- Neural networks visual course ( If you know nothing about neural networks start here) ( I strongly recommend you subscribe to anything Grant Sanderson does e.g. 3blue1brown and his personal channel where he gives other talks )
- Large Language Models explained briefly ( Smooth introduction )
- Transformers (how LLMs work) explained visually ( Deeper visual understanding )
- What does it mean for computers to understand language? ( a conceptual understanding of language models from the ground up )
- Let's build GPT: from scratch, in code, spelled out. ( Once you are bit more aware, watch and read everything Andrej Karpathy says or writes)
- History of language models
-Fast AI ( A bit outdated but provides a very solid understanding of machine learning especially this course.)
- mlabonne llm course (I think it's the same course that Huggingface provides since he's the top scientist there) (Use this chat to ask questions and test your knowledge) (Follow his work)
- LLM University (Not tried this but people speak highly of it. Take a look at both and choose one.)
- Hugging Face (Main portal for open source models) (Also has many courses on Hugging Face Learn)
- Hugging Face smolagents[python] (simple/powerful and 100% OSS)
- mastra-ai [typescript] (new kid on the block)
- langchain [python,javascript] (very popular and powerful but overly complex. OSS but potential lock-in)
There are more but you'll discover them eventually...
- Check the openrouter rankings e.g. this one for programming (You'll see Claude Sonnet 3.7 is by far the superior choice here...for now)
- Hugging face
(You'll need a beefy machine with a good graphics card but on my machine
qwen2.5:7b(4Gb) performs pretty decently for programming tasks too)
(Check the llm stats) (Both openrouter and llm stats usually coincide but always check both)
(I don't bother with OpenAI since they are far more expensive)
- deepseek R1 (Use this for planning!! See next article)
- LLM codegen workflow (Read how to make maximum use of existing AI tools) (tl:dr; Brainstorm spec, then plan a plan, then execute using LLM codegen. Discrete loops. Then magic!)
-Cline (Addy Osmani (google chrome top dev) has very powerful reasoning for using Cline. READ THIS!!) (Use it everyday. Don't look back!!!)
One of the many advantages over Cursor is you can use any API or local model.
For programming, first buy 20 euros worth of anthropic api to start with.
Cline, will show the cost of different tasks. Then you can start exploring using different cheaper models for certain tasks using other model apis (e.g. gemini flash).
Cline also supports openrouter which can automatically route to cheaper models based on your prompt.
Awesome MCP Server list: https://github.com/punkpeye/awesome-mcp-servers?tab=readme-ov-file#browser-automation
-Cursor ( Similar to the article above this article is more related to Cursor but explains THE FUNDAMENTAL GAME CHANGING IDEA OF AI USE FOR PROGRAMMERS IN THE FUTURE. If you don't read anything else, read this!!!. You should then try to apply this thinking to using other AI codegen tools (e.g. cline) See clinerules & memory-bank )
- Windsurf AI (Not tried it because I think Cline is better)
(Check out others in the OpenRouter app leaderboard)
If you aren't familiar with the terminal shells (bash/zsh/fish) then first get familiar with them.
-
Warp (But on free plan you have limits)
-
Aider (Has overlap with Cline but also works for commands on the terminal and allows opensource models. Not tried it yet for codegen though because Cline is better)
-
Claude AI from Anthropic (Usually my go to choice, especially for programming questions. Though with cline and co-pilot, I need this less and less)
-
Deepseek (Anytime you want more careful planning)
-
Mistral Le Chat (Good alternative when you hit Claude's limits)
-
Open Router's char (Portal to other chatbots)
-
Groq (Extremely fast inference)
Mostly for web apps...
- Lovable.dev (Build full fledged (static) single page web apps from a single prompt. The better the prompt the better the app. So plan your prompt with a thinking model and then improve it with a conversational model)
- Vercel v0) (Like lovable but also very good. Excellant for prototyping...Landing pages etc)
- Replit agent (Can build and deploy full apps including setting up infrastructure e.g. dbs, caches, networking etc...lock-in and currently limited to python apps)
(There are others and more will be built. The trend is going in this direction. Explore them to see what's already possible.)
- Alphasignal (I basically use this, hacker news and friends)
- Hacker news
- Deep Learning.AI
- Deep Learning.AI
- OpenRouter AI App Leaderboard (Check this often to see the cool new AI tools. There's also a lot of crap and places that will waste your time so focus!)
- OpenRouter AI Model Rankings