Pradeep Pandey

Apple flipped 9 privacy settings to ON the moment you activated your iPhone. Location tracking. Microphone access. Your exact shopping patterns in physical stores. All quietly running in the background. Here are 9 settings to turn off in the next 5 minutes

@Damn_coder This insight really sheds light on how we engage. Thanks for sharing your thoughts!

X’s open-source algorithm repo revealed something most creators still don’t understand: The platform is no longer ranking “good content.” It’s ranking behavioral probability. Not: • quality • accuracy • expertise But: • likelihood of reply • likelihood of pause • likelihood of outrage • likelihood of sharing • likelihood you stay on the app longer The feed is basically a prediction engine now. Here’s what the repo quietly exposed: 1. X predicts your behavior in real time The system scores tweets based on: • likes • replies • profile clicks • dwell time • shares • negative feedback Then combines those predictions into a final ranking score. 2. Dwell time matters more than most people think A post people STOP for can outperform a post people casually like. That explains why: • curiosity hooks work • cliffhangers work • long-form posts work • emotionally charged posts spread faster The algorithm rewards retained attention. 3. The “For You” feed is mostly discovery now The system aggressively pushes content from people you don’t follow. Meaning: • virality matters more than followers • niche accounts can explode overnight • one strong post can outrank massive creators 4. Replies are algorithm fuel Reply-heavy posts create: • longer sessions • repeated interactions • deeper engagement loops That’s why controversial posts dominate the feed. 5. The biggest shift: Old Twitter: “Who do you follow?” Modern X: “What behavior can we predict?” That’s the real story behind the repo. The feed isn’t curated anymore. It’s engineered attention.

@Rixhabh__ 😂😂

This guy used AI to put himself into Squid Game and fixed everything

@thetripathi58 Love seeing smaller models punch way above their weight

Everyone is chasing bigger models. More parameters. More compute. More cloud dependency. OpenBMB went the other way. MiniCPM-V 4.6 - a ~1B vision-language model that outperforms models three times its size. I ran it on my iPhone. Here's what happened:

@hey_mujeebahmed @renoiseai Looks good.

Been using AI tools for months. Generation part is solved. Keeping everything organized? Still a mess. Tested Canvas by @renoiseai. Finally feels like someone built this right. #RenoiseCanvas

That's a wrap! If you liked this thread: 1. Follow @Div_pradeep for more such content around AI, web development. 2. Bookmark and repost the first tweet to share it with other people.

Want more AI drops like this? Join 30,000+ readers getting the latest AI tools, product breakdowns, and step-by-step guides delivered every week. Subscribe free: joinaiinsights.com/subscribe

Your iPhone is hiding 47GB of storage you did not know existed. Not in photos. Not in apps. In places Apple designed you to never check. I found them all in 10 minutes. Here is where:

Most multimodal AI models can't do this. One prompt → full 3-panel manga story. Line art. Shading. Sequential panels. Consistent tone across all frames. SenseNova U1 thinks and draws in the same space. And it's open-source. Here's how the architecture actually works: 👇

@Rixhabh__ 2 second inference on H100 is going to make a lot of other models look slow.

@Damn_coder Worth watching.

You can scroll Netflix for an hour. Or spend 2 hours learning how LLMs actually work at a level most AI employees never reach. This Stanford lecture is the clearest breakdown I've seen. Bookmark it now

@iamjordan Must have tool.

🎉 JUST IN: Wan 2.7 and Seedance2.0 now available on GlobalGPT — free to try! Create AI videos with dynamic comic IPs, lively ads, energetic dance videos, and unique running videos in various styles, all powered by Happy Horse and Seedance. No limits. No regional barriers. No invite codes. 👇