Darius (@lam_darius) 's Twitter Profile
Darius

@lam_darius

Founder @ NEX . Scale Maximalist . Prev Cerebras Systems, DoD, Harvard . Views my own

ID: 2575152494

linkhttp://nex.art calendar_today18-06-2014 16:44:51

607 Tweet

1,1K Takipçi

2,2K Takip Edilen

Darius (@lam_darius) 's Twitter Profile Photo

Very likely the greatest downside to GANs is their numeric instability. Not just instability with training objective -- but instability with the model numerics too. At least with diffusion we can rest assured that the loss is stable & focus only on model numerics

Darius (@lam_darius) 's Twitter Profile Photo

My o1 workflow for research: - Auto collect & convert all relevant arxiv papers to MD - Do some basic content filtering - Add all relevant papers to o1 context w/ XML tag breakdown The annoying part is having to select relevant papers every time. But the added multi-paper

My o1 workflow for research:
- Auto collect & convert all relevant arxiv papers to MD 
- Do some basic content filtering 
- Add all relevant papers to o1 context w/ XML tag breakdown

The annoying part is having to select relevant papers every time. But the added multi-paper
Darius (@lam_darius) 's Twitter Profile Photo

There are two types of AI products: 1. Low-Autonomy (eg Pika, Cursor): Usually called 'tools'. Requires non-trivial skill to operate & extract value from. 2. High-Autonomy (eg Devin, Decagon): I put in dollars and get out labor. The end state is to have human guidance but not

Darius (@lam_darius) 's Twitter Profile Photo

Our Key innovation was to move the upscaling process out of latent space. Where existing methods like SUPIR, ResShift, and others rely on pre-trained stable diffusion models, operating in latent space, we built the NEX Upscaler OUTSIDE the latent space. Thanks to our highly

Our Key innovation was to move the upscaling process out of latent space. Where existing methods like SUPIR, ResShift, and others rely on pre-trained stable diffusion models, operating in latent space, we built the NEX Upscaler OUTSIDE the latent space. 

Thanks to our highly
Darius (@lam_darius) 's Twitter Profile Photo

Btw, the NEX team literally only has 4 engineers We're looking for fullstack and AI engineers to come join us while we're still small, and are offering a $15,000 award for successful referrals. At NEX we have one mission: Solve Marketing We believe Marketing is an

Btw, the NEX team literally only has 4 engineers

We're looking for fullstack and AI engineers to come join us while we're still small, and are offering a $15,000 award for successful referrals. 

At NEX we have one mission: Solve Marketing 

We believe Marketing is an
Alex Reichenbach (@a_reichenbach_) 's Twitter Profile Photo

Excited to share that Structify has raised a $4M seed round, led by Bain Capital Ventures & @8VC! We’ve been heads-down training our model & building our UI—and we’re just getting started. 🚀

Darius (@lam_darius) 's Twitter Profile Photo

Time is a flat circle ... Years ago (back in 2017!) my computer vision advisor built a consumer company (acq by VSCO) to edit short form videos for Insta & TikTok - Cursor for Video Editing! - but today's founders have forgetten the past My advisor's approach was to build a

Darius (@lam_darius) 's Twitter Profile Photo

Most recent Claude Code update is going to reduce my Cursor usage to near 0 My biggest gripe has been the inability to show in-line diffs in a code editor. But now it's possible & just as good as the Cursor integration Claude can also use model MCPs to debate with O3 and Gemini

Most recent Claude Code update is going to reduce my Cursor usage to near 0 
My biggest gripe has been the inability to show in-line diffs in a code editor. But now it's possible & just as good as the Cursor integration

Claude can also use model MCPs to debate with O3 and Gemini