Tech, reimagined with AI 🤖 Daily AI tools, workflows & use-cases 🎯 Turn AI into income 💰
- Tweets 16,836
- Following 270
- Followers 59,462
- Likes 21,456
Pinned Tweet
AI video just got a CTRL+Z for terrible audio.
AI Videos finally sound real.
Introducing Mirelo SFX 2.0.
It watches your video, cuts the music you never prompted, keeps the dialogue, and builds a full sound design synced to the action and consistent across any length clips.
You’ve made it look right. We’ll make it sound right.
Live today through our API, on @runwayml and @fal.
Other platforms coming soon. (Thread)
9
34
72
45,670
When an AI agent makes dozens of model calls, latency can quickly add up. Fast inference infrastructure matters.
@General_Compute is adding @positron_ai’s Atlas systems and has committed to its next-generation Titan systems.
The idea is to use different chips for different stages of inference: GPUs for compute-heavy prompt processing and memory-first systems for token generation.
With Titan targeting larger models and longer context, this partnership is worth watching as AI agents become more capable.
Atlas today, Titan next. Available Q1 2027.
Our heterogeneous cloud just got better. We're announcing a cluster scale deployment of @positron_ai as our next partner.
We believe demand for fast inference will outstrip supply for the next decade. Our bet on Positron is another step towards getting fast inference for opensource models into builders' hands.
Hardware goes live Q1 2027.
Thanks to @mitesh711, @trsohmers and the whole Positron team! We're excited for this partnership.
LFG!
3
1
23
14,979
codemarch retweeted
Our heterogeneous cloud just got better. We're announcing a cluster scale deployment of @positron_ai as our next partner.
We believe demand for fast inference will outstrip supply for the next decade. Our bet on Positron is another step towards getting fast inference for opensource models into builders' hands.
Hardware goes live Q1 2027.
Thanks to @mitesh711, @trsohmers and the whole Positron team! We're excited for this partnership.
LFG!
101
54
501
172,703
codemarch retweeted
The future of robotics isn't just about what robots can do. It's also about how they fit into our lives.
That's what makes Tuesday Labs' Voxel interesting.
This droid butler is being designed around everyday tasks, from putting things back where they belong to fetching items and tending to plants.
I like that the team is thinking about the experience of living with a robot, not just its capabilities.
Voxel is expected in 2028.
10
2
53
35,292
codemarch retweeted
Please stop letting AI generate images that make every brand look identical. We can tell. 👀
The problem isn't AI itself. It's that image models often reproduce the same familiar patterns while getting logos, fonts, and product details wrong.
Vyb by @cocoailabs starts with your real brand assets and builds each design element on its own editable layer.
Change the headline without rebuilding the entire graphic, keep your actual product photos, and export to Figma when you're ready to continue.
That's a more interesting direction for AI design.
I launched the first image and video generation models at Google, and kept asking why all AI designs look the same.
AI has brought us the WORST era of design.
It made content creation effortless, but it also made 90% of it look painfully generic.
My co-founder and I spent 2 years building in stealth to change that.
Meet Vyb, the first AI design model that understands your taste like a designer would.
RT + comment your website URL and we’ll make your first design.
10
3
62
29,392
codemarch retweeted
Imagine using a parser's uncertain answers to decide its next lesson.
The lesson plan works like this: a VLM reviews outputs, repeated runs expose disagreement, clustering sorts the causes, and the next training data targets those problems.
That's OXR's RSI approach during post-training, the first in any document parser.
Classroom setup: a separate layout model, with recognition and reading order in parallel.
Report card: open-source speed #1 (101s vs GLM-OCR's 308s, single RTX PRO 6000) and first on olmOCR-Bench, OmniDocBench, DocParsingBench.
Layout: replaceable.
github.com/SoMarkAI/OXR
Beyond OCR. Meet OXR.
Only 1.3B. Open source.
#1 on olmOCR-Bench, OmniDocBench & DocParsingBench🏆
#1 in parsing speed among open-source models⚡
RSI-powered. A first in document parsing.
Turn documents into Agent-Friendly data.
If OXR helps, give us a star on GitHub! 🌟
16
2
69
8,159
codemarch retweeted
Imagine asking AI to create a product ad and getting back a version of your product that doesn't quite look like the real thing. Not ideal.
Vyb by @cocoailabs uses your actual product photos instead of redrawing them.
