A year ago, a congregation in San Francisco invited many people to “The Woodstock of AICollected 5,000 devotees of “open source” AI models—that is, where the underlying code, and sometimes the model weights and training methods, are publicly available for researchers and developers to develop.
The event, hosted by open-source AI hub Hugging Face and featuring live llamas (referring to Meta’s Llama model), kicked off an open-source AI boom that hasn’t stopped since. The landscape now includes unicorn startups like Mistral and Together AI, and boasts a steady barrage of new open-source AI models that come close to beating OpenAI’s flagship GPT-4 on various performance benchmarks. They are becoming more and more. Over the past few weeks, open source LLMs have been released from top companies like Databricks, Cerebros, AI21, and Coher.
However, a recent survey by venture capital firm a16z found that for large companies adopting generative AI, OpenAI’s closed, proprietary models have so far been the most popular — especially for use cases that actually I am in production. But it also showed signs of change: Six months ago, for example, most organizations were experimenting with just one model — mostly from OpenAI — and mostly common in areas like marketing, coding, and customer support. Stuck in use cases. But in 2024, they’re opening up to experiment with more AI model options—often open source.
More organizations are experimenting with open source models.
Sarah Wang, general partner at A16z who co-authored the survey, said that OpenAI’s biggest advantage so far has been the first mover. In addition, it was difficult to top it off, he explained, because for most of the last year, GPT-4 has been considered the best model available, as well as directly through or through the API. Direct access is easy. Microsoft Azure.
“I think it was easiest to plug and play and say this model is great, let’s see what the use cases come up with,” he said. The survey estimates the 2023 market share of closed-source models at 80%–90%, with the majority going to OpenAI. No updated market share was provided for this year, but 46% of respondents said they prefer or strongly prefer open source models.
“Each enterprise said they were testing more than one model family,” Wang said, noting that two of the top six model families in use are open source—Llama and Mistral. “So certainly it’s still early, but I think it’s probably an important indicator for use down the road,” he said.
OpenAI has made ‘tremendous growth’ in enterprise offerings.
Perhaps with its eye in the rear-view mirror and its foot on the gas, OpenAI, led by CEO Sam Altman, is working hard to solidify its lead with corporate customers. It published a new blog post last week announcing new features for its “self-serve fine-tuning API” — which allows for some customization — and shared case studies from companies like SK Telecom have customized and fine-tuned OpenAI models. The post also announced an expanded “support fine-tuning offering” for companies to “collaborate with OpenAI technical teams to leverage techniques beyond the fine-tuning API.”
Also, OpenAI COO Brad Lightcap mentioned the ‘tremendous growth’ in the enterprise version of ChatGPT in an interview with Bloomberg on Friday — claiming that more than 600,000 people have now signed up to use ChatGPT Enterprise. What’s up, up from about 150,000 in January, and 2024 ‘the year of AI adoption in the enterprise.’
Still, the market for generative AI models is fragmenting based on enterprise use cases, according to Kjell Karlsson, head of AI strategy at Domino DataLab and former Forrester analyst.
“OpenAI has a dominant position thanks to its head start but more than that, its relationship with Microsoft. and their powerful sales teams,” he said, referring to OpenAI’s partnership with Microsoft, which includes offering OpenAI models through its cloud arm, Azure. OpenAI models are increasingly being used for common use cases such as ad hoc user queries or customer service chatbots when companies are trying to build various generative AI applications and paradigms — such as biotech businesses. Companies that conduct AI-powered drug discovery — and want to protect their data due to regulatory or security concerns — are often turning to other vendors and open-source models.
“I have yet to speak to a company that says they use OpenAI’s models because of any inherent technological advantages they have today,” he said.
Cost, control, and customization
While cost is often cited as a reason for turning to open source—for example, Meta’s Llama 2 was shown to be 10-20 times cheaper than OpenAI’s GPT-4 to generate 1 million tokens. Hai-Wang said respondents widely cited other reasons. For willingness to adopt AI models. These reasons include control (protecting proprietary data and understanding why models produce certain outputs) and customization (the ability to effectively fine-tune for a given use case).
“The fact that you can self-host a fine-tuned model on your own data with an open source model was very attractive to a lot of enterprises,” he said.
Ali Ghodsi, CEO of data and AI platform Databrx (which recently released a powerful new open-source large-scale language model called DBRX), agreed with this assessment, noting that the wholesale initiative for open-source AI in 2024 termed as an “under-reported phenomenon”. Enterprises want to customize AI models to their specific data and tasks and consequently own the intellectual property, he added.
“I think that will continue regardless if there are really smart models that come out by proprietary vendors,” he said. “Enterprises want to be competitive in their market. They want to create their own recipes.”
Potential effect of GPT-5
Of course, none of the current predictions about corporate adoption of generative AI take into account the release of OpenAI’s next major language model, the highly anticipated GPT-5, which Wang says is coming very soon. However, he pointed out that in discussions with enterprise companies, the cost of switching models is very low—so it’s likely that organizations will continue to experiment with a mix of closed and open-source models.
“They can change models very easily on the back end,” he said, so companies don’t have vendor lock-in issues with something like a database business. Plus, the landscape is getting more crowded with newcomers, so people are open to testing.
That said, GPT-5 could come out and “blow everyone away,” Wang admitted, adding that OpenAI maintained its market share. “It can throw everything. It’s hard to predict.”