Renowned tech commentator Robert Cringely has joined forces with a team of experts to co-found AI startup 2Brains Inc., tackling one of the most pressing issues in large language model (LLM) research: hallucinations.
Cringely, 73, has a storied career in tech, having worked at the Stanford Artificial Intelligence Lab as far back as 1978. His most recent venture is a direct response to the limitations of current LLMs, which often produce fictional or false information, known as hallucinations. Cringely’s site, which has been on hiatus for two years due to health issues, has now resumed operations and will serve as a platform to promote 2Brains Inc.’s mission.
Dr. Sam Altman’s former colleague at OpenAI, Ilya Sutskever, has argued that the problem of hallucinations is inherently tied to the scaling of computation in LLMs. However, Sutskever’s colleague at Meta AI, Dr. Jason Weston, has proposed a different explanation: that these inaccuracies might be the result of complex interactions between model components, rather than simply a matter of scaling.
Addressing the Elephant in the Room
2Brains Inc.’s approach to solving hallucinations is centered around the development of more sophisticated tools for identifying and addressing these inaccuracies. By harnessing the power of multi-modal learning and incorporating techniques from cognitive science, the startup aims to create LLMs that are more reliable and trustworthy.
Practical Implications
What this means for users is a significant improvement in the accuracy and reliability of AI-driven applications. Whether it’s AI-generated content, decision-making systems, or chatbots, reduced hallucinations will lead to more dependable and trustworthy interactions. As AI becomes increasingly ubiquitous in our daily lives, addressing concerns around hallucinations is critical to maintaining public trust and ensuring the safe deployment of AI technologies.
What’s Next
2Brains Inc. has announced plans to release a series of research papers and prototypes demonstrating the effectiveness of their approach. As the startup continues to make progress, it will be interesting to see how their solutions are received by the wider AI research community and whether they can provide a meaningful step forward in solving the problem of hallucinations.
