Join our daily and weekly newsletters for updates and exclusive content on top AI industry news.
Yoav Shoham, Co-CEO of AI21, a leading language model company, recently responded to comments from Amazon and others suggesting that language models are losing their uniqueness. He disagrees, stating that models do differentiate.
AI21 specializes in creating AI systems for enterprise companies, especially those that excel in specific tasks like text summarization.
During an interview with Shoham, we discussed recent events in generative AI, including issues at OpenAI and announcements from Amazon AWS.
Shoham addressed the idea that language models might be losing their distinctiveness and that the real value of generative AI could lie in proprietary data. Amazon’s Swami Sivasubramanian argues that differentiation comes from data and that companies must leverage their unique data to create standout AI applications. This sentiment was echoed by Albertsons’ VP of AI and Data Science, Miguel Paredes, who suggested that with widespread access to models like OpenAI’s ChatGPT and Bard, companies need to focus on utilizing their data to build superior AI systems.
Shoham agrees on the importance of data but believes that, for now, developing excellent AI systems shouldn’t shift focus away from LLMs. He noted that creating a high-quality language model is challenging and that understanding their capabilities and limitations can take time. Prototyping and benchmarks offer limited insights into a model’s performance.
Shoham highlighted that even basic tasks like text summarization can be complex for a language model, but focusing on specific tasks can significantly improve performance. He mentioned that AI21’s text summarization model outperformed GPT-4, ChatGPT, and Claude in tests conducted by a major financial institution.
Looking ahead, Shoham anticipates that the focus will broaden beyond just language models. Future AI systems will incorporate language models but offer a wider range of functionalities, presenting ample opportunities for innovation.