The Great AI Explosion - The Migration to AI and Agentic Systems (Part 1 - Introductions)
We go over the explosion of LLM (Large Language Models) and some examples of their current capabilities.

AI (Artificial Intelligence) systems are currently expanding at a exponential rate. They can literally do what takes you weeks in minutes. Many of the basic ones are now free. They are invasive and persuasive for anyone that does anything on the internet. Even though they started off as basic 'chat bots' that were not powerful - they expanded extremely quickly. They are transformers that are about to transform everyone's life.
I was one of the first to adopt Grok 4 when it dropped (literally buying a years access out the gate.) It was to surmise - shocking. What would take hours to code, write or configure would be done in a few minutes (if you prompted it right.) Was it perfect - no but so close out the gate that it literally was only limited by the users imagination. Stuff of your imagination could literally be generated in minutes, and it worked. In short - These AI was about to change everything.
- The concept of existing outside this next AI industrial revolution is like the concept of leaving electricity to go live in a wood cabin. It's just not even remotely feasible.
- Instead the user is strongly encourage to leverage AI as much as possible to improve their own lives, it cannot be stopped.
- Much of this article was assembled with the help of Grok 4 Expert..
Large AI Models (Top 20)
- Costing 100's of millions to build and train these AI models - it is far outside the realm of the basic small company and individual to run in a home setting. You can either subscribe or access through their API. They are banking on millions of subscribers.
Rank | Model Name | Training Cost (USD) | FLOPs | Monthly Cost (USD) |
---|---|---|---|---|
1 | GPT-4.5 / GPT-5 | ~$200-500M | ~1e26-1e27 | ~$20-200 |
2 | Gemini 2.5 Pro | ~$150-300M | ~5e25 | ~$0-19.99 |
3 | Claude 4 Sonnet | ~$100-250M | ~2e25 | ~$0-15 |
4 | Grok-4 | ~$100-200M | ~1e25 | ~$0-15 |
5 | Llama 4 (400B) | ~$50-150M | ~1e25 | N/A |
6 | DeepSeek R1 (671B) | ~$20-50M | ~5e24 | ~$0-5 |
7 | Qwen 2.5 (72B) | ~$30-80M | ~2e24 | N/A |
8 | Mistral Large 2 | ~$40-100M | ~1e24 | ~$0-14.99 |
9 | Nemotron-5 (340B) | ~$80-150M | ~3e25 | N/A |
10 | Nova Pro | ~$50-120M | ~1e25 | N/A |
11 | Mixtral 8x22B | ~$20-50M | ~5e23 | ~$0-14.99 |
12 | Command R+ | ~$40-90M | ~8e23 | N/A |
13 | Phi-4 | ~$10-30M | ~1e23 | ~$0-20 |
14 | Falcon 2 (180B) | ~$30-70M | ~2e24 | N/A |
15 | GPT-4o (multimodal) | ~$100M | ~1e25 | ~$20-200 |
16 | Claude 3.7 Sonnet | ~$80-150M | ~1e25 | ~$0-15 |
17 | Llama 3.1 (70B) | ~$20-50M | ~5e23 | N/A |
18 | DeepSeek V2 | ~$15-40M | ~3e23 | ~$0-5 |
19 | Gemini 1.5 Ultra | ~$100-200M | ~2e25 | ~$0-19.99 |
20 | Grok-3 | ~$50-100M | ~5e24 | ~$0-15 |
