ChatGPT at Enterprise
We know about ChatGPT for personal use and how ChatGPT is being integrated with Microsoft products to increase productivity. But what about how Enterprises use it for their industry verticals/domains?
With ChatGPT, it is the latest of the Large Language Models that took off in 2018, and now it is being scaled globally with this free research preview, with the simple interface of the prompt and outputs in texts. Every knowledge-based field, it can ‘mimic.’ I was talking to a colleague who is an expert in text analytics about LLMs recently, and really, this ChatGPT may be a fad, as Language Models have been around for years, and only took off exponentially, beginning in 2018, with the evolution of compute to handle Big Data and Deep Learning at the same time. So with LLM’s it follows the evolution of compute horsepower to train on such massive corpuses (corpi? lol) and run them through deep learning algorithms- all resource intensive.
In my previous stints from 2015 to 2020 in AI startups, I have experience with labeling corpuses (corpi? lol) and training language models, in the medical/psychiatry domain, and ultimately built an AI psychiatry classifier that diagnosed seasonal affective disorder (seasonal depression) from mock clinical notes. This AI Psychiatry classifier was was validated against other psychiatric disorders and the ground truth validated by a domain expert, a psychiatrist (yours truly). This was a Language model, albeit on a small scale, but with expert input and validation.
So language models are not new- they have been around for years, but ChatGPT is monetizing it by scaling it globally, making it available free for the public, in this research preview. The scale of both the billions of parameters on which the LLM was trained, and the scale of its reach for global use, is truly impressive. But that is where it stops, as this LLM, as with all Language Models, needs to be validated by experts in the domain that the LLM is attempting to provide an answer. At the end of the day, nobody can be certain of the validity of the outputs, as those outputs need to be labeled as correct or incorrect, and the optimal answer provided to ‘correct’ it. You see that the ChatGPT is utilizing a research preview, and there is a thumbs up and thumbs down, for the user to give their opinion what is right or wrong answers. This is called RLHL (reinforcement learning human in the loop), where the model ‘learns’ from the human, if the model is right or wrong. Ideally, you would like an expert in that domain to train it, but it appears that ANYONE can train the ChatGPT model, and that in itself is concerning, as the model is ‘learning’ the wrong things from the public. Ideally, the validations, comparing to the ground truth, comes from experts in the domain, not the public. So this is a major concern of using it for the Enterprise, when using it in a particular domain.
And on the other side, the corpus on which the LLM was trained, we have to validate how the corpus was labeled, and if they indeed utilized experts in the domain to train the corpus in that domain. A colleague was sharing with me how this was done overseas, with some questionable methodology on who was labeling the corpus, and how they were labeled.
Another major concern involves plagiarism. My text analytics colleague was hearing in her NLP circles that they were calling ChatGPT a ‘plagiarism machine.’ When ChatGPT gives answers, we don’t know where the sources are coming from, and it does not give a reference of where the information is pulled from. You can ask at the prompt to give references, but both those lineages and tracking still needs to be validated.
I’m afraid there will not too many uses for it for our Enterprise business clients in their domains, but there may be other workloads which are not affected by this fatal flaw of the outputs in a particular domain not yet validated by the right domain experts, as that is where the ground truth really is from. In the meantime, for the Enterprise, we have to assume this is a ‘plagiarism machine’ until it is validated properly. This validation is most likely where those who want to monetize should focus their efforts.
If you're looking for support from me, here are a few options:
Enterprise Data Science Consultancy: With my consult team comprised of a Senior Data Scientist, Senior ML Engineer, Senior Data Engineer, and Senior Cloud Engineer, we will help you architect and build your Enterprise Data Science platform, and transfer knowledge to your IT team to maintain and optimize it. We will also overlay an MLOps framework to manage the AI solutions you build on this platform. If you don’t have an MLOps team, we will help you build one. Please get in touch about this consultancy here
Coaching and Mentorship: I offer coaching and mentorship; book a coaching session here