OpenAI – ChatGPT Enterprise

The feature is the bug. 

• ChatGPT Enterprise looks like a marketing gimmick to make money rather than a serious attempt to offer generative AI services to corporations as, in my opinion, this service is not suited to serve enterprise requirements for generative AI.
• OpenAI has unveiled a new service called ChatGPT Enterprise which not surprisingly is a version of ChatGPT with features tailored towards enterprise use.
• However, it is very far from what RFM Research has concluded (see here) will be a generative AI product that companies can derive real value from.
• ChatGPT Enterprise is an instance (or instances) of ChatGPT (based on GPT-4) that resides in a public cloud (Azure) where many of the restrictions have been removed in return for a higher fee.
• For example, ChatGPT Enterprise is faster, allows longer prompts, has some customisation options, has higher privacy and has no limits on usage by any one user.
• However, it is still a shared instance in the cloud and the model will not be trained on the data of the customer in any way.
• This, in my opinion, is where this proposition falls flat on its face.
• RFM Research has concluded that the most useful generative AI will be a foundation model (such as GPT-4, Bert, LlaMa etc.) which has been fine-tuned on the dataset of the company.
• This will mean that all of the unstructured data in the company will have been ingested and the generative AI will be able to recall it in response to user requests.
• I have long been of the opinion that most companies make use of only a tiny fraction of the data that they have mostly because no one inside the company knows that it is there.
• A generative model trained on the data will know about all the data and be able to surface it on command although double-checking and verification will be required by the human requestor.
• ChatGPT Enterprise will do none of this because it is trained only on generic data meaning that this service will be unable to provide the real value that RFM sees in generative AI for the enterprise.
• Furthermore, using this service will mean priming the AI with long prompts in order to get superior responses which effectively means that company information will be sent into a public cloud.
• OpenAI says that it has taken precautions to prevent this data from leaking but with these models being large black boxes, I suspect that it will be difficult for OpenAI to guarantee that data will not leak.
• Consequently, if I was running a company, I would not trust it.
• Instead, I would have my own instance of the model running inside my own infrastructure to ensure that there is no data leakage which would also allow the company to train the model on its own data.
• This is the enterprise equivalent of running inference at the edge which is where RFM Research sees the bulk of inference taking place in the long-term.
• This is how I think companies can make the most of generative AI and I do not see how a cloud-based version of ChatGPT Enterprise can offer much value to companies beyond that which is already available using GPT-4 at $20 per month.
• Many companies are in a fluster and have decided that they need to use generative AI in their business even if many of them are not quite sure what the use case is and ChatGPT Enterprise is the perfect way to tick this box.
• Hence, in the longer term, as companies figure out what they need generative AI for or whether they need it at all, a much more sophisticated industry is likely to emerge with off-the-shelf foundation models and services to help companies fine-tune and use them on their data.
• Hence, I don’t think that ChatGPT Enterprise has much of a long-term future although it will help OpenAI generate some badly needed revenue to cover its huge compute costs.
• It does however have a future for as long as the current frenzy and FOMO continues to drive the generative AI sector.