The Hype and the Problem
The hype around generative AI and large language models (LLMs) has been immense. These technologies, led by OpenAI’s ChatGPT, have captured the attention of investors and developers alike. However, one major issue with LLMs is their tendency to hallucinate or make things up. This can be problematic when the model presents these fabrications as facts. The rate of hallucination can range from 15% to 20% and even as high as 27% in some cases.
Understanding the Purpose
These hallucinations are an inherent aspect of LLMs. Jon McLoone, Director of Technical Communication and Strategy at Wolfram Research, explains that LLMs are designed with a purpose in mind – to sound plausible and human-like, rather than to provide accurate information. As a result, they may say things that sound right but are nonsensical when examined closely.
The Need for Objectivity
To address this issue, a solution is needed to inject objectivity into LLMs. Wolfram, a computational technology company, has developed a ChatGPT plugin that provides access to powerful computation, accurate math, curated knowledge, real-time data, and visualization. By integrating Wolfram’s expertise, LLMs can benefit from a more objective and structured approach.
The Role of Wolfram
Wolfram has a long history of computational technology and sits on the symbolic side of the AI spectrum, which focuses on logical reasoning rather than statistical AI. The company’s collaboration with OpenAI aims to combine the strengths of both approaches. Wolfram’s plugin teaches LLMs to recognize and utilize the curated knowledge and computation capabilities of Wolfram|Alpha, their knowledge engine.
Expanding Use Cases
The combination of ChatGPT’s language mastery and Wolfram’s computational mathematics opens up various use cases beyond chat interactions. For example, Wolfram’s plugin can be used to perform data science on unstructured medical records, correcting errors and identifying correlations within the data. The potential for LLMs extends beyond chat-based applications, as they excel in handling unstructured data.
The Future of LLMs
According to McLoone, incremental improvements in LLMs can be expected, along with better training practices and potential hardware acceleration. However, a significant sea-change like the past year is unlikely due to high compute costs and potential copyright limitations on training sets. While LLMs have limitations in computational tasks, they can excel in synthesizing new knowledge when combined with computation-oriented approaches.
The Combination in Action
When LLMs are given strong instructions and prompted to use Wolfram’s tools, the combination proves effective. McLoone emphasizes the importance of prompt engineering to guide the LLM’s responses. When computation generates knowledge and injects it into an LLM, the model tends to adhere to the facts. It’s akin to whispering facts to a loudmouth person at a pub – they’ll readily take credit for the information.
Wolfram will be demonstrating the capabilities of their plugin at the upcoming AI & Big Data Expo Global event in London on November 30-December 1.
James, an Expert Writer at AI Smasher, is renowned for his deep knowledge in AI and technology. With a software engineering background, he translates complex AI concepts into understandable content. Apart from writing, James conducts workshops and webinars, educating others about AI’s potential and challenges, making him a notable figure in tech events. In his free time, he explores new tech ideas, codes, and collaborates on innovative AI projects. James welcomes inquiries.