If you wonder what you can 'do' with an AI model - here is an example using Grok 4 Expert as this is the one subscribed to:
In under 4 minutes it produced the following:
import numpy as np
import matplotlib.pyplot as plt
def plot_ellipse(a, e, color, label, ax, offset=(0, 0)):
theta = np.linspace(0, 2 * np.pi, 100)
x = a * (np.cos(theta) - e) + offset[0]
y = a * np.sqrt(1 - e**2) * np.sin(theta) + offset[1]
ax.plot(x, y, color=color, label=label)
# Heliocentric orbits data (a in AU, e)
heli_data = {
'Mercury': {'a': 0.387, 'e': 0.206, 'color': 'gray'},
'Venus': {'a': 0.723, 'e': 0.007, 'color': 'orange'},
'Earth': {'a': 1.000, 'e': 0.017, 'color': 'blue'},
'Mars': {'a': 1.524, 'e': 0.093, 'color': 'red'},
'Jupiter': {'a': 5.203, 'e': 0.048, 'color': 'brown'},
'Saturn': {'a': 9.537, 'e': 0.054, 'color': 'gold'},
'Uranus': {'a': 19.191, 'e': 0.047, 'color': 'lightblue'},
'Neptune': {'a': 30.069, 'e': 0.009, 'color': 'darkblue'},
'Pluto': {'a': 39.482, 'e': 0.249, 'color': 'plum'},
'Eris': {'a': 68.051, 'e': 0.436, 'color': 'pink'}
}
# Moon orbits data (a in AU, e) - converted from km using 1 AU = 149597870.7 km
moon_data = {
'Earth': {
'moons': {
'Moon': {'a': 0.002569, 'e': 0.0549, 'color': 'gray'}
},
'color': 'blue'
},
'Jupiter': {
'moons': {
'Io': {'a': 0.002818, 'e': 0.0041, 'color': 'red'},
'Europa': {'a': 0.004485, 'e': 0.009, 'color': 'lightgray'},
'Ganymede': {'a': 0.007157, 'e': 0.0013, 'color': 'gray'},
'Callisto': {'a': 0.012585, 'e': 0.0074, 'color': 'darkgray'}
},
'color': 'brown'
},
'Saturn': {
'moons': {
'Titan': {'a': 0.008168, 'e': 0.0288, 'color': 'orange'},
'Rhea': {'a': 0.003523, 'e': 0.001, 'color': 'gray'}
},
'color': 'gold'
},
'Uranus': {
'moons': {
'Titania': {'a': 0.002914, 'e': 0.0011, 'color': 'lightblue'}
},
'color': 'lightblue'
},
'Neptune': {
'moons': {
'Triton': {'a': 0.002371, 'e': 0.000016, 'color': 'blue'}
},
'color': 'darkblue'
}
}
# Plot heliocentric orbits
fig1, ax1 = plt.subplots(figsize=(12, 12))
ax1.scatter(0, 0, color='yellow', label='Sun', s=100)
for body, info in heli_data.items():
plot_ellipse(info['a'], info['e'], info['color'], body, ax1)
ax1.set_title('Heliocentric Orbital Paths of Planets and Dwarf Planets (in AU)')
ax1.set_xlabel('AU')
ax1.set_ylabel('AU')
ax1.set_xlim(-80, 80)
ax1.set_ylim(-80, 80)
ax1.set_aspect('equal')
ax1.legend(loc='upper right')
ax1.grid(True)
# Create separate figures for moon systems
for planet, data in moon_data.items():
fig, ax = plt.subplots(figsize=(8, 8))
ax.scatter(0, 0, color=data['color'], label=planet, s=50)
for body, info in data['moons'].items():
plot_ellipse(info['a'], info['e'], info['color'], body, ax)
ax.set_title(f'Orbital Paths around {planet} (in AU)')
ax.set_xlabel('AU')
ax.set_ylabel('AU')
ax.set_aspect('equal')
ax.legend()
ax.grid(True)
plt.show()
Which produced the following plot along with a dozen others:

