1.  Biggest Cloud Providers Will Drive AI

Google, Microsoft, and Amazon have the most incentive to push AI because it is ultimately about consuming paid cloud resources (in this case a premium resource) in larger and larger amounts. Amazon is the biggest cloud here, but Google and Microsoft have shown the most innovation so far.

2. AI Models Will Continue to Evolve at a Dizzying Pace

It seems Google and OpenAI try to leapfrog each other every three months. In recent months GPT4.0 Turbo has been launched for prototype tasks. Google Gemini will have an impact in early 2024. Other names to look out for in 2024 are Baidu ERNIE, Amazon Olympus, Anthropic’s Claude 2 (or Claude-Next).  New big models from Meta are being trained now and GPT-5 may also appear in the 2024 timeframe for production use cases and likely with limited use in the Summer 24. That is incredible from an innovation perspective but will present enterprises with a huge headache to keep up.

3.  Vector Search Will Mature Significantly with Increased Deployment

The potential applications of vector search are vast and continue to expand. Recent advancements in machine learning and algorithms have made vector search more efficient and scalable than ever before. This has removed some of the technical barriers that limited its adoption. As the technology continues to evolve and become more accessible, we can expect to see even more widespread adoption this year.

4.  Retrieval Augmented Generation (RAG) Will Prove Itself as an Architecture

RAG is being used to mitigate the LLM hallucination problem and facilitate generative answers over secure data by retrieving factual and contextually relevant information for a model to employ while respecting access controls. Organizations with any amount of subscription access to authoritative and unique content will want to offer RAG, either through their search bot or via a chatbot. I expect RAG to be a standard offering of all search and hosted AI offerings in 2024.

5.  The Economics of Embedded AI Will Become Clearer – But Just a Bit

Prices continue to plummet for pay-per-sip use of large AI models. Making the economics of AI-driven applications at least seem more reasonable. Competitive forces also help. But we do not know enough about the finances behind OpenAI and Google to know where the dust will settle, and where AI companies can make a lot of money while consumers of AI can gain benefit to offset the cost. As new models become available and at pre-existing cost levels, the legacy models will get cheaper and cheaper to access.

6.  Multi-Modal AI Models – Combining Text and Images – Will Become the Norm

When you read a document that includes images, the content should seamlessly contribute to your brain’s understanding of the knowledge. That’s why multi-model models are likely to become the norm for knowledge extraction. A real estate property description may not mention much about the garden or about the modernness of a kitchen, but Multi Modal models will be able to pull out features of images in the listing and ‘sell’ based of that data. For example, AI might suggest “The garden is ideal for making your own stamp on it,” or “The kitchen is ripe for refurbishment,” all based on analysis of images of the property.

7.  Generative AI Will Help Consumers Visualize their Future State

As per the example above, generative AI will be able to take a current photograph of an existing kitchen and generate images of distinctive designs in different styles. Or maybe generate different scenarios of a garden makeover on the fly from pre-existing images of a drab garden.