sequenceDiagram
participant User
participant LLM
User->>LLM: Prompt: What is the capital of France?
LLM->>User: Response: The capital of France is Paris.
Fall 26
To prompt an LLM with text you type your question and submit it
sequenceDiagram
participant User
participant LLM
User->>LLM: Prompt: What is the capital of France?
LLM->>User: Response: The capital of France is Paris.
The model responds with some text.
Example: ChatGPT 3.5 (Nov 22)
Prompts can include
Responses can include: Text, Image, Audio, Video
Example: Gemini 2.0 Flash (Dec 24)
To prompt the LLM you speak into a microphone on your computer
sequenceDiagram
participant User
participant LLM
User->>LLM: Audio Prompt 🔊: Can you hear me?
LLM->>User: Audio Response 🔊: Yes! I hear you? How can I help you today?
The model responds in audio and/or text that you can hear and see immediately
Using a webcam you speak and show the model what you want
sequenceDiagram
participant User
participant LLM
User->>LLM: Webcam Prompt 📹+🔊: Can you see and hear me?
LLM->>User: Audio Response 🔊: Yes! I can both see and hear you. How can I help you today?
The model responds in audio and/or text
In Spring 25 a new model called Deep Seek caused Wall Street to panic and lose Billions of dollars. The reason is the Deep Seek was released as an open source model that anyone could download and run on their own, thus challenging the idea that only companies with proprietary models would dominate the AI space. The models from OpenAI, Anthropic, Google all are proprietary models:
These models cannot be run by anyone the way Deep Seek can.
Deep Seeks models are competive with the state of the art closed source models but are created at a fraction of the cost.
Notice the difference when posting prompts to the standard V3 models versus the “reasoning model” (Button “DeepThink”). The reasoning model shows you its thinking process.
Some examples of prompting and their costs using openrouter:
LLMs can call Tools (external programs)
This ability is sometimes called agentic.
Examples (Tools):
On the following slides we show a web application with no AI, and then show how it turns into a modern AI application.
This is the FIT Retail Index (non-AI). It uses financial data for retail companies and lets you compare one company against another, or against the segment it belongs to. There is no AI anywhere in this version.
This is the FIT Retail Index (AI version) which now has LLM Chat added.
Two things are new:
The AI can drive the user interface and access the same data.
Strudel.cc is a musical language for live coding and DJing It is based on a language called Tidal Cycles which is written in a programming language called Haskell. It is compact, technical, and mathematical language.
It is amazing in capability.
Examples:
Call and Response
Chords And Melodies
Claude working with Patterns/Channels:
More Claude.ai to FL Studio (Mac):
DeepSeek working with a DAW: