Superintelligence explains one AI concept at a time — clearly, and with no hype. Agents, LLMs, RAG, and everything you keep hearing about, finally made simple.
New episode is live 🧠 How Do We Know If AI Is Actually Smart? — how AI benchmarks and evals really work, and the catch that can make a leaderboard lie.
New episode is live 🧠 How AI Gets Smaller Without Getting Dumber (Quantization) — how a model that needs a data-center GPU gets squeezed down to run on your laptop, even your phone.
New episode is live 🧠 How AI learns to be helpful — RLHF explained with no hype: people rank answers, a reward model scores them, and the model gets nudged toward what we prefer (plus the sycophancy catch).
New episode is live 🧠 Agent, Workflow, or Assistant? The difference that actually matters — and the one question that tells them apart: who decides the next step?
New episode is live 🧠 How AI Gets Tricked — prompt injection, where one hidden line of text can hijack an AI assistant. What it is, why it works, and how to stay safe.
New episode is live 🧠 MCP — the "USB-C for AI", explained. Why a raw model is sealed off from your files and tools, how the old "custom connector for every app × every tool" tangle collapses into one open standard, and the client/server loop that lets any AI use any tool.
New episode is live 🧠 Why "thinking" AI models are different — reasoning & test-time compute.
We test it on one dumb question ("how many R's in strawberry?") and watch a thinking model actually get it right, then unpack test-time compute, System 1 vs System 2, and when it's worth the wait.
Superintelligence
New episode is live 🧠 How Do We Know If AI Is Actually Smart? — how AI benchmarks and evals really work, and the catch that can make a leaderboard lie.
▶ Full video: https://www.youtube.com/watch?v=RuHxT...
⚡ 60-sec version: https://www.youtube.com/watch?v=glWQd...
2 months ago | [YT] | 0
View 0 replies
Superintelligence
New episode is live 🧠 How AI Gets Smaller Without Getting Dumber (Quantization) — how a model that needs a data-center GPU gets squeezed down to run on your laptop, even your phone.
▶ Full video: https://www.youtube.com/watch?v=lMSw3...
⚡ 60-sec version: www.youtube.com/shorts/2DHoM5...
2 months ago | [YT] | 0
View 0 replies
Superintelligence
New episode is live 🧠 How AI Uses a Team of Experts (Mixture of Experts)
▶ Full video: https://www.youtube.com/watch?v=MWvRx...
⚡ 60-sec version: https://www.youtube.com/watch?v=Ac8S3...
2 months ago | [YT] | 0
View 0 replies
Superintelligence
New episode is live 🧠 How AI learns to be helpful — RLHF explained with no hype: people rank answers, a reward model scores them, and the model gets nudged toward what we prefer (plus the sycophancy catch).
▶ Full video: https://www.youtube.com/watch?v=86SWr...
⚡ 60-sec version: https://www.youtube.com/watch?v=wWipt...
2 months ago | [YT] | 0
View 0 replies
Superintelligence
New episode is live 🧠 Agent, Workflow, or Assistant? The difference that actually matters — and the one question that tells them apart: who decides the next step?
▶ Full video: https://www.youtube.com/watch?v=Yb4yx...
⚡ 60-sec version: https://www.youtube.com/watch?v=np0cT...
3 months ago | [YT] | 0
View 0 replies
Superintelligence
New episode is live 🧠 How AI Gets Tricked — prompt injection, where one hidden line of text can hijack an AI assistant. What it is, why it works, and how to stay safe.
▶ Full video: https://www.youtube.com/watch?v=g5bQ_...
⚡ 60-sec version: https://www.youtube.com/watch?v=8HMtS...
3 months ago | [YT] | 0
View 0 replies
Superintelligence
New episode is live 🧠 MCP — the "USB-C for AI", explained. Why a raw model is sealed off from your files and tools, how the old "custom connector for every app × every tool" tangle collapses into one open standard, and the client/server loop that lets any AI use any tool.
▶ Full video: https://www.youtube.com/watch?v=9LJRZ...
⚡ 60-sec version: https://www.youtube.com/watch?v=i0z9n...
3 months ago | [YT] | 0
View 0 replies
Superintelligence
New episode is live 🧠 Why "thinking" AI models are different — reasoning & test-time compute.
We test it on one dumb question ("how many R's in strawberry?") and watch a thinking model actually get it right, then unpack test-time compute, System 1 vs System 2, and when it's worth the wait.
▶ Full video: https://www.youtube.com/watch?v=DTuhy...
⚡ 60-sec version: https://www.youtube.com/watch?v=iKDcG...
3 months ago | [YT] | 0
View 0 replies
Superintelligence
New episode is live 🧠 How AI Makes Images (Diffusion)
It doesn't paint — it starts with a screen of pure static and removes it, step by step, until your picture is all that's left.
▶ Full video: https://www.youtube.com/watch?v=F9RAp...
⚡ 60-sec version: https://www.youtube.com/watch?v=BCech...
3 months ago | [YT] | 0
View 0 replies
Superintelligence
New episode is live 🧠 Open vs Closed AI — what a model's "weights" actually are, and why that one file splits the whole AI world in two.
▶ Full video: https://www.youtube.com/watch?v=3rZxv...
⚡ 60-sec version: https://www.youtube.com/watch?v=QCHcE...
3 months ago | [YT] | 0
View 0 replies
Load more