Marking the launch of ChatGPT In the era of large linguistic models. In addition to OpenAI’s offerings, other LLM programs include Google’s LLM family (including Bard), the BLOOM project (a collaboration between groups at Microsoft, Nvidia, and other organizations), Meta’s LLaMA, and Anthropic’s Claude.
No doubt more will be created. In fact, A April 2023 Ares Survey It found that 53% of respondents planned to publish their LLMs within the next year or sooner. One approach to doing this is to create a “vertical” MBA that starts with an existing MBA and carefully retrains it in domain-specific knowledge. This tactic can work in life sciences, pharmaceuticals, insurance, finance, and other business sectors.
Deploying an LLM can provide a powerful competitive advantage—but only if it’s done well.
LLMs have already led to newsworthy issues, such as their tendency to “hallucinate” incorrect information. This is a serious problem, and can distract leadership from core concerns with the processes that generate those outputs, which can be similarly problematic.
Challenges of LLM training and deployment
One of the problems with using LLMs is the huge running overhead because the computational demand to train and run them is very intense (they’re not called large language models for nothing).
LLM degrees are exciting, but developing and accrediting them requires overcoming many feasibility hurdles.
First, the hardware needed to run the models is expensive. Nvidia’s H100 GPU, a popular choice for LLM degree holders, was sold on the secondary market for about $40,000 per chip. One source estimated it would take approx 6000 slices To train LLM similar to ChatGPT-3.5. That’s nearly $240 million on GPUs alone.
Another important cost is the operation of those chips. It is estimated that simply training the model would require approx 10 gigawatt-hour (GWh) of energy, equivalent to the annual electrical use of 1,000 American homes. Once the model is trained, the cost of electricity will vary but may become prohibitive. This source estimated that the energy consumption to run ChatGPT-3.5 is about 1 GWh per day, or the combined daily energy use of 33,000 households.
Power consumption can also be a potential pitfall of user experience when running LLMs on mobile devices. This is because heavy usage on the device can drain its battery very quickly, which can be a major barrier to consumer adoption.