I’m not sure why all the demos are single tasks that are already very optimized. Agents can do so much more. Like booking flights or buying food delivery.
Like let’s see multitask objectives that have no single happy path I.e. optimizing airline miles and credit card points across multiple systems to get discounted flights.
We're giving away $1,000,000 of free pizza right now
Download and order from the Hark app: https://t.co/gXfv7WktGR
Type "order a free pizza" in chat. It's first-come, first-served, delivered only in the U.S. (review details below)
I’m suprised E-Ink tablet menus haven’t been adopted by high end restaurants, especially ones with large wine books.
No more people busting out their phones to read the menu in the dark? Easy to change to reprints? They probably already spend 8-10k a year on printing 🤷♂️
The economics are starting to check out already… 1 RTX pro blackwell hosting of Qwen 3.8-27B with continuous batching, you can rip inference in aggregate >350k tok/s …
20% below market rate at 50% saturation - payback period is ~a year, earning $855/month
Now obviously hitting that saturation is the tricky part but there will eventually be marketplaces that will handle routing and if you’re pushing price down you should be able to saturate harder. Also you can probably optimize way way more and get wayyyy higher TPS