Mohit
banner
mohitmayank.com
Mohit
@mohitmayank.com
4 followers 1 following 150 posts
The AI Guy | Helping Startups 10x AI Game 🤖 | Author “Lazy Data Science Guide” | Creator of Jaal
Posts Media Videos Starter Packs
🔥 Kimo K2 Thinking > GPT-5 or Sonnet 4.5
How: Powered by GPT-4o, it extracts story details and suggests edits as you interact with visuals. Built with TypeScript, React, and Vite, it runs locally using your OpenAI API key securely.

Code:
github.com/m-damien/Vi...
Paper:
arxiv.org/pdf/2410.07486
GitHub - m-damien/VisualStoryWriting: 🧙‍♂️ Writing by manipulating visual representations of stories
🧙‍♂️ Writing by manipulating visual representations of stories - m-damien/VisualStoryWriting
github.com
🧙‍♂️ Imagine rewriting your story just by moving characters and events visually!

Why: Visual Story-Writing aims to make story crafting intuitive and dynamic by bridging visual interaction with AI-driven text generation.
- Optimization steps: OpenMP threading, CUDA matrix multiplication, kernel fusion, memory tricks, and loop unrolling for maximum speed.
- Benchmark: Achieves 63 tok/s on a RTX 4090, surpassing projects like llama.cpp.

3/4
Andrew covers:

- Building LLM inference architecture: attention, feedforward, custom transformer blocks, and KV cache use.
- Hardware bottlenecks, focusing on memory bandwidth and the benefits of quantization.

2/4
Building a Lightning-Fast LLM from Scratch? Here’s How It’s Really Done!

If you want to deeply understand the difference between “just running a model” and “squeezing every ounce of performance with your bare hands,” Andrew Chan’s deep-dive blog is your blueprint.

1/4
Maya Research, founded by two 23-year-old Indians, open-sourced Maya1 TTS model that is #20 on the global leaderboard! It was created by finetuned LLaMa model + SNAC audio codec. In case you want to know what SNAC is, here is an article mohitmayank.com/a_lazy_data...
WindSurf's Codemaps release shows that VibeCoding SaaS recognizes users often struggle due to a lack of understanding of their own codebase. "It's not us, it's you" 🫵
Did you get your free ChatGPT subscriptions? 🇮🇳

here is how to get it - Go to chatgpt.com/ > Click on "Try Go" button > Click on "Upgrade to Go" > Add Card (you will not be charged) > Click "Purchase". Done!
Trainable vs. Frozen Parameters 👀
I’m surprised how many people STILL don’t know that with LoRA, you don’t have to fine-tune the entire embedding matrix - just the new tokens!
#1 Did you know you can learn the basics of Agentic AI for free on DeepLearning.AI!?
Should you train your own model or not?
ditch the netflix binge this weekend and dive into this instead
I'm surprised how many people don't use Python profilers when it is so easy
Just use FP16 and be happy.
When you cancel your @netflix subscription, they show a banner on mobile app, where you can’t click on “No, thanks” 🤦
Each method weighs components differently—characters, tokens, sequences, or meanings. Choosing the best algorithm requires testing them against your dataset and requirements.

Read more here:
mohitmayank.com/a_lazy_data...
🔍 Exploring the diverse world of text similarity algorithms reveals there’s no one-size-fits-all solution; it all depends on your specific use case.