Yoav Shoham, Co-CEO of a leading large language model (LLM) company A121, reacts to comments made recently by Amazon and others that imply LLMs are losing their differentiation.
He disagrees, saying: “Models do differentiate.”
Shoham’s company, AI21, is a provider of AI systems for enterprise companies, and it specializes in offering task-specific LLMs, including models that do text summarization very well.
I interviewed Shoham yesterday to get his perspective on all of the news developments recently around generative AI, including the debacle at leading LLM provider, OpenAI, and the slew of announcements made by Amazon AWS this week.
Click on the video above to see his full comments.
Shoham made his comments about LLMs in reaction to a point I made about how people are saying LLMs may be starting to lose their differentation, and that the true value of generative AI may be elsewhere, for example in proprietary data.
“If all of the providers end up actually building models that look very similar, where is the differentiation?” Swami Sivasubramanian, Amazon AWS’s vice president of Data and AI, said in an interview with me Monday. “That differentiation comes to data,” he said, arguing that now it is up to enterprise companies to leverage their proprietary data correctly to create differentiated AI applications.
This echoes comments made by other leaders. Miguel Paredes, VP of AI and Data Science at Albertsons told us in an interview earlier this month: “These models are becoming commodities,” he said, adding that all companies will be able to access OpenAI’s ChatGPT, Bard and other models equally. So companies wanting to build excellent AI systems will have to shift their focus away to instead find ways to compete by leveraging their own data.
Shoham responded by saying the he agreed that data is critically important, but disagreed that the focus for building excellent AI systems is shifting away from LLMs toward leveraging data, at least for now.
“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.” Generally available benchmarks, and even prototyping LLMs directly, will only give weak signals on how good a model is, he said.
Even basic functionality like text summarization can be difficult to do for an LLM, but by focusing on such specific tasks, they can be made much better. He gave the example of A121’s text summarization model, which was ranked better than GPT-4, ChatGPT and Claude by a large margin, when tested by a large financial institution.
At the same time, Shoham acknowledged that in a year, the focus will move away from just 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.”
Are you ready to bring more awareness to your brand? Consider becoming a sponsor for The AI Impact Tour. Learn more about the opportunities here.
Yoav Shoham, Co-CEO of a leading large language model (LLM) company A121, reacts to comments made recently by Amazon and others that imply LLMs are losing their differentiation.
He disagrees, saying: “Models do differentiate.”
Shoham’s company, AI21, is a provider of AI systems for enterprise companies, and it specializes in offering task-specific LLMs, including models that do text summarization very well.
I interviewed Shoham yesterday to get his perspective on all of the news developments recently around generative AI, including the debacle at leading LLM provider, OpenAI, and the slew of announcements made by Amazon AWS this week.
VB Event
The AI Impact Tour
Connect with the enterprise AI community at VentureBeat’s AI Impact Tour coming to a city near you!
Click on the video above to see his full comments.
Shoham made his comments about LLMs in reaction to a point I made about how people are saying LLMs may be starting to lose their differentation, and that the true value of generative AI may be elsewhere, for example in proprietary data.
“If all of the providers end up actually building models that look very similar, where is the differentiation?” Swami Sivasubramanian, Amazon AWS’s vice president of Data and AI, said in an interview with me Monday. “That differentiation comes to data,” he said, arguing that now it is up to enterprise companies to leverage their proprietary data correctly to create differentiated AI applications.
This echoes comments made by other leaders. Miguel Paredes, VP of AI and Data Science at Albertsons told us in an interview earlier this month: “These models are becoming commodities,” he said, adding that all companies will be able to access OpenAI’s ChatGPT, Bard and other models equally. So companies wanting to build excellent AI systems will have to shift their focus away to instead find ways to compete by leveraging their own data.
Shoham responded by saying the he agreed that data is critically important, but disagreed that the focus for building excellent AI systems is shifting away from LLMs toward leveraging data, at least for now.
“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.” Generally available benchmarks, and even prototyping LLMs directly, will only give weak signals on how good a model is, he said.
Even basic functionality like text summarization can be difficult to do for an LLM, but by focusing on such specific tasks, they can be made much better. He gave the example of A121’s text summarization model, which was ranked better than GPT-4, ChatGPT and Claude by a large margin, when tested by a large financial institution.
At the same time, Shoham acknowledged that in a year, the focus will move away from just 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.”
VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.
Author: Matt Marshall
Source: Venturebeat
Reviewed By: Editorial Team
