Jin Daily AI Trivia – PewDiePie Discovers AI Agents and the Limits of SLMs

Jin Daily AI Trivia – PewDiePie Discovers AI Agents and the Limits of SLMs

PewDiePie just dropped a video showing how he built his own local AI setup — and the tech community went wild.

He’s running massive local models like GPT-OSS 120B, Qwen 235B, and GLM-4.6, all fine-tuned and quantized to fit into over 200GB of total VRAM (yes, multiple RTX 4090s were harmed in the making of this experiment).

He even built his own WebUI, integrated RAG and deep local search agent workflows, and started experimenting with model fine-tuning. Apparently, he plans to release his own fine-tuned model next month too.

Why is PewDiePie one of my all-time favorite YouTubers? Because he casually explains the evolution of today’s AI scene just by messing around with his own discoveries.

“How do you get better answers from a dumb agent?” Let them fight each other and punish the bad ones (hello RLHF).

“How do we stop AI from lying to us?” Easy — use a dumber model.

“What’s the limit of Small Language Models (SLMs)?” They suck at deep reasoning — but pair them with tool calling and good RAG context, and they can automate tons of simple tasks for you.

Honestly, PewDiePie explained AI way better than half of the “AI gurus” on LinkedIn. Love this man. Never change, king. 👑