A Guide to Vibe-Coding Large Projects . I tried to come up with a guide for vibe-coding large greenfield projects. Here's my suggestion—wondering what else we should add to this list:. 0- You do not need to pay for an IDE.Tools like Cursor, Windsurf, or Lovable are great for.

Maybe AI models trained on free and public data should also be free and public.

Agentic AI Summit

@grok ♟️Why has no one created a chess engine that can play accurately at a specified Elo? Engines are either super strong or they randomly blunder to pretend they’re rated 1500.

♟️Why has no one created a chess engine that can play accurately at a specified Elo score? Engines are either super strong or they randomly blunder to pretend they’re rated 1500.

Or #RAG = Ridiculously Atrocious Garbage.

Unpopular opinion: I think RAG systems are overhyped. They often don’t quite work the way people hope they will. #RAG = Really Average Gibberish.

[Question] At some point, the cost of some of these AI tools will go up to $5k per month for individuals (right now it is ~$200 for each top tier AI tool) How are regular people supposed to keep up with the wealthy at that point?.

Me finding Firebase after so many years of working in tech feels like I have rediscovered fire! 🔥 #firebase.

🤔 Is there a way to add a PDF/TextFile to the knowledge base of any of the big AI UIs? . I know how to build a separate RAG system on the Gemini API but I like the convenience and the single payment model of the UIs.

📝 Pushed the vibe-coding guide to Github and added a few scripts to it -->

Business idea: an LLM that is trained to help people navigate Google Cloud's complex AIM roles. Gemini Pro constantly suggests using Admin roles for accounts that should not have admin roles.

Has anyone ever learned a language using Duolingo?.

I believe the way we develop large systems is changing dramatically. The technology is already available for solo developers to build large systems independently. For example, using Google Cloud and Gemini, I built and deployed a domain-specific Gemini question-answering system.

Synthetic data trick: by just zeroing your classifier-free guidance (CFG) you can just create unlimited synthetic training data for your image generation

"MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention"

Unpopular opinion in agile development: project backlog items should live on the same board as the rest of sprint items.

Gemini Pro and O3-Mini-High were provided the same prompt. Below are the results

I feel like there's zero stickiness or moat with foundation models. I was a paying Claude customer for a while, but I switched to ChatGPT when it became better. Now, I'm paying for Gemini Pro because it works best for my use case. It's great for consumers!.