9 AI Trends to Watch in 2024: From Reality Checks to Shadow AI

We're a little ways into 2024 now, and the pace of AI certainly isn't slowing down. But where will it be by the end of the year? Well, I've put together 9 trends that we expect to emerge throughout the year. Some of them are broad and high level, and some are a bit more technical. So let's get into them. Oh, and if you've stumbled across this video in 2025, let us know how we did.

Trend #1: The Year of the Reality Check

This is the year of more realistic expectations. When generative AI first hit mass awareness, it was met with breathless news coverage. Everyone was messing around with Chat GPT, DALL-E, and the like. Now that the dust has settled, we're starting to develop a more refined understanding of what AI-powered solutions can do. Many generative AI tools are now being implemented as integrated elements rather than standalone chatbots. They enhance and complement existing tools rather than revolutionize or replace them. Think Copilot features in Microsoft Office or generative fill in Adobe Photoshop. Embedding AI into everyday workflows like these helps us better understand what generative AI can and cannot do in its current form.

Trend #2: Multimodal AI

Generative AI is extending its capabilities in multimodal AI. Multimodal models can take multiple layers of data as input, and we already have interdisciplinary models like Open AI's GPT-4v and Google Gemini that can move freely between natural language processing and computer vision tasks. For example, users can ask about an image and receive a natural language answer, or ask for instructions to repair something and receive visual aids alongside step-by-step text instructions. New models are also bringing video into the fold. Multimodal AI allows models to process more diverse data inputs, expanding the information available for training and inference, such as ingesting data captured by video cameras for holistic learning.

Trend #3: Smaller Models

Massive models jump-started the generative AI age, but they're not without drawbacks. According to one estimate from the University of Washington, training a single GPT-3 size model requires the yearly electricity consumption of over 1,000 households. You might think, sure, that's training, but what about inference? A standard day of Chat GPT queries rivals the daily energy consumption of about 33,000 households. Smaller models, meanwhile, are far less resource-intensive. Much of the ongoing innovation in LLMs is focused on yielding greater output from fewer parameters. GPT-4 is rumored to have around 1.76 trillion parameters, but many open-source models have seen success with model sizes in the 3 to 17 billion parameter range.

In December last year, Mistral released Mixtral, a mixture of experts (MoE) model integrating eight neural networks, each with 7 billion parameters. Mistral claims Mixtral outperforms the 70 billion parameter variant of Llama 2 on most benchmarks at six times faster inference speeds and even matches or outperforms Open AI's far larger GPT-3.5 on most standard benchmarks. Smaller parameter models can be run at lower cost and locally on many devices, like personal laptops.

Trend #4: GPU and Cloud Costs

The trend towards smaller models is driven as much by necessity as by entrepreneurial vigor. The larger the model, the higher the requirement on GPUs for training and inference. Relatively few AI adopters maintain their own infrastructure, putting upward pressure on cloud costs as providers update and optimize their infrastructure to meet gen AI demands, all while everyone scrambles to obtain the necessary GPUs to power the infrastructure.

Trend #5: Model Optimization

This past year, we've seen the adoption of techniques for training, tweaking, and fine-tuning pre-trained models, like quantization. You know how you can reduce the file size of an audio or video file just by lowering its bitrate? Well, quantization lowers the precision used to represent model data points, for example, from 16-bit floating point to 8-bit integer, reducing memory usage and speeding up inference. Rather than directly fine-tuning billions of model parameters, Low-Rank Adaptation (LoRA) entails freezing pre-trained model weights and injecting trainable layers in each transformer block. LoRA reduces the number of parameters that need to be updated, dramatically speeding up fine-tuning and reducing the memory needed to store model updates. Expect to see more model optimization techniques emerge this year.

Trend #6: Custom Local Models

Open-source models afford the opportunity to develop powerful custom AI models trained on an organization's proprietary data and fine-tuned for their specific needs. Keeping AI training and inference local avoids the risk of proprietary data or sensitive personal information being used to train closed-source models or otherwise passed to third parties. Using techniques like Retrieval Augmented Generation (RAG) to access relevant information, rather than storing all of that information directly within the LLM itself, helps to reduce model size.

Trend #7: Virtual Agents

Virtual agents go beyond the straightforward customer experience chatbot. They relate to task automation, where agents will get stuff done for you, make reservations, complete checklist tasks, or connect to other services. There's lots more to come in this area.

Trend #8: Regulation

In December of last year, the European Union reached a provisional agreement on the Artificial Intelligence Act. The role of copyrighted material in the training of AI models used for content generation remains a hotly contested issue. Expect much more to come in the area of regulation.

Trend #9: Shadow AI

Shadow AI refers to the unofficial personal use of AI in the workplace by employees, using gen AI without going through IT for approval or oversight. In one study from Ernst and Young, 90% of respondents said they used AI at work. Without corporate AI policies in place—and, importantly, policies that are observed—this can lead to issues regarding security, privacy, and compliance. For example, an employee might unknowingly feed trade secrets to a public-facing AI model that continually trains on user input, or use copyright-protected material to train a proprietary model, exposing the company to legal action. The dangers of generative AI rise almost linearly with its capabilities, and that line's going up. With great power comes great responsibility.

So, there you have it: Nine important AI trends for this year. But why nine? Don't these things almost always come in tens? Why, yes, yes they do. And that's YOUR job. What is the one AI trend for 2024 that we haven't covered here? The missing 10th trend. Let us know in the comments.