The AI world moves fast. We bring you the moments that matter. Tools ✦ Tips & Tricks ✦ Insights ✉️ Biz [email protected]
California Usa 🇺🇸
Joined September 2026
- Tweets262
- Following155
- Followers11K
- Likes363
Maya retweeted
Your notes app is full of startup ideas. Your GitHub is waiting for its first commit. 😂
Now you have fewer excuses.
@boltdotnew is free all weekend, giving you a chance to turn one of those ideas into an actual product.
Builders Week also has $10K up for grabs, with submissions closing October 20.
Pick an idea. Open Bolt. Start building.
🤝 Paid partnership
Bolt is free for EVERYONE, all weekend long.
And you could win $10k!
Turn your idea into a product.
Submit it by Oct 20 for your shot at the prize.
Bolt Builders Week starts now 🏁
Details here 👉 bolt.new/builders-week
Maya retweeted
This is why efficient AI matters.
MiniCPM-V 4.7 didn't just read a humidifier's display. It interpreted the dial, spotted a hidden switch, and explained the settings, all on a phone.
No cloud. Around 3GB of memory.
The next big AI breakthrough might be intelligence that fits in your pocket.
2 a.m. Furnace running, air bone-dry, and a humidifier control in the basement I've never touched. The Wi-Fi barely reaches down there.
So I asked the model on my phone. MiniCPM-V 4.7: 35B total, 3B active, fully on-device in about 3 GB of memory. No cloud.
One photo of the panel, one question: "Help me understand the right settings for the winter."
It read the LCD 43% and told me that's already inside the 30–50% winter range. It read the dial: 1–7, "Less" on the left, "More" on the right, keep it around 4–5. Then the part I didn't expect it caught the tiny Blower Activation switch at the bottom of the panel and reminded me it has to be ON, or the humidifier does nothing. It hedged on exactly which number "Normal" is ("likely 4 or 5"), which is about what a person reading the same photo would say.
How a 35B MoE model runs on a phone at all: Lookahead Expert Scheduling. Experts live on storage, and the runtime predicts which ones the next layer needs and prefetches them while the current layer computes OpenBMB reports 13+ tok/s decode in just 3 GB of memory on an iPhone 17 Pro Max, about a 5× speedup from LES.
Model (open weights): huggingface.co/collections/o…
Code: github.com/OpenBMB/MiniCPM-V
LES write-up: github.com/tc-mb/llama.cpp-o…
This is why efficient AI matters.
MiniCPM-V 4.7 didn't just read a humidifier's display. It interpreted the dial, spotted a hidden switch, and explained the settings, all on a phone.
No cloud. Around 3GB of memory.
The next big AI breakthrough might be intelligence that fits in your pocket.
2 a.m. Furnace running, air bone-dry, and a humidifier control in the basement I've never touched. The Wi-Fi barely reaches down there.
So I asked the model on my phone. MiniCPM-V 4.7: 35B total, 3B active, fully on-device in about 3 GB of memory. No cloud.
One photo of the panel, one question: "Help me understand the right settings for the winter."
It read the LCD 43% and told me that's already inside the 30–50% winter range. It read the dial: 1–7, "Less" on the left, "More" on the right, keep it around 4–5. Then the part I didn't expect it caught the tiny Blower Activation switch at the bottom of the panel and reminded me it has to be ON, or the humidifier does nothing. It hedged on exactly which number "Normal" is ("likely 4 or 5"), which is about what a person reading the same photo would say.
How a 35B MoE model runs on a phone at all: Lookahead Expert Scheduling. Experts live on storage, and the runtime predicts which ones the next layer needs and prefetches them while the current layer computes OpenBMB reports 13+ tok/s decode in just 3 GB of memory on an iPhone 17 Pro Max, about a 5× speedup from LES.
Model (open weights): huggingface.co/collections/o…
Code: github.com/OpenBMB/MiniCPM-V
LES write-up: github.com/tc-mb/llama.cpp-o…
Maya retweeted
October 20–30 is your window for 60% off Pro and 40% off Ultra, only on CapCut Desktop.
Jean Phil brings a useful angle to Black Friday planning: build a campaign around an AI character people can recognize.
Use CapCut PC to design an original character, generate AI video, and refine the result manually on the multitrack timeline.
For a vertical edit, adjust the clip's scale and position so the character's face and the offer text both have enough space.
Sound helps complete the personality. Seed Audio 1.5 is coming, with one-prompt generation of emotional dialogue, ambience, sound effects, and music, alongside separate tracks and precise timestamp control.
CapCut PC gives you room to make the layout work. Save 60% on Pro or 40% on Ultra, October 20–30, exclusively on CapCut Desktop.
#capcutpc #capcutmarketing #capcutsale #capcutblackfriday
🤝 Paid partnership
Maya retweeted
AI can describe a video, but can it actually understand what's happening?
There's a big difference between recognizing objects and figuring out how they move, interact, and change over time.
That's the problem @ElorianAI is tackling with Glimpse.
Built by a team whose founders led work on Gemini and Apple's MM-1, it's focused on visual reasoning, not just image descriptions.
This is a space worth watching.
🤝 Paid partnership
Dreamina’s $1.5/month Basic deal is still live. 4K on Seedance 2.5 now costs less than 720p!
Draft at 480p. Reroll. Find the keeper. Then use Creative Upscale to take your favorite clip to 4K.
