Wolfram Research: Enhancing the Trustworthiness of Generative AI

The Revolutionary Potential of Large Language Models (LLMs) and the Future of AI

The Hype and Investment in Large Language Models (LLMs)

– The hype surrounding generative AI and large language models is palpable; it’s inescapable.
– More than 25% of US startup investments this year were directed towards AI-related companies.
– OpenAI’s ChatGPT is one of the fastest-growing services of all time.

The Flaw: Hallucinations in LLMs

– A major issue with LLMs is their tendency to hallucinate, making up information.
– Frequently cited rates of hallucination range from 15%-27%, making the issue significant.
– LLMs sometimes come across with assertiveness even when they are presenting false information.

Understanding the Nature of LLMs

– LLMs are designed to be fluid and say plausible things, not necessarily to be factually correct.
– An understanding of the purpose of LLMs is crucial – they are designed to sound plausible, not to provide accurate information.
– The inevitable nature of hallucinations in LLMs is a product of their designed purpose.

Wolfram’s Intervention in LLM Technology

– Wolfram’s ChatGPT plugin aims to make ChatGPT smarter by providing access to powerful computation, accurate math, curated knowledge, real-time data, and visualization.
– Wolfram’s approach to knowledge generation is different from scraping the web – they incorporate human-curated data sets for structured knowledge.
– Wolfram’s long history and expertise in computational technology make them well-suited for this role.

The Role of Symbolic AI and Statistical AI

– Wolfram is on the symbolic side of AI, focusing on logical reasoning, while statistical AI focuses on pattern recognition and object classification.
– Both approaches, however, share a common goal of using computation to automate knowledge.

Use Cases and Future Developments for LLMs

– Wolfram’s plugin can be used for a variety of purposes, including performing data science on unstructured medical records.
– Incremental improvements, better training practices, and potential hardware acceleration will contribute to the development of LLMs in the coming years.
– Copyright rulings and compute costs may impact LLM development in the future.

Challenges and Solutions for LLMs

– The reliability problem for LLMs in computational tasks remains a challenge.
– The combination of computational knowledge and LLM response seems to be working, as long as strong instructions are given.

Upcoming AI & Big Data Expo

– Wolfram will be showcasing their ChatGPT plugin and discussing the future of LLMs.
– The AI & Big Data Expo is an opportunity to learn more about AI and big data from industry leaders.

In conclusion, the potential for large language models is vast, but challenges remain. Wolfram’s intervention in this domain represents a step towards addressing these challenges and making LLMs more reliable. The future of LLM development holds promise, but also requires careful consideration and innovation.