Uploaded November 2026 | Updated November 2026, 2 hours ago
oxylabs.io/siraj
$5 in free credits, no credit card, no time limit. Use code SIRAJ for 20% off additional Oxylabs credits.
This video is a paid promotion for Oxylabs.
I ran the same agent twice. One version answers from training data alone. The other searches the live internet through the Oxylabs Web API. Same model, same prompts, four scored rounds and a two hour mission where it has to find something I could actually buy.
Every answer gets checked against the real source on camera. Where there is a measurable price difference I work out what relying on the stale answer could have cost, and I am careful to call that hypothetical exposure rather than money lost.
The blind agent is confident and wrong. It quotes prices that moved, libraries that changed under it, and listings that are already gone. The cited agent opens the actual page, reads it, and links where every claim came from. The scrape endpoint returns clean markdown instead of raw HTML, which matters more than it sounds: raw pages burn the context window before the model gets to think.
Chapters
0:00 Same question, two agents
0:28 Why a training cutoff costs you money
0:50 The rules: four rounds and one mission
2:08 What the Oxylabs Web API actually does
3:55 Round 1, prices that moved
5:37 Changing one parameter: geo targeting
6:16 Round 2, what is true today
7:58 Round 3, real estate listings
9:43 Round 4, code against live docs
11:21 Scoreboard, and why I verify everything
12:59 Inside the dashboard and usage limits
14:36 The two hour mission
15:56 Results, and what I will not call profit
17:30 Final scoreboard
18:21 Three takeaways
The MCP server used in the coding round: github.com/oxylabs/web-api-mcp
Every number on screen was measured during the run, including the ones that went against me.
Subscribe for more real-world AI builds: youtube.com/@SirajRaval
Business inquiries: hello@sirajraval.com
X: https://x.com/sirajraval
Instagram: instagram.com/sirajraval
LinkedIn: linkedin.com/in/sirajraval
oxylabs.io/siraj
$5 in free credits, no credit card, no time limit. Use code SIRAJ for 20% off additional Oxylabs credits.
This video is a paid promotion for Oxylabs.
I ran the same agent twice. One version answers from training data alone. The other searches the live internet through the Oxylabs Web API. Same model, same prompts, four scored rounds and a two hour mission where it has to find something I could actually buy.
Every answer gets checked against the real source on camera. Where there is a measurable price difference I work out what relying on the stale answer could have cost, and I am careful to call that hypothetical exposure rather than money lost.
The blind agent is confident and wrong. It quotes prices that moved, libraries that changed under it, and listings that are already gone. The cited agent opens the actual page, reads it, and links where every claim came from. The scrape endpoint returns clean markdown instead of raw HTML, which matters more than it sounds: raw pages burn the context window before the model gets to think.
Chapters
0:00 Same question, two agents
0:28 Why a training cutoff costs you money
0:50 The rules: four rounds and one mission
2:08 What the Oxylabs Web API actually does
3:55 Round 1, prices that moved
5:37 Changing one parameter: geo targeting
6:16 Round 2, what is true today
7:58 Round 3, real estate listings
9:43 Round 4, code against live docs
11:21 Scoreboard, and why I verify everything
12:59 Inside the dashboard and usage limits
14:36 The two hour mission
15:56 Results, and what I will not call profit
17:30 Final scoreboard
18:21 Three takeaways
The MCP server used in the coding round: github.com/oxylabs/web-api-mcp
Every number on screen was measured during the run, including the ones that went against me.
Subscribe for more real-world AI builds: youtube.com/@SirajRaval
Business inquiries: hello@sirajraval.com
X: https://x.com/sirajraval
Instagram: instagram.com/sirajraval
LinkedIn: linkedin.com/in/sirajraval






![I Used AlphaFold 3 To Cure Cancer (Tutorial)
I asked Claude to find a cancer protein with no known cure and design a potential cure (that it codenamed DualStrike). Then I used Googles AlphaFold 3 to simulate its effectiveness. The results showed that DualStrike (a novel treatment for Lymphoma [Blood Cancer]) can theoretically increase the 5 year survival rate of from 80% up to 90-95%. Finally, I asked ChatGPTs o1-preview to write an end-to-end research paper on the process to enable any elementary biochemist to verify the in-silico results with a wet-lab. In this video, Ill explain each step I took so that you can do the same for your own creative ideas around Scientific discovery. The year is 2025 and AI is truly democratizing Science to a level where any human on Earth with enough motivation can make a meaningful contribution using AI tools effectively. I hope this video inspires Wizards around the world to use AI for Scientific discovery. Enjoy!
Code & Research Paper:
https://github.com/llSourcell/DualStrike_AI_For_Lymphoma
Sign up for HPC-AI:
https://hpc-ai.com/account/signup?invitation_code=invite_c3HJ41XXWr
Want more AI/ML education? Connect with me here:
X: https://x.com/sirajraval
Instagram: https://www.instagram.com/sirajraval
Linkedin: https://www.linkedin.com/in/sirajraval/
TikTok: https://www.tiktok.com/@sirajraval
Facebook: https://www.facebook.com/sirajology
Design your own AI trading bot (no code):
https://www.tradergpt.co
Join my AI Sports Betting community:
https://wagergpt.xyz/
⏱️ **Chapters:**
0:00 Introduction: Using AI to Design a Novel Cancer Treatment
1:16 Introducing Sponsor: HPC AIs Computing Cluster
1:59 The Big Deal: Introduction to AlphaFold
2:06 Evolution of AlphaFold: From Version 1 to 3
2:59 Finding a Missing Protein in the AlphaFold Database
4:17 Identifying the NPM-ALK Fusion Protein
4:43 Generating the FASTA Sequence with Claude
5:41 Understanding Protein Folding and AIs Role
6:35 Designing a Potential Treatment with AI
7:03 Developing a Dual-Site Inhibitor with Claude
7:54 Generating 3D Models with AlphaFold
8:38 Testing the Inhibitor with Molecular Docking Software (MolModa)
10:30 Generating the Research Paper and Code with o1-preview
13:27 Prompt Hack: Improving AI Outputs
13:41 Utilizing HPC AIs Computing Cluster for Experiments
14:31 Conclusion and Encouragement to Innovate with AI
15:14 Outro and Closing Remarks
Dont forget to like, share, and subscribe for more deep dives into AI advancements!
AlphaFold 1 explained:
https://www.youtube.com/watch?v=cw6_OP5An8s
AlphaFold 2 explained:
https://www.youtube.com/watch?v=GQH-zWUylPY
I Asked O1-Preview to reproduce O1-Preview:
https://youtu.be/sf7Ntg72qCI
I Built a Sports Betting Bot with ChatGPT:
https://youtu.be/IDthta5sUGQ
I Built a Trading Bot with ChatGPT:
https://www.youtube.com/watch?v=fhBw3j_O9LE
Watch ChatGPT Build an AI Startup:
https://www.youtube.com/watch?v=hL2hLFUwuqQ
Watch ChatGPT Build a Finance Startup:
https://www.youtube.com/watch?v=on623d5EfPw
Watch Me Build a Startup Playlist:
https://www.youtube.com/watch?v=oeraUtRgsbI&list=PL2-dafEMk2A4n3aP_rzgqA8lVGknkSN2p&index=1
🔔 Subscribe and hit the notification bell to join the AI revolution! I Used AlphaFold 3 To Cure Cancer (Tutorial)](https://i.ytimg.com/vi/kll90oM9CoQ/mqdefault.jpg)



