OpenAI has been going through a lot of ups and downs ever since it released ChatGPT. Acquiring millions of users, then a dip in revenue, reporting losses, and then finally releasing ChatGPT Plus and Enterprise to earn a little bit of revenue. But it seems like the company has been successively going under the bus when it comes to sales, as people are looking for other cheaper alternatives.
For enterprise, GPT-4 is 50 times more expensive than Llama 2, specifically for summarisation of the Wikipedia text into half its size. Imagine, for countless other use cases.
Meanwhile, just a year post-launch of ChatGPT, Salesforce and Wix, who were its early customers, have now decided to explore other options. They’re now frolicking with rival AI providers, touting cost-effectiveness and finding the alternatives budget friendly.
Salesforce has been one of the first customers and poster boys to use OpenAI’s GPT-4, using them to automatically draft emails or distil endless meeting chatter into manageable snippets. But now, Salesforce is also eyeing open-source models and creating their in-house creations, such as Einstein GPT, which rumour has it, are less expensive for them.
Jayesh Govindarajan, the senior vice president of AI at Salesforce, puts it, “we’re at the very beginning of this cost-reduction exercise in AI. It’s only going to become more important as these AI products reach greater scale and we begin to focus on achieving cost effectiveness.”
Even Zoho, which currently uses OpenAI products, is working on developing its own cost-effective, proprietary LLMs, and is in talks with NVIDIA for GPU acquisitions.
Morgan Stanley, another flagship OpenAI customer, is also testing out Microsoft Azure’s other offering, exploring whether the Azure service is a long-term match for OpenAI’s. Wix, too, once a dedicated OpenAI fan, is eyeing the competition, testing open-source models and even Google’s offerings to cut costs. The list goes on and on.
What’s the issue with ChatGPT Enterprise?
When OpenAI announced that it would release ChatGPT Enterprise soon, we said that it would fail, as Microsoft was just using OpenAI to do the dirty work. But when OpenAI actually launched it, it seemed as though the company had finally learned how to do business, directly from Microsoft.
Most LLM applications don’t need a generalist model.
For those, you’ll save a ton of money and get better accuracy by fine-tuning something like Mistral-7B.— Mark Tenenholtz (@marktenenholtz) October 17, 2023
But now, Microsoft, the so-called OpenAI backer, is directly choking ChatGPT Enterprise by the neck. According to its recent earnings report, the tech giant’s revenue has risen 13% to $56.5 billion, as the sale of its AI cloud through Azure has accelerated, all thanks to OpenAI’s GPT.
When customers buy OpenAI through Azure, Microsoft snags a fatter slice of the pie. Moreover, enterprises have been considering the costs of building LLMs, and are finding open source models cheaper for them, that includes going through Microsoft.
Though it is not entirely clear if Microsoft’s cloud sale through AI investment is just because of OpenAI’s offering, or if it also includes other open source offerings or not. But Microsoft is dabbling in the open-source playground to cut costs as well. For example, it also provides Meta’s Llama 2 on its cloud platform. Developers are finding that these open-source offerings can replace OpenAI’s models for less demanding chores.
On the other hand, while its revenue has skyrocketed as Sam Altman said, and the company is on its way to the billion-dollar club, most of that moolah for OpenAI comes from ChatGPT Plus subscriptions. It is still learning the ropes in the enterprise world.
When it comes to open source, OpenAI had already raised the alarm on open source AI dangers long back. This was probably because it realised that these models would turn out to be dangerous for the company.
LLMs' hallucination is a major problem we face in enterprise AI application. While it is not a big deal if you occasionally get some incorrect answers when you chat with ChatGPT, it would be embarrassing if AI outputs nonsense when reviewing a legal document.
— Fei Li (@lifeitech) September 30, 2023
Apart from this, LLM hallucinations remain one of the biggest problems for ChatGPT Enterprise. It might be fine for a chatbot to hallucinate when asking random questions, but when it comes to handling legal and financial documents, hallucinations cannot be touted as a feature, but a bug.
But OpenAI is continually fixing the hallucinations problem, and its autonomous AI agents, which are touted to be released next month, would make it all possible.
Busy impressing customers, not enterprise
Recently, OpenAI made DALL-E 3 available for all the ChatGPT Plus and ChatGPT Enterprise users. But, enterprise customers are still not impressed as they crave for more specialised business use cases, and not generalised.
OpenAI isn’t entirely on the losing side. Some customers of other AI services (AWS Sagemaker, Google’s Vertex AI, etc.) are seeking more variety. Fidelity Investments, for instance, who was loyal to Amazon’s SageMaker, has started testing with Microsoft’s Azure OpenAI Service to give OpenAI’s models a whirl. But there are still a lot of use cases that OpenAI has to explore directly, and not via Microsoft.
The Altman led firm is also at the risk of being outshone by open-source models, which are smaller and simpler but pack enough punch for many tasks. Mistral AI’s new models have been also outperforming OpenAI’s generalist models, and enterprises are utilising them.
For instance, Pete Hunt, the founder of developer tools startup Dagster, switched from OpenAI’s GPT-3.5 model to a nifty open-source model from Mistral AI, for his Summarize.tech service, saving a bundle without compromising quality. He’s now eyeing that elusive mortgage payment, thanks to open-source cost savings.
Same goes for Oracle’s billion-dollar baby, Cohere, which is seeing seamless enterprise-wide adoption at scale. OpenAI’s ChatGPT is nowhere to be found in enterprise, except in Microsoft Azure OpenAI Service. Even the early adopters are yet to figure it out.
The post What’s Up with ChatGPT Enterprise appeared first on Analytics India Magazine.