Taming a Stochastic Parrot

How the Enterprise Can Benefit from Generative AI

In the past few months, we’ve seen record-breaking amounts of news about ChatGPT. It’s being positioned as the next big disruptor and the questions are pouring in:

• Will this replace search?
• Is it accurate?
• Is it secure?
• Can I use this for my business?

That’s just a small sample. Answering the above questions with respect to ChatGPT in its current state: it's unlikely.

At its core, ChatGPT uses Large Language Models (LLMs) to process input queries and generate outputs that appear convincingly human. In other - somewhat layman - words, it finds a bunch of terms that are related to the original input and recombines them according to the probability of each related term's relevancy to produce a valid-looking output. This is why it’s often referred to as a “stochastic parrot.” Consequently, this can result in outputs that are partially incorrect, a phenomenon that is referred to as hallucination.

Two important elements that can make ChatGPT (and generative AI more broadly) successful are the scope and timeliness of the data that it is trained on. If it only has access to a small set of outdated data, the outputs it produces won’t be able to offer much value - especially in the enterprise. Additionally, LLMs need to satisfy enterprise’s existing requirements related to security and Access Control Lists. In case it gets access to internal/confidential information, companies need assurance that information handling fully satisfies their requirements. Once these hurdles - and a few others - can be overcome, the answer to the questions at the top becomes: it’s possible.

Generative AI is having an impact, but not so much on the enterprise's internal data in its current state. This will happen in conjunction with existing technologies, namely insights engines. We see the integration of generative AI with the insight engine being the most successful path for enterprises taking advantage of this technology

Gartner states: “We expect to see this evolution of the insight engines continue through the combination of generative AI assistants, metadata-driven knowledge graphs, and integration via APIs to and from analytics and BI systems. The result will be a richer and personalized knowledge worker experience across structured and unstructured data and content.”¹

Squirro is helping drive this revolution by introducing its enterprise-ready generative AI solution. The solution brings all the benefits of generative AI to enterprises while eliminating its limitations.

Squirro’s solution combines LLMs with its own Composite AI - fusing different AI technologies, like Machine Learning and Knowledge Graph, - and Insight Engine technologies. This means enterprises can generate accurate and fully contextualized results based on their own data and eliminate hallucinations, by adding the source of information to the LLM response as evidence and providing transparency and explainability in a secure and compliant way.

To learn more about Squirro’s Generative AI solution, download this e-book and FAQ.

¹Source: Quick Answer: What Are the Short-Term and Midterm Implications of ChatGPT for Data and Analytics?, Rita Sallam, Jim Hare, Bern Elliot, David Pidsley, Erick Brethenoux, Roxane Edjlali, Radu Miclaus, Pieter den Hamer, Afraz Jaffri, Kurt Schlegel, Julian Sun, Georgia O'Callaghan, 23 February 2023 - ID G00786939

Gartner does not endorse any vendor, product or service depicted in its research publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner research publications consist of the opinions of Gartner’s research organization and should not be construed as statements of fact. Gartner disclaims all warranties, express or implied, with respect to this research, including any warranties of merchantability or fitness for a particular purpose.