Flavors of AI

Not all AI is the same. And the differences matter more than most people realize.

When people say “AI,” they usually picture one thing. A familiar prompt to ask any question, a filter to make you look better, or maybe even a self-driving car. But AI isn’t a single technology. It’s a family of very different tools, each built to do something distinct. And many of these technologies have been around since long before the current LLM (large language model) or WAYMO craze.

If you’ve ever looked at your credit card app for the free credit score, you have seen Predictive Analytics. PA isn’t just used in a FICO score, though. It is in anything that uses historical data to forecast what happens next. Customer churn, fraud detection, equipment failure. Predictive Analytics is the workhorse behind recommendation engines, supply chains, and insurance pricing. When Netflix tells you that you might like that new British detective series, it is using predictive analytics to do it.

Beyond using “Face ID” to unlock your phone, have you ever sat back and considered how Computer Vision, or AI that sees things, has replaced fingerprints with “face prints?” Even disguised or from obtuse angles, it analyzes images and video detecting objects, reading text from photos, and identifying faces for potential terrorist threats at Customs. It can flag anomalies in medical scans or on assembly lines as well as it does potential criminals trying to disguise themselves.

Whenever you have to type the swear word starting with “f” over and over because your phone keeps saying “duck,” you have seen Natural Language Processing. While “duck” was forced into the models by Google and Apple, NLP long predates LLMs like ChatGPT. It remains its own discipline. Classifying sentiment and pulling key entities from unstructured text. NLP applications use smaller, faster, cheaper models purpose built for specific tasks like helping complete sentences with the words you are likely to choose next statistically, which is not necessarily what you intended to say.

In 2018 Amazon used their own programmer hiring data to screen new applicants and it downgraded female applicants because previous programmer hiring had been male dominated. Deep Learning powers everything from voice assistants to computer Chess Masters to rapid drug discovery. The tradeoff is that these models need enormous training data, and they can absorb biases baked into that data without anyone noticing until the outputs go sideways.

You’ve heard of ChatGPT, Claude, Gemini. These are deep learning models trained on vast text to predict what comes next in a sequence which makes them Large Language Models that can write, summarize, translate, code, reason through problems at times, and hold extended conversations. They don’t “understand” language the way you do. They are exceptionally sophisticated pattern completers. But the practical gap is shrinking if you’re willing to put up with a few mistakes.

Have you ever been duped by a fake podcast that’s actually a giant advertisement for a product? Or more excitingly, have you ridden in a Waymo self-driving taxi? That’s Multimodal AI. The latest systems process audio, images, and video alongside language. You can now speak to an AI, hand it a photo, or feed it a recording and get a transcript with analysis. Voice capable models are not speech to text bolted onto a chatbot. They process tone, pacing, and context in ways that make the interaction feel qualitatively different.

Where will medical practices find the most AI value long term? What if you could clone junior versions of providers and key staff to help them get their massive amount of work done fast enough to make enough money to survive? Agentic AI is the bleeding edge of AI tech today. Think of it as the difference between asking someone a question and handing someone a project. Agents can do more than answer questions. They can now take action, representing a fundamental shift from AI as a tool you use to AI you can delegate to.

We’re excited to leverage all the AI types, but creating Agents to work beside you offers the most promise of managing growing workloads with static or negative reimbursement changes. The next time a vendor tries to woo you with claims that they use AI you’ll be able to see if what they have is just a minor convenience or a major revolution. At Venture, we’re swinging for the fences. We want to permanently ease your workload.

Come talk to us about how we can help you.

‍