AI – we’ve all been inundated with the subject. I’m sure everyone has used ChatGTP. But where is it going? (This article was written by a human but with augmentation and assistance by AI, which is how it should be…)
There are several major trends to keep an eye on: Multimodal AI, Agentic AI, on-device AI, enterprise AI assistants, the surge in open source and AI+custom data (“RAG”).
Let’s break each one down.
Multimodal AI – These are models that handle text, images, audio and video inputs and outputs. This enables a more natural, human-like interaction and broader use, like in customer service and education. ChatGPT’s 4-o (“omni”) works this way. It can understand images, speech, even tone of speech.
Agentic AI – An incredibly powerful next wave of AI, these are agents that act autonomously to do things for you. Think of giving an AI agent a goal (“find the best price for eggs locally” or “schedule a time with Jack to meet”) and having that AI agent perform the actions to get to that goal. This moves from tools like ChatGTP to actual assistants that can do things for you. Thing of OpenAI’s AutoGPT, Meta’s Llama Agents, ReAct frameworks. Of note: ChatGTP Operator is itself a very junior version of Agentic AI. It just uses a browser combined with 4-o’s multimodal functionality to do things on the web for you; agentic AI is a true assistant that could go on the web to research something, open Outlook to set a calendar appointment, make a call for you, etc. It’s usable for things like automating workflows, customer service, software development, tasks, etc.
On-Device AI – These are AI models that run on a device – like a phone, or the rumored AI device that OpenAI just acquired.
Enterprise AI and AI assistants – Organic implementation of AI in your apps, and you’re already seeing this and you’ll be seeing a lot more of it. This is AI being embedded thoughout the technology stack, whether it’s in your CRM, Office 365, GitHub Co-Pilot, Google Duet, etc.
The explosion of open source – Another key development in AI is the explosion of open source tools. This is a key development to democratize AI for all – meaning, you could download an open source LLMÂ (Large Language Model – like ChatGTP) package and start your own LLM inside your company, running privately. Hugging Face is a good place to start in evaluating open source models. Other key players are Meta (Llama 3), Mistral and Stability AI.
AI + Custom Data / Retrieval-Augmented Generation (RAG) – Okay, you can now sound cool talking at a party talking about “RAGs”. So what is it? Pretty simple, actually: it’s combining LLMs with private stores of data. For example, you could combine a customer service agent based on an LLM (like ChatGTP) with internal data to assist your customers. Think of OpenAI (ChatGPT + File Search or custom GPTs), Cohere, Anthropic and the LangChain ecosystem. The real gold rush isn’t just building models, it’s applying RAG on enterprise data to create high-trust copilots.
So there’s your quick look at the landscape of AI! We’ll keep publishing new information to keep all of you informed.
Photo by Possessed Photography