It also pulls in your brand's logo, fonts, and colors, then builds the design in editable layers.
Need to change just the headline? You can edit that element instead of starting over.
And when you're ready, export the layers to Figma.
This is the kind of control AI-assisted design needs.
I launched the first image and video generation models at Google, and kept asking why all AI designs look the same.
AI has brought us the WORST era of design.
It made content creation effortless, but it also made 90% of it look painfully generic.
My co-founder and I spent 2 years building in stealth to change that.
Meet Vyb, the first AI design model that understands your taste like a designer would.
RT + comment your website URL and we’ll make your first design.
9
6
62
14,911
Imagine asking AI to create a product ad and getting back a version of your product that doesn't quite look like the real thing. Not ideal.
Vyb by @cocoailabs uses your actual product photos instead of redrawing them.
It also pulls in your brand's logo, fonts, and colors, then builds the design in editable layers.
Need to change just the headline? You can edit that element instead of starting over.
And when you're ready, export the layers to Figma.
This is the kind of control AI-assisted design needs.
I launched the first image and video generation models at Google, and kept asking why all AI designs look the same.
AI has brought us the WORST era of design.
It made content creation effortless, but it also made 90% of it look painfully generic.
My co-founder and I spent 2 years building in stealth to change that.
Meet Vyb, the first AI design model that understands your taste like a designer would.
RT + comment your website URL and we’ll make your first design.
9
6
62
14,911
codemarch retweeted
I launched the first image and video generation models at Google, and kept asking why all AI designs look the same.
AI has brought us the WORST era of design.
It made content creation effortless, but it also made 90% of it look painfully generic.
My co-founder and I spent 2 years building in stealth to change that.
Meet Vyb, the first AI design model that understands your taste like a designer would.
RT + comment your website URL and we’ll make your first design.
184
143
732
295,270
A robot with the personality of a gentleman butler? Now that's a fun direction for robotics.
Meet Voxel from Tuesday Labs, a home robot designed to keep things tidy, find misplaced items, fetch things, and care for plants.
The focus on character and thoughtful design makes this more than just another robotics announcement.
I'm curious to see how Voxel develops.
Expected to ship in 2028. 👀
19
4
63
39,888
Marketing is getting a serious AI upgrade.
Zavi finds your biggest growth bottlenecks and handles ads, SEO, social media, and emails on autopilot.
You set the budget. AI does the heavy lifting.
Introducing Zavi.
Growth on autopilot. ✨
Give it your website. Zavi finds what’s holding your growth back, then runs your ads, SEO, social posts, and emails.
12
29
83
12,490
codemarch retweeted
Took one full-body photo of myself [where you took it] and gave it to Viggle's JST-2
it rebuilt the room around me in 3D, with a 3D me standing in the middle of it. no 3D software, no scanning, just one photo from my phone
then I [walked the camera around myself / looked at the room from behind me / made 3D me dance]. [the detail that got you]
memes were fun, but this is me inside an actual 3D world.
Today we are releasing JST-2, the world's first full-stack physics-native world foundation model.
Many world models learn in pixel/video space. We went back to first principles and built the entire model stack in 3D.
Generate a 3D scene in seconds and expand it in real time. Create characters with motion built in, all through one general architecture.
4
3
60
23,920
An AI learning experience gets interesting when agreement stops being automatic.
Growth doesn't come from a tool validating everything a learner says—sometimes, it takes a little constructive friction.
Making that work in real time takes more than a capable model.
@gakkuai is a bet on a critical layer missing from today’s AI stack: an interaction layer that keeps voice, visuals, tools, and learner state coherent.
78
121
262
64,449
Across different sessions, persistent context determines how a new task opens up. Raku remembers that the learner loves football and uses Ronaldo to invite them into a new reasoning exercise.
Instead of treating memory as a background log, this recollection changes what happens next.
The child steps into a fresh challenge through a topic they already care about, making the next prompt feel like a natural continuation of their own interests.
1
3
73
The showroom simulation shows Gakku’s core design in action: gakku.ai
Coordinating real-time voice, visuals, and learner state allows Raku to respond to developing thinking—stepping in with a question, a hint, or an alternative perspective at just the right moment.
That dynamic layer is what makes Gakku an AI-native consumer app built around the idea that the learning process itself is the product, rather than just generating a quick answer.
2
71