{gpt-4-vision-preview}
DigitalPurpose.ch
// Measure the distance between two points
const distanceBetween = ( p1, p2 ) => {
const dx = p1[0]-p2[0];
const dy = p1[1]-p2[1];
return Math.sqrt( dx*dx + dy*dy );
}
distanceBetween([10,10], [50,50])// OpenAI Non-vision
{ "model": "gpt-3.5-turbo",
"messages": [
{"role": "system", "content": "Please fix my truncated .json file"},
{"role": "user", "content": " {\"field1\": 0, \"field} "}
] }
// OpenAI Vision
{ "model": "gpt-4-vision-preview",
"messages": [{
"role": "user", "content": [
{"type": "text", "text": "What are in these images? Is there any difference between them?"},
{"type": "image_url", "image_url": {"url": "https://image_url.jpg", } },
{"type": "image_url", "image_url": {"url": "https://image_url.jpg", } }
] } ] }Using the Vision API Endpoint
Classifier: When Is Tennis Being Played?
0-Shot Classifications: {"299": 0, "252": 0, "69":1, "148": 1}
System Prompt: "Classify each sub-frame as PLAYING (with label 1) or NON-PLAYING (with label 0):
PLAYING scenes only show the tennis court from the standard broadcast angle, fully visible and vertically & horizontally aligned with the court lines (the audience should at most be barely visible).
NON-PLAYING scenes show any other content that is not a standard broadcast view of the court, including low-angle, court-side, zoom-in, or audience shots. Respond just in a JSON using the centered number (white text with black outline) in each square as the index for each frame."
Extract Structured Data: Reading < Seeing
Founding Team Keith Corso Co-Founder & CEO Phil Dunn Co-Founder & CPO Met with 500+ Transportation Directors in the Fmr. Tech Dir. for Broward Schools, 6th past 5 yrs largest U.S. district Allocated $1OOM+ of K12 technology Presented to 750+ Transportation Directors at spend annually school bus conferences in the U.S: Applied Math at Cornell, Quantitative Fmr. investor at MassMutual Ventures Methodology at Columbia, and Law at Seton Hall University. BusRight's team members consist of both talented technologists and experienced industry practitioners. Sc Penn COLUMBIA Coitn nntan COLLEGE 4TW Antora Ventures MassMutual Google Wiai elabkhed 1915 Department of NVIDIA BROWARD NC Education robinhood County Public Schools PUBLIC _ wich , "RSONALIZCD 'LEARNIN
{
"FoundingTeam": {
"Member1": {
"Name": "Keith Corso",
"Title": "Co-Founder & CEO",
"Achievements": [
"Met with 500+ Transportation Directors in the past 5 yrs",
"Presented to 750+ Transportation Directors at school bus conferences in the U.S.",
"Fmr. investor at MassMutual Ventures"
],
"Education": []
},
"Member2": {
"Name": "Phil Dunn",
"Title": "Co-Founder & CPO",
"Achievements": [
"Fmr. Tech Dir. for Broward Schools, the 6th largest U.S. district",
"Allocated $100M+ of K12 technology spend annually"
],
"Education": [
"Applied Math at Cornell",
"Quantitative Methodology at Columbia",
"Law at Seton Hall University"
]
}
},
"TeamDescription": "BusRight's team members consist of both talented technologists and experienced industry practitioners.",
"AffiliationsLogos": [
"Penn",
"Columbia",
"Cornell",
"MassMutual",
"Google",
"NVIDIA",
"Broward County Public Schools",
"NC Department of Education",
"robinhood"
]
}
→
GPT4v Uses-Cases & Benefits
✅Can do magic with text & documents (images so-so)
✅Quick Classifier Prototyping (0-shot ~90% reliability)
✅Delegate Dataset Labelling / Enrichment
❌Can't Customise Model — ✅but don't need to? 🤞🍀
❌Configure Own GPU Server vs. ✅Rely on Microsoft API
❌API Rate Limits 😅 (500 requests per day)
{gpt-4-vision-preview}
By Jay Teach
