Rahul Agrawal’s Post

Do you know what in-context learning, RAG or a context window for LLMs is? Scroll below to find out! 🔄 These terms are just the tip of the iceberg when it comes to learning about LLMs, and new concepts are constantly appearing. But don’t worry, we got you covered! Find all the definitions and buzzwords you need to know in our Big Dictionary of MLOps and LLMOps. https://bit.ly/3WHzgbE

I believe one of the greatest advantages of LLMs can also become a significant drawback for your performance. Over time, I’ve noticed myself relying heavily on LLMs for code samples, syntax, and directly copying references. While this saves time, it has also affected my ability to remember basic library functions and code syntax. It feels like a step backward in terms of memory retention and individuality as a programmer. This hit me hard during situations where recalling concepts on the spot is critical. Without access to LLMs or documentation, I found myself struggling with even well-known concepts, regretting and thinking, "I could have done this so easily if I had memorized it. My humble advice: Use LLMs / Docs to explore ideas, clarify doubts, and find references, but don’t stop there. Take the time to fully understand the code, memorize key functions and arguments, and practice typing it out yourself—just like we did in the good old days. This habit will sharpen your skills, improve your confidence, and prepare you for situations where quick recall is essential. Sharing this because I don’t want someone else to lose things simply and regret because of our daily habits. It may be a joke to a brilliant person, but for ordinary people like me, it is a practical experience. Thanks!

LLM-PowerHouse: Unleash LLMs' potential through curated tutorials, best practices, and ready-to-use code for custom training and inferencing. Keep Learning, Keep Growing!!

How to program with LLMs.

Did people who made compilers from scratch use guides or rely on intuition, and how can templates help in learning? Compiler #BuildingFromScratch #CodingResources Have you ever heard someone say they built a compiler from scratch and wondered what that really means? 🤔 Let's break it down into two possibilities - following a guide with their own tools and low-level stuff, or relying solely on pure intuition. 📚 Utilizing Guides: If someone followed a guide but built their own tools and low-level components, they likely had a basis to start from but still put in the effort to customize... Read more: https://lnkd.in/gw-X3hDW

LLMs write median code. This is great in most cases. But it also is a good way to tell if what’s in your head is above median.

This is a great LLM test question " What are the last 4 digits of the sum of the first 70 prime" I've managed to do this with llama3.1 70B Q4. With Temp = 0.0 it's consistent I've tweaked the prompt a bit but I think it's still generic enough. System Prompt: "You are a helpful assistant and always try to answer user queries to the best of your ability. Always say out everything do the task step by step use Chain of Thought. Keep everything on order. Use helper text. Use numbered lists to keep track of the things. Don’t take shortcuts. You have no tools you have to do all the work (like calculations) and also double check your results" I've also managed to reproduce with ChatGPT. The free version doesn't state which model it uses it's not really transparent but the chat prompt stated "Analyzing..." so maybe it's used o1? I can't stop thinking why OpenAI has removed the system prompt, have the feeling they also play with the System prompt, and might run multiple generations but not display the data?

The way to learn a new stack is not to learn a new stack unless you have a depth of knowledge in at least one.

Here's the truth why coding with LLMs is so challenging. 😡 We know #LLMs are probabilistic algorithms. They predict the most probable "word" based on previous context. So, if you consider that most the code used to train them was 💩, it's no wonder we sometimes have to battle for hours to get LLM-generated code to work. Something to think about.

Day 5 Harvard's CS50x course Today is still all about problem set 0 I am learning to write pseudo code so I can better articulate my formal code But first, what is pseudo code?- - it's the process of converting instructions into code. Pseudo code is the human readable version of your code. "Why should I pseudo code again?" you'd say! When you pseudo code before creating your formal code it allows you to THINK through the logic of your problem in advance & it will help you in providing the logic of your problem to other people that wants to understand your coding decisions & how the code works. When I am done with the pseudo code of my problem set Zero I will update but here's an example of a pseudo code Say we want to write a code for an address book program to find a name in the book - say the search bar code- the pseudo code will look something like this ! -Go to the phonebook -Open to the middle of the phone book -look at page -if person is on page - Call person -Else if person is earlier in book - Open to middle left half of book - Go back to line 3 -Else if person is later in the book - Open to middle of right half of book - Go back to line 3 -Else -Quit.