Matt Macosko Creater of Divine Tribe & Nice Dreamz https://xitter.cf/t.co/6dEjmThOfj https://xitter.cf/t.co/v3poDJ7y9R https://xitter.cf/t.co/w0xMMVFey1

california
Joined November 2012
NiceDreamz retweeted
Deadass thought she was a zombie
158
1,424
112
27,477
2,136,962
Apple says you need a Mac to build iPhone apps. Today I built one on Omarchy Linux, submitted it to the App Store, installed it on my iPhone over USB, and stopped it at a breakpoint in the debugger. No Mac anywhere. It's all open source: github.com/joshuaswarren/oma…
335
877
154
9,593
538,616
After Opus 5.5, I think it's time to retire "vibe coding." This is just how programming works now.
180
288
43
10,729
569,377
qwen 3.8 vs gemma 4, both running local on my M5 mac. same prompt, one try, and a script decides pass or fail. a fur ball you can pet, a spinning ball, and a playable breakout. gemma passed 2 of 3, qwen 1 of 3 but twice as fast. youtu.be/EG_tYjaZbY0
278
Which local AI should drive your browser? 🧠 I gave Gemma 4 and Qwen 3.8 four jobs in four browser tabs at once, all on my Mac, no cloud Both got every answer right. Gemma was over 2x faster Want a browser agent that multitasks? Start with Gemma 4 #LocalAI #AIAgents
78
I gave 2 local AI models 4 browser tabs and 4 jobs at the same time 🧠 Gemma 4 vs Qwen 3.8, all local on my M5 MacBook, 128GB Both went 16/16. Gemma ~30% faster Twist: 4 tabs at once barely helped. The brain takes turns. Fixing that next Who gets the rematch? #LocalAI #MLX
69
NiceDreamz retweeted
What an absolute cracker of a report... Just facts, no drama or exaggeration, and goosebumps at the end! 🙌 Another fact- No Indian media person can ever ever match this...
181
1,600
128
7,933
523,099
NiceDreamz retweeted
A significant amount of Suno AI's source data has been leaked. Upon investigation, it was found that: Their primary tech stack is MERT25, and they’ve used web crawlers to gather a large amount of YouTube Music content....
25
217
26
2,356
351,200
NiceDreamz retweeted
You can’t predict the breakthroughs.
At 10:06pm on Aug 3, I watched a robot do something I thought was years away. We were working on few-gradient learning with GEN-1.5, wondering how far we were from landing physical prompting: show the robot a task once, and it just does it. We YOLOed it, and the robot imitated exactly what we demonstrated, with 0 fine-tuning. We gave more prompts for different tasks. More successful rollouts landed. I could not believe my eyes. Looking back, this represents the culmination of a huge amount of work across the entire team, many failed experiments, and years of groundwork from the community. I just happened to be the one standing in front of the robot when it all clicked. @andyzengineer once asked me: "If Generalist failed tomorrow, what would we want to have left behind that could benefit humanity?" I believe we found one answer: an existence proof of physical prompting.
10
13
319
42,027
A young Nigerian built AI smart glasses that help blind people walk without a cane or a guide dog. Here's how it works. The glasses use AI, sensors, and a camera to detect obstacles in real time. If something is in the way, they vibrate and trigger an alarm before the user reaches it. The clever part built-in GPS tracking. If the user gets lost, their exact location is sent to their caregiver's phone via WhatsApp or Email. He built the whole thing with an ESP32 microcontroller. Handmade. From Kano, Nigeria. Now he's adding AI voice so the glasses can describe surroundings out loud not just detect obstacles but explain what's around them. His name is Khalifa Aminu. @inventorkhalifa
54
629
32
2,274
107,773
I abliterated NVIDIA's Nemotron Omni for Apple Silicon and kept all three senses intact. It reads text, sees images, and hears audio, fully uncensored, running in MLX. As far as I can tell, it's the first one like this on a Mac. Three sizes, so it fits anything from a 32GB MacBook on up. Free on Hugging Face.
1
1
103
abliterated Meta's new Muse-Glimmer 30B and kept the vision tower, so one uncensored model both answers and sees, 100% local on a Mac. no cloud. badge blurred, it still ID'd the exact Jaguar F-Type from the spoiler. read a neon sign. pulled Q4 off a chart. weights + code open: github.com/nicedreamzapp/mlx…
1
90
NiceDreamz retweeted
In May 2016, 15-year-old Karinya “Yaya” Chen was in hospice care in Austin, Texas after years of fighting cancer. She had tickets to see Florence The Machine, but by the night of the concert, she was simply too sick to go. So the people at Hospice Austin reached out to the band. And Florence Welch came to her. Florence and guitarist Rob Ackroyd walked into Karinya’s room and gave her the concert she couldn't make it to. They sang “Shake It Out” and “Dog Days Are Over,” with Karinya singing and harmonizing right alongside them. A few months later, Karinya lost her battle with cancer at just 15 years old. But for one afternoon, none of that seemed to matter. There was just a young girl, one of her favorite singers, a guitar, a room full of people who loved her... and music. Sometimes the smallest act of kindness can become one of the biggest moments of someone's life. This is what being human is supposed to look like.🙏
858
5,813
510
90,642
5,133,078
my local AI was pausing 7 seconds before every reply once a conversation got long. same model, same MacBook. turned out the KV cache silently stops being reusable past Gemma's sliding window, so every turn re-prefills the whole transcript. no error, it just gets slow. one fix later: 0.36s. story and code in the first comment.
1
36
nvidias nemotron omni sees, hears and reasons, but on a mac only its text half loads with standard mlx tooling. the vision and audio towers had no apple silicon runtime, so i wrote one. 23 of 23 parity tests against nvidias own pytorch reference. 22gb peak, wifi off.
1
42