They use the same LLMs in the background, but certainly where I work (large financial sector company) it's a very different product than what the public see and use.As someone who is working with it in a business context, it's largely the exact same AI! The only real guardrails in place are ones that prevent the AI learning from the information given to it to avoid e.g. GDPR breaches. All AI hallucinates, though you can reduce the risk by careful use of prompting such as specifically instructing it to evidence everything it says. But the models themselves are the same ones.
Whilst we do have access to a chatbot type interface for general queries most of our work is done through specialised interfaces, sandboxed within our organisation which specifically uses only internal documents and does not train outside the business models. We've been testing and manually validating the outputs consistently over the past 18 months and it is actually insane the leaps and bounds it has come on in that short space of time.
Most of the AI testing we do is repetitive tasks (agents) or analysis (ad-hoc) and we've been running it alongside the human equivalent workers and the accuracy from the AI has....
Last edited:
