12 Dec The promise of ChatGPT and the reality of the enterprise
A Q&A with CEO Ram Menon
The recent release of GPT-3 (Generative Pre-trained Transformer 3) has generated a lot of excitement in the tech world. With 175 billion parameters, GPT-3 is the most powerful language processing model ever created, and it has the potential to revolutionize many aspects of our lives, from language translation to personal assistants to chatbots. However, while GPT-3 is undoubtedly impressive, there are also limitations and challenges to using it in enterprise settings. We sat down with Ram Menon, CEO of Avaamo, to discuss the broader implications of this innovation and the realities of deploying this technology in the enterprise.
Q: Why is there so much hype about ChatGPT? What makes it different from previous innovations in conversational AI?
A: Well, anyone who has a a free account is posting on social media and that has caused quite a stir. But as a practical matter LLMs (large language models) are trained on internet data and present a great opportunity for prosumers and consumer-oriented creators to build stuff.
Now that being said, ChatGPT is an incredibly exciting development in Conversational AI. Chatbots allow interaction in a seemingly ‘intelligent’ conversational manner, however ChatGPT3 produces output that appears to have actually understood the question, the content and the context. This creates a very human-like natural interaction, often insightful, sometimes profound.
But it’s still nascent technology; ChatGPT in its current form is best used for creative endeavors like content creation and ideation. It is still not great at things like precise answers to internal content or complex math equations. ChatGPT should be seen as a glimpse of what is yet to come, namely, an exciting future for AI helping humans in meaningful ways.
Q: So is ChatGPT applicable to the enterprise? What are the challenges of LLMs?
A: While this is a great leap forward in language models and this should truly be commended, sufficiently large language models’ ability to be both generic and generative, indicates to me that there’s a substantial risk we do not yet fully understand when and if the model is trained on enterprise data. In the short term, I don’t see an enterprise customer feeding their employee data or for our healthcare customers, patient data into a large language model of this sort. The security and compliance will simply not pass muster at virtually any company in this stage of AI evolution for GPT-3.
What is frequently lost in the shuffle in Silicon valley as we rave about these new developments is that enterprises are not dependent on internet data. Rather, they have their own proprietary data; customer records, patient records, medication records, IT ticket records, and they demand precise answers with utmost accuracy. The expectation is that companies like Avaamo run their AI models on this proprietary data and they are what they are – closed and carefully monitored. The price of wrong answers is very high. This last 10% is the hardest to do.
A case in point, one of the charming but flawed issues with large language models is a behavior called “stochastic parroting” or statistically analyzing text strings and “making thing up.” It sounds good but it is false. Sam Altman, the CEO and Co-founder of OpenAI, admitted as much in a tweet over the weekend saying “it’s a mistake to be relying on it [ChatGPT] for anything important right now.”
ChatGPT is incredibly limited, but good enough at some things to create a misleading impression of greatness.— Sam Altman (@sama) December 11, 2022
it's a mistake to be relying on it for anything important right now. it’s a preview of progress; we have lots of work to do on robustness and truthfulness.