina Inc., have spearheaded the company’s evolution from a project within EleutherAI’s philanthropic AI research lab. Synth Labs has secured funding from various sources, including M12 and Eric Schmidt’s Innovation Endeavors.
Castricato emphasized that the company has developed sophisticated tools in recent months capable of swiftly assessing extensive language models on intricate subjects. The primary objective is to democratize access to user-friendly tools for comparing and contrasting AI models efficiently.
A recent collaborative study by Castricato, Lile, and Biderman outlines the company’s strategic approach: leveraging responses from OpenAI’s GPT-4 and Stability AI’s Stable Beluga 2 AI models to construct a database. This data was then utilized in an automated process to train a robot to shift conversations away from certain topics towards others.
According to Lile, the focus is on empowering users to tailor configurations according to their organizational or personal preferences, offering flexibility in managing early-stage resources.
