Shreyas Kumar

@Im_IrushiK Skill

I'm a vibe coder, scare me with one word.

@Jason i'm in

We started an AI founder twitter group... reply with "I'm in" if you're a founder and want to be added

I just published my first Novel✨ The Kettle and the Lantern: a quiet tea shop, a small town, and found family. Free on Kindle for 2 days: amazon.com/dp/B0GY5V4DHY #cozyfantasy #amwriting #indieauthor

@ParagArora I am currently in my 3rd year btech want to start startup on interview by ai. Built an small mvp but doesn't know how to start it. Looking for your guidance. I have followed can you message in private

IITD 2008. Sold my first startup to Paytm right out of college. Second one, backed by YC. Third one, building now. If you're in tech and seek help building, ask me anything. Here or DM.

I built a Notion template that tells you when to buy a stock.📈 Auto-updating portfolio 🟩🟥 Visual buy-zone bars 🧪 Graham's value checks 🧠 Bias checklist before every tradeStop guessing. Invest with math. #NotionTemplate" target="_blank" rel="nofollow noopener">payhip.com/b/9fwzL#Notion… #ValueInvesting

I built a Notion template that tells you when to buy a stock.📈 Auto-updating portfolio 🟩🟥 Visual buy-zone bars 🧪 Graham's value checks 🧠 Bias checklist before every tradeStop guessing. Invest with math.🔗 #NotionTemplate" target="_blank" rel="nofollow noopener">payhip.com/b/9fwzL#Notion… #ValueInvesting

AI coding agents are changing software development. Devin and Claude computer use show what autonomous coding looks like. The shift is from AI as autocomplete to AI as a collaborator that plans, writes, runs and debugs code end to end. This is just getting started.

Diffusion models took generative AI beyond text. Stable Diffusion, DALL E and Sora all use score based diffusion. The core idea is learning to denoise progressively. Understanding diffusion math opens up a huge portion of modern generative AI research.

State Space Models like Mamba are a real alternative to Transformers. Linear complexity vs quadratic for long sequences. On genomics, time series and long documents SSMs can outperform attention in both speed and accuracy. Worth tracking closely if you work on sequence modeling.

Fine tuning vs prompting is something every ML engineer faces. Few shot prompting with a strong model handles most cases. But for consistent output format, domain knowledge or low latency, fine tuning wins. Knowing when to fine tune is as important as knowing how.

Test Time Compute is a new scaling axis. Models like o3 spend more compute at inference to reason harder. Chain of thought and self reflection at test time let smaller models beat larger ones. This changes how we define model capability entirely.

Vision Language Models like GPT-4o and Gemini 1.5 Pro are making multimodal AI mainstream. Images, audio and text together in one model unlocks capabilities single modality systems cannot match. Multimodal reasoning is now the baseline for frontier AI.

Multi agent systems are the next big shift in AI. Specialized agents collaborating beats one LLM doing everything. Frameworks like LangGraph and AutoGen make this practical. The orchestration layer is where real AI engineering happens now.

Retrieval Augmented Generation tackles hallucination head on. By grounding LLM responses in retrieved documents at inference time, RAG makes models far more reliable. Vector databases like Pinecone and Qdrant are now central to every serious production AI system.

RLHF changed how we align LLMs with human preferences. Now RLAIF takes it further by using AI feedback instead of human labelers. This scales alignment research massively. Constitutional AI combined with RLAIF is one of the most exciting frontiers in the field today.

Mixture of Experts is reshaping LLM scaling. MoE routes each token to only a few expert networks instead of all parameters. Models like Mixtral get massive capacity with low inference cost. Sparse activation is a must know concept for AI engineers in 2026.

@fdotinc Building win-prep.vercel.app WinPrep is an AI-backed interview preparation platform that combines realistic interview simulations, AI-generated questions and feedback, live video rooms, resume parsing, and printable PDF reports - all in a single

We want to invest $100k-$200k in someone international. Show us a demo of what you're building, if it's cool: You'll join Canopy, a 5 week program: > can join in person at our SF campus or online > on demand access to our team + $50k in credits. > work w/ 500 founders making the same push. > extremely direct support on what you're shipping. > constant talks and lectures from $1b founders. not in SF? good. apply.

Big news! I'll be interning at @Microsoft! 🥳🎉 Honored by this opportunity, can't wait to get started! Thanks @Microsoft & @ChennaiVit🙏 #Microsoft #MicrosoftIntern #Intern2026 #Internship #TechInternship

@AndrewBolis Guide

You can make $17,292 per month with a Faceless YouTube Channel, if you have: 1. Wifi 2. Laptop 3. Two hours per day I have prepared an in-depth guide for this. It's absolutely FREE: Like + Comment 'Guide' and I will DM you my proven guide for free. (Follow me to get DM) Free for 24 hours

@piyascode9 DS

DeepSeek is a crypto money printer 🚨 I wrote a Bot, that's made me $35K from $100 last night. This isn’t clickbait—just an AI Trading Bot powered by DeepSeek. FREE for 24 hrs then deleted! Simply, 1. RT 2. Like 3. Reply "DS" and Follow me and I'll send you a DM.