rishabh retweeted
👇 Day: 04 Of Learning with @rishabh10x
Completed upto positional embeddings, where it provides the information about the order of words in the sentence when tokens are sent in parallel.
#ArtificialIntelligence #TechTrends #MachineLearning #DigitalTransformation
rishabh retweeted
Happy Sunday so far :)
- Started learning Kubernetes — completed the first lecture and implementation
- Fixed a major project bug
- Went deep into fine-tuning
One task left before sleep DSA which i'm doing.
@rishabh10x @kirat_tw
rishabh retweeted
LLM Obserability by @rishabh10x . Got great project ideas also from this.
rishabh retweeted
Fine-tuning, and what actually happens inside the model
Prompting and RAG don’t change the model. Fine-tuning does. RAG adds knowledge, fine-tuning changes behavior and style.
LoRA freezes the base model and trains tiny adapters....more
rishabh retweeted
Did deep dive into Recursive Language Models RLMs today. Fascinating how they turn context into code variables, helping reduce context rot and enabling work with much larger contexts.
Started working on a project "Code Reviewer" where I’ll be using RLMs.
Thanks to @rishabh10x 🙌
rishabh retweeted
Going live today at 10 PM IST.
Talking poker, the inner studio and our upcoming raise on @backabledotbiz
This Week on MCG Powered by @fomo, @MetaDAOProject, times in ET
Monday 10/19
12:30 - $POKER @TheInnerTable w/@kirat_tw (@MetaDAOproject eco)
1:00 - $STATICS @StaticsProtocol w/@hooftly
Tuesday 10/20
12:30 - $RAM @RamsesExchange w/@northpool9
1:00 - $CLAWHUNTER @mystudiosfun w/@lycavallo (@Clawpump eco)
Wednesday 10/21
12:30 - $FABLES @FablesFi w/@0xcs361
1:00 - $RAXOL @axol_io w/@DROOdotFOO @0xblooberries
Thursday 10/22
12:30 - $BORRO @borromarkets w/@BorroDev
1:00 - Open slot, DM MCG to book
Friday 10/23
12:30/1pm Open Slots - DM MCG to book
MCG Research terminal: mcgsignal.com
Subscribe/Notifications on: youtube.com/@MCG_Live
Learn to LP with @met_lparmy on @MeteoraAG
Convert to $USDv with @solomon_labs, use link in bio.
rishabh retweeted
Completed Week 16 of @100xDevs AI/ML course: RL Environments by @rishabh10x bhaiya
Today's learning:
1/What is RL ?
Reinforcement Learning= Try actions → Get rewards → Learn what works
For LLMs:Generate responses → Score them → Update the model to improve future response
rishabh retweeted
❄️ WinterArc - Day 7
Built and trained a small modern LLM from scratch in PyTorch!
RMSNorm
RoPE
Grouped-Query Attention (GQA)
SwiGLU
Trained a 2.7M parameter model on Tiny Shakespeare and got the validation loss from 4.22 → 1.50 in ~11 minutes.
rishabh retweeted
Completed Week 14 of @100xDevs
AI/ML course: RLHF & DPO
by @rishabh10x bhiaya.
Learned preference tuning:
RLHF learns from feedback
Negative example show boundaries SFT only teaches center
PPO optimization
DPO skips reward model, it learns from chosen AND rejected
#ML #AI
rishabh retweeted
Completed Week 15 of @100xDevs AI/ML course: RLVR by @rishabh10x Bhaiya
how DeepSeek-R1 changed everything:
1/Test-time scaling
Don't just spend compute during training. Let models think longer during inference
Pre-o1: Prompt → Answer
o1: Prompt → Think 30sec → Answer
Career update 🚀 : Joined @Infosys as Senior Associate Consultant - Gen AI.
The journey wasn’t easy but a huge thanks to @Hiteshdotcom, @piyushgarg_dev and @rishabh10x for generously sharing your knowledge. It truly wouldn’t have been possible without your support.
rishabh retweeted
❄️ WinterArc: Day 6
Transformers look pretty different today compared to 2017.
If you look under the hood of Llama, Mistral, or DeepSeek, almost every core component has been swapped out for better stability and efficiency during scaling:
rishabh retweeted
Just now Completed Week 13 of @100xDevs AI/ML course Finetuning Part 2 by @rishabh10x bhaiya.
Learned Supervised Fine-Tuning :
Converts autocompleter to assistant
Alpaca format with EOS boundaries
Packing multiple examples efficiently
Attention and loss masking
#ML #AI #GENAI
rishabh retweeted
❄️ WinterArc — Day 5
Today I learned what’s actually happening under the hood of DL.
Tensors = how data is represented
Matrix multiplication = how data gets transformed
Dot products = how models compare things
GPUs = doing tons of these calculations in parallel
rishabh retweeted
❄️WinterArc : Day 4
Today I went a little deeper into how Transformers actually work.
Got a better understanding of what happens inside an LLM:
→ Positional Encoding helps the model understand word order
→SelfAttention uses Q, K & V to figure out what each token should focus
rishabh retweeted
Completed Week 12 class of @100xDevs AI/ML course Fine-tuning Part 1 by @rishabh10x bhaiya.
Finetuning changes model behavior
Learned about:
CPT, SFT, LoRA, QLoRA,
Key hyperparameters: rank, alpha, target modules, dropout and catastrophic forgetting
#MachineLearning #AI
rishabh retweeted
❄️ WinterArc : Day 3
Ever wondered how AI models like ChatGPT actually read text?
Hint:
They don’t see words the way we do.
Everything gets converted into numbers first! Here is a simple breakdown of how Transformers turn text into something they can understand:
rishabh retweeted
❄️WinterArc : Day 2
🧠How do Neural Networks actually learn?
Diving deeper into the fundamentals of Deep Learning.
Today I learned:
• Neurons = inputs + weights + bias + activation
• Non-linearity lets networks learn complex patterns



