In case if you wonder - it is far past just an LLM, it is reasoning:

and.. it just kept going..

<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Rotating 10-Sided Tumbler with Falling Ball</title>
</head>
<body>
<canvas id="tumbler" width="400" height="400" style="border: 1px solid black;"></canvas>
<script>
var canvas = document.getElementById('tumbler');
var ctx = canvas.getContext('2d');
var centerX = canvas.width / 2;
var centerY = canvas.height / 2;
var tumblerRadius = 150;
var rockRadius = 20;
var rockX = centerX;
var rockY = centerY + (tumblerRadius - rockRadius); // Start at bottom
var vx = 0;
var vy = 0;
var rockAngle = 0;
var angularVelocity = 0;
var gravity = 0.2;
var damping = 0.8;
var friction = 0.5;
var tumblerAngle = 0;
var tumblerAngularSpeed = 0.05;
function getVertices() {
var vertices = [];
for (var i = 0; i < 10; i++) {
var theta = tumblerAngle + 2 * Math.PI * i / 10;
var x = centerX + tumblerRadius * Math.cos(theta);
var y = centerY + tumblerRadius * Math.sin(theta);
vertices.push({x: x, y: y});
}
return vertices;
}
function draw() {
ctx.fillStyle = 'beige';
ctx.fillRect(0, 0, canvas.width, canvas.height);
// Draw rotating tumbler
var verts = getVertices();
ctx.beginPath();
ctx.moveTo(verts[0].x, verts[0].y);
for (var i = 1; i < 10; i++) {
ctx.lineTo(verts[i].x, verts[i].y);
}
ctx.closePath();
ctx.fillStyle = 'white';
ctx.fill();
ctx.strokeStyle = 'red';
ctx.lineWidth = 5;
ctx.stroke();
// Update physics
vy += gravity;
rockX += vx;
rockY += vy;
rockAngle += angularVelocity;
// Collision detection with polygon sides
verts = getVertices();
for (var i = 0; i < 10; i++) {
var p1 = verts[i];
var p2 = verts[(i + 1) % 10];
var dx = p2.x - p1.x;
var dy = p2.y - p1.y;
var len = Math.sqrt(dx * dx + dy * dy);
if (len === 0) continue;
var ux = dx / len;
var uy = dy / len;
var nx_out = -dy / len; // Outward normal
var ny_out = dx / len;
var dxr = rockX - p1.x;
var dyr = rockY - p1.y;
var proj = dxr * ux + dyr * uy;
proj = Math.max(0, Math.min(len, proj));
var closestX = p1.x + proj * ux;
var closestY = p1.y + proj * uy;
var distX = rockX - closestX;
var distY = rockY - closestY;
var dist = Math.sqrt(distX * distX + distY * distY);
var dot = (distX / dist) * nx_out + (distY / dist) * ny_out;
if (dist < rockRadius && dot > 0) {
var overlap = rockRadius - dist;
var unit_distX = distX / dist;
var unit_distY = distY / dist;
rockX -= overlap * unit_distX;
rockY -= overlap * unit_distY;
var nx = unit_distX;
var ny = unit_distY;
// v_wall
var r_dist = Math.sqrt((closestX - centerX) ** 2 + (closestY - centerY) ** 2);
var unit_rad_x = (closestX - centerX) / r_dist;
var unit_rad_y = (closestY - centerY) / r_dist;
var unit_tan_x = -unit_rad_y;
var unit_tan_y = unit_rad_x;
var v_wall_x = tumblerAngularSpeed * r_dist * unit_tan_x;
var v_wall_y = tumblerAngularSpeed * r_dist * unit_tan_y;
// Normal reflection
var vrad = (vx - v_wall_x) * nx + (vy - v_wall_y) * ny;
vx -= (1 + damping) * vrad * nx;
vy -= (1 + damping) * vrad * ny;
// Damping on whole velocity (approximate)
vx *= damping;
vy *= damping;
// Tangential
var tx = -ny;
var ty = nx;
var v_tan = vx * tx + vy * ty;
var v_wall_tan = v_wall_x * tx + v_wall_y * ty;
var relative_vtan = v_tan - v_wall_tan;
angularVelocity = (v_wall_tan - v_tan) / rockRadius;
var frictionForce = friction;
if (Math.abs(relative_vtan) > frictionForce) {
var sign = relative_vtan > 0 ? 1 : -1;
vx -= sign * frictionForce * tx;
vy -= sign * frictionForce * ty;
angularVelocity += sign * frictionForce / rockRadius;
} else {
vx -= relative_vtan * tx;
vy -= relative_vtan * ty;
angularVelocity = (v_wall_tan - v_tan) / rockRadius;
}
}
}
// Draw ball
ctx.save();
ctx.translate(rockX, rockY);
ctx.rotate(rockAngle);
ctx.beginPath();
ctx.arc(0, 0, rockRadius, 0, 2 * Math.PI);
ctx.fillStyle = 'orange';
ctx.fill();
ctx.beginPath();
ctx.moveTo(0, 0);
ctx.lineTo(rockRadius, 0);
ctx.strokeStyle = 'black';
ctx.lineWidth = 2;
ctx.stroke();
ctx.restore();
// Update tumbler angle
tumblerAngle += tumblerAngularSpeed;
requestAnimationFrame(draw);
}
draw();
</script>
</body>
</html>
Producing the following in a browser:

- In summary it requires a very rapid adoptation of how we work, produce, and automate.
- Wikipedia was the decommodification of information. What that implied simply was - you had access to the information once only held by elite institutions and engineers.
- AI Agents will bring the decommodification of knowledge. This is entirely different - because not only it will know about lanugages, programming, and obscure facts, but you will simply be able to activate this knowledge as was shown. You won't need an engineer or an architect, just enough reasoning to know 'its not quite what I want.'
- The AI is still not a mind-reader. It will require good prompting to get good results.
- Advanced quatum mechanical and scientific papers will be tutored, routing entire teaching industries.