As everyone rushes to use OpenAI models, build atop its APIs, or use services like ChatGPT in their workstreams, many wonder if the dawn of super abundant (conversational) AI is already here.
As someone working in NLP and Deep learning in a non-MAANG or non-startup-born-in-today's-digital-era, my view is that adoption of Large Language Models (LLMs), even open sourced ones, in existing service/product industries (such as healthcare, publishing, etc), will be slower than you think
The idea of LLMs have been around for a while, most notably since BERT models in 2018, but I've seen resistance to its adoption in some companies. One might argue that the recent ChatGPT wind has woken up management professionals to the possibilities of LLMs (although I have heard too many people refer to LLMs and Generative AI and ChatGPT interchangeably - and that worries me). However the integration of such models into company workflows has been slow. This is due to a number of reasons.
Being risk averse due to the giddying pace of development. Model architectures, model artifacts (e.g, different flavours of GPT models), and tools built atop these approaches is a very dynamic space, with reports of companies often evaluating build-vs-buy decisions on a weekly basis. Many companies are waiting for waters to still, and stable models to emerge before adoption.
Being risk averse to non-deterministic nature and lack of interpretability of generative AI (e.g. in healthcare with strict regulations and costs of mis-predicting)
Lack of culture/expertise in being able to run dynamic experiments at scale. Many industries/management decisions are top-down, and not conducive to bottoms-up approach. Thus experimentation is not always encouraged/financed, which makes it harder to nurture inhouse experience for productionisng LLMs.
Hardware requirements are daunting, and often unaffordable, often requiring multiple GPUs operated in a distributed manner. This is formidably expensive for many, and is difficult to get budget approval of that kind of hardware requirements in experimentation phases of projects. (Note there are several third party efforts focused on simplifying distributed training/inference)
Within a single company, LLM applications are developed in silos, making reuse and maintenance hard. Companies that invest in managed ML services/platforms are able to produce off the shelf ML components. In the absence of such centralised efforts, it is challenging for individual teams to secure the correct resources/expertise to quickly & reliably apply LLMs.
Mandatory disclaimer: I do see the winds of change, and believe the adoption will improve soon.