Post

x.com/15650288283299…

More from @heyshrutimishra

Shruti

@heyshrutimishra

Aug 27

🚨 MIT just dropped a brutal report on AI adoption.

95% of AI projects fail because of overpriced models.

That’s why I tried a lean model to build a viral content app at 1/10th the cost.

Here’s the exact code + setup I used : 🧵👇

2/ While companies fail with "their own worse version" (MIT's actual words)...

Meanwhile, Kimi is doing the same job at a much lower cost:

→ GPT-4.1 (base): $2 input / $8 output
→ Claude Sonnet 4: $3 input / $15 output
→ Kimi K2: $0.15 input / $2.50 output

Kimi stays the cheapest. Same quality. ⚡️

3/ Step 1: Get your API key →

Here’s how simple it is to get started: platform.moonshot.ai

Read 10 tweets

Shruti

@heyshrutimishra

Aug 26

🚨 Google just dropped Nano Banana inside Gemini… and it’s WILD 🍌

SPOILER: it makes Photoshop look ancient.

People are already calling it the best AI photo editor in the world.

Here’s why everyone’s freaking out (with wild examples + how to use it): 👇

Google DeepMind built the #1 rated image editor in the world and it’s now inside the Gemini app... you guessed it right, it's nano-banana

Now You can:
• Keep your look while changing outfits, places, or even decades
• Blend photos of you + your pet into one scene
• Edit in steps (paint walls → add furniture → redesign a room)
• Transfer styles (butterfly wings → dress pattern)

Check post from @OfficialLoganK

2/ Until now, AI photo editing had one fatal flaw:

👉 It could change your photo…
👉 But it couldn’t keep your face consistent.

One edit and suddenly you looked like your distant cousin.

Gemini just fixed that.

Read 11 tweets

Shruti

@heyshrutimishra

Aug 24

Hugging Face quietly dropped FREE courses with certification

It cover everything from LLMs to diffusion models.

Here are the best ones you should bookmark today 🧵👇

1. LLM Course

Train, fine-tune, and deploy large language models.

🔗 huggingface.co/learn/llm-cour…

In this chapter, you will learn:

- Using the pipeline() function for NLP tasks like text generation and classification
- Understanding the Transformer architecture
- Differentiating between encoder, decoder, and encoder-decoder architectures and their uses

2. Agents Course

Build AI agents with multi-step reasoning using LangChain + HF.

🔗 huggingface.co/learn/agents-c…

In this course, you will:

📖 Explore AI Agents in theory and design.
🧑‍💻 Use AI libraries like smolagents, LlamaIndex, and LangGraph.
💾 Share and discover agents on Hugging Face Hub.
🏆 Compete in challenges to test your agents.
🎓 Earn a completion certificate after assignments.

Read 11 tweets

Shruti

@heyshrutimishra

Aug 22

Every coder should be using this stack:

→ Claude
→ CodeRabbit
→ Warp

It’s fast, AI-native, and feels like magic when it all comes together.

Here’s why you’ll never code the same way again: 👇

1. The CodeRabbit extension works with VSCode, Cursor, and WindSurf.

It reviews your code as you write, checks your code and identifies bugs

At any time, just ask it to "Review all changes."

2. Claude Code → your AI pair programmer

Generate functions, fix bugs, explain logic.

Ask it to “build a Chrome extension” or “debug this API”... it delivers.

Read 6 tweets

Shruti

@heyshrutimishra

Aug 20

🚨 BREAKING: Google just killed the old smartphone.

Phones, watches, buds, foldable … all powered by Gemini.

This might be the biggest phone upgrade in 10 years: 🧵👇

1. @google JUST dropped the Pixel 10 Series

Pixel 10, 10 Pro, 10 Pro XL.

→ New design with Material 3 Expressive UI
→ Pixel Snap with Qi2 magnetic accessories
→ 7 years of OS + Pixel Drops

Preorders → Aug 28 #MadeByGoogle

2. Pro Res Zoom → 100x with generative AI

The Pixel 10 Pro and Pro XL take zoom to a new level. This isn't just cropping, it's an on-device generative imaging model that intelligently refines details at extreme distances

Delivering stunning clarity even at 100x magnification. It's powered by the new Tensor G5 chip, making it perfect for capturing distant subjects.

Read 18 tweets

Shruti

@heyshrutimishra

Aug 5

🚨 BREAKING: Google DeepMind just launched Genie 3

You can build interactive environments - live, at 24 FPS, from just a prompt.

This could change games, agents, films… even AGI research.

But what it reveals about the future of AI is even bigger: 🧵👇

1/ Most AI models just generate pixels.

Genie 3 goes further - it simulates a world.

You type a prompt.
It renders a live environment.
You can walk around, interact, revisit places… and it remembers.

AI just got a memory of the world. And a body.

2/ Let’s say you type:

> “Walking through a Zen garden in Kyoto. Morning light. Bamboo fence. Still pond.”

Genie 3 will render it in real time.

You can walk, turn, and revisit places and the model remembers what it generated earlier.