@hsyntax
Account based inUnited States
About this account
- Account based in
- United States
- Connected via
- United States App Store
Account-level information from X, not a live location or the device used for a specific post.
https://xitter.cf/t.co/Gk4ajJafY1
San Francisco
Joined May 2013
- Tweets1.3K
- Following352
- Followers165
- Likes1.9K
Duet is a great name. Probably the best of the lot
Its amazing that I can raise PR's like this with just high-level architectural reviews and not even have to look at the code much
I've cancelled my Xero subscription because its easier to just take the bank feeds as an excel file and give it to an agent to do everything.
I’m finally building my own personal cloud stack starting from the bottom layer of the stack. Storage and compute.
Harish retweeted
Introducing Expand Fetch, a new way to scrape and crawl the web.
Your agent has never seen the web. No wonder it gets basic questions wrong.
Traditional Markdown APIs lose critical details. So we trained two System one models to change that:
Iris, a vision model that sees pages like you do, preserving tables, icons, and layouts.
X-Ray, a JSON model that uncovers the hidden data behind every page.
Every fetch also comes with a Wayback Machine. Every line links back to its source, so you can verify exactly what your agent saw.
Three products. One API.
On our benchmark of 1,094 questions across 326 challenging web pages, agents using Expand got 74% right. The next best API got 55%.
Available today at $1 per 1,000 fetches.
Try it at expand.ai
X has done a great job keeping me in my echo chamber.
A gentle reminder to look outside once in a while.
dots <> tailscale. wen?
team landed a bunch of updates to dots 🙌
new things:
- create your dot on iOS / Android + open ChatGPT straight into it
- kick off Codex work and follow up on threads, with context from your chats, Codex threads and automations
- review + edit Scheduled Tasks
and a whole bunch of fixes:
- faster cloud browsing + sidebar loading, fewer pauses, smoother scrolling, less network usage
- fewer phone notifications when you’re already using ChatGPT
- cleaner code blocks + attachments, fixed Safari / Firefox character rendering
- smarter thread reuse, automatic renaming, fewer repeated “created task” messages
- clearer plans, retries when activity won’t load, more stable safety banners + approval buttons that don’t get clipped or hidden
- Enterprise web + voice access fixes
keep the feedback coming!!
I think the personal AI cloud is more important
Okay, f*ck it. I’m going to be the one to say it: A LOT OF RED/VAPORWARE FLAGS WITH THIS ONE.
First: There’s no way this is profitable. I simply don’t see how they’re making money on this.
Second: Performance won’t be that great:
>They’re selling this as if it has 88 GB of unified memory, when in reality only the 24 GB of VRAM will be efficient for inference. DDR5 with that processor will offer around 70 GB/s of bandwidth. If a model doesn’t fit entirely in VRAM, you’re going to have a hard time. So be prepared to run smaller models under 35B parameters or heavily quantized larger ones.
>An RTX 3090 could outperform this thing for roughly a third of the price. It has more than twice the memory bandwidth (936 GB/s versus 432 GB/s), more CUDA cores, and better potential for scaling thanks to NVLink support.
The only area where that RTX Pro 4000 wins is power consumption. But if that’s the priority… M-series Macs do exist.