LLMs Still Differentiate in the AI Landscape
Yoav Shoham, Co-CEO of AI21, disagrees with recent comments suggesting that Large Language Models (LLMs) are losing their differentiation. As a provider of AI systems for enterprise companies, AI21 specializes in offering task-specific LLMs, particularly in text summarization. In an interview, Shoham asserts that LLMs do indeed differentiate themselves through their capabilities.
“Models do differentiate,”
- Yoav Shoham, Co-CEO of AI21
The Role of Data in AI Systems
While some argue that the value of generative AI may lie in proprietary data, Shoham maintains that the focus for building excellent AI systems should still center around LLMs. He acknowledges the importance of data but believes that understanding the capabilities and limitations of language models is crucial.
“It’s very hard to create an excellent language model, and it can take a while to understand how good they are, and their limitations.”
- Yoav Shoham, Co-CEO of AI21
Shoham emphasizes the difficulty in evaluating language models, even with generally available benchmarks. He highlights the significance of focusing on specific tasks, such as text summarization, to enhance the performance of LLMs. He cites A121’s text summarization model, which outperformed GPT-4, ChatGPT, and Claude by a large margin in testing conducted by a financial institution.
While Shoham recognizes that the future of AI will involve systems that go beyond LLMs, he believes that there is still a lot of innovation to be explored within the realm of LLMs.
“We’ll be speaking about AI systems that include large language models, but they’ll do a lot of other things…It’s a blue ocean. There’s a lot of innovation to be had there.”
- Yoav Shoham, Co-CEO of AI21
