Survey: Large Language Model Adoption Reaches Tipping Point
With a dizzying array of research papers and new tools, it’s an exciting time to be working at the cutting edge of AI. Given that the space is so new, some remain skeptical that LLM-powered applications are enterprise-ready – or that they can be deployed safely and reliably.
How common is this concern – and is it slowing the development of LLM systems? To answer, we surveyed over 350 AI engineers, data scientists, developers, technical business executives, and others in September of 2023.
Here are some major takeaways from the survey.
LLM Adoption Is Accelerating
A lot can change in five months! In all, 61.7% of developers and machine learning (ML) teams now have or are planning to have an LLM app in production within a year – up from 51.7% in April. Over one in ten (14.7%) are already in production, compared to 8.3% in April.
The Foundation Model Market Is Getting More Competitive
While OpenAI still dominates with 59.1% of technical teams relying on the company’s LLMs, Meta’s Llama 2 and other alternatives are becoming more popular. In the “Other” category, Google PaLM 2 leads in adoption (20.7% of those surveyed), followed by Databricks (Dolly) at 14.9% and MosaicML at 5.6%.
Barriers To Adoption Are Evolving
Concerns about data privacy, needing a business case, or responsible deployment are trending downward while barriers like “require on-prem” and “accuracy of responses and hallucinations” are up from the April survey – both likely pointing to the seriousness of adoption and need for tools around governance and LLM observability.
Builders Don’t Back AI Regulation
Overall, 43.5% of technical teams prefer to hold off on new AI regulation or to better enforce existing regulations, while 42.4% are undecided or neutral. Only 14.1% want more regulation, perhaps suggesting a difference of opinion between the typical rank-and-file AI engineer and more senior executives who are often actively shaping regulatory agendas.
Open Source Or Proprietary
Most developers and ML teams surveyed prefer a third-party public API, followed by a proprietary fine-tuned model.
What Are Teams Implementing?
Intuitively, most teams (60.6%) using LLMs are implementing prompt engineering. Adoption of LLMOps tools also appears to be accelerating, with 40.9% of teams saying they use a vector database and 30.1% reporting that they leverage LLM observability (i.e. Arize).
Use Cases: Viva La RAG
Perhaps unsurprisingly, retrieval augmented generation is the most common use case among teams planning to leverage LLMs.
While the space is rapidly evolving, this survey makes one thing abundantly clear: LLM adoption is not a passing fad. As the industry shifts, the need for LLM observability and other tools to ensure companies can maximize its benefits likely takes on heightened importance.