Why upscale every attempt when you can focus on the one worth keeping?
Start creating on Dreamina 👇
dreamina.capcut.com/ai-tool/…
#Dreamina #DreaminaPartner
🤝 Paid partnership · 🤖 Made with AI
Let's break down five attempts on the exact same 10-second shot: 📉
720p × 5 = $9
480p × 5 = $4
One 4K upscale = $1
You end up spending just $5 instead of $9, and your final output is still in stunning 4K.
#AIcreator
Maya retweeted
I installed MiniCPM-V 4.7 on my iPhone. 35B total, 3B active, fully on-device in about 3 GB of memory. No cloud.
Thanks @aijoey for the ideation！
Pointed it at the Einstein poster on my wall and asked one thing: "What is this photo of?"
It got Einstein. It got the pop-art style, the rainbow hair, the tongue out. Then it did the part I didn't expect — it read the text in the background: E=mc², "GENIUS GONE MAD". It fumbled one line of tiny print, but that's from a photo taken across the room.
How a 35B MoE model runs on a phone at all: Lookahead Expert Scheduling. Experts live on storage, and the runtime predicts which ones the next layer needs and prefetches them while the current layer computes — OpenBMB reports 13+ tok/s decode in just 3 GB of memory on an iPhone 17 Pro Max, about a 5× speedup from LES.
Model (open weights): huggingface.co/collections/o…
Code: github.com/OpenBMB/MiniCPM-V
LES write-up: github.com/tc-mb/llama.cpp-o…
Maya retweeted
Here's a question AI filmmakers need to think about: when AI creates the video, how do you prove what YOU contributed?
@buildwithVN is tackling this with VN Studio, combining AI filmmaking tools with human authorship tracking.
Your edits, creative choices, and revisions matter. They shouldn't disappear behind a single prompt.
The future of AI filmmaking needs more than generation. It needs attribution.
🤝 Paid partnership
Your notes app is full of startup ideas. Your GitHub is waiting for its first commit. 😂
Now you have fewer excuses.
@boltdotnew is free all weekend, giving you a chance to turn one of those ideas into an actual product.
Builders Week also has $10K up for grabs, with submissions closing October 20.
Pick an idea. Open Bolt. Start building.
🤝 Paid partnership
Bolt is free for EVERYONE, all weekend long.
And you could win $10k!
Turn your idea into a product.
Submit it by Oct 20 for your shot at the prize.
Bolt Builders Week starts now 🏁
Details here 👉 bolt.new/builders-week
Maya retweeted
MiniCPM-V 4.7 35B-A3B, running on an iPhone.
A 35B MoE vision model with ~3B active, fully on-device with just 3 GB of memory. One photo of a weathered window, one question — and it actually read the photo 🧵
GitHub: github.com/OpenBMB/MiniCPM-V
HF: huggingface.co/collections/o…
Maya retweeted
Hot take: you don't need AI to generate another image. You need AI to help you get the image you actually want.
That's what makes @envato interesting.
With 29M+ human-made creative assets and Edit with AI, you can start with a real photo, change the setting, adjust the lighting, and make it fit your vision.
Start real. Riff freely.
🤝 Paid partnership
October 20–30 is your window for 60% off Pro and 40% off Ultra, only on CapCut Desktop.
Jean Phil brings a useful angle to Black Friday planning: build a campaign around an AI character people can recognize.
Use CapCut PC to design an original character, generate AI video, and refine the result manually on the multitrack timeline.
For a vertical edit, adjust the clip's scale and position so the character's face and the offer text both have enough space.
Sound helps complete the personality. Seed Audio 1.5 is coming, with one-prompt generation of emotional dialogue, ambience, sound effects, and music, alongside separate tracks and precise timestamp control.
CapCut PC gives you room to make the layout work. Save 60% on Pro or 40% on Ultra, October 20–30, exclusively on CapCut Desktop.
#capcutpc #capcutmarketing #capcutsale #capcutblackfriday
🤝 Paid partnership
Check out : tinyurl.com/2v23rn22
Maya retweeted
The researchers behind Gemini + Apple's MM-1 just dropped a new AI model.
Elorian's Glimpse is built to actually understand what it sees.
Tricky counting. Visual puzzles. Video reasoning.
Visual Thinking
🤝 Paid partnership
Maya retweeted
Everyone is talking about AI coding agents.
Meanwhile, one of the biggest gaps in AI is still understanding what happens in the real world.
Videos contain movement, interactions, changing perspectives, and sequences that a single frame can't explain.
Glimpse from @ElorianAI aims to tackle that with deeper visual reasoning.
If AI learns to understand videos instead of merely describing them, a whole new set of applications opens up.
🤝 Paid partnership
Maya retweeted
The biggest problem with AI agents isn't what they can do.
It's how much access you need to give them.
@Selfxyz just introduced Agent Vault, letting agents pay and log in without exposing your real credentials.
You stay in control with spending caps, approvals, and a kill switch.
This is how AI agents should work.
🤝 Paid partnership
AI can describe a video, but can it actually understand what's happening?
There's a big difference between recognizing objects and figuring out how they move, interact, and change over time.
That's the problem @ElorianAI is tackling with Glimpse.
Built by a team whose founders led work on Gemini and Apple's MM-1, it's focused on visual reasoning, not just image descriptions.
This is a space worth watching.
🤝 Paid partnership