Frontier pretraining is said to be a big-lab-only game. We don’t have 100k chips yet, so there’s only one way: algorithmic efficiency. Our new recipe matches DeepSeek V4 Pro’s pretrain using 50x less compute – that’s roughly half the FLOPs used for GPT3, or ~$0.5M on GB200.
https://t.co/uBs5o5Arsu
Activate AI with any compute engine, through Palantir AIP.
Palantir Dev Matthew Bayer shows @chadwahl how lightweight transforms are powering radically cheaper, faster data transforms — and moving beyond monolithic compute frameworks.
@EricSteinb You’re right, any long term solution needs to address tuition fees and predatory loan practices. The UK government loan is an interesting example of less predatory oans. Any time you make something better people could complain that it’s unfair because their experience was worse.
This picture, in fact, came from the launch of Long March 3B/Fengyun-4B on June 2nd this year, and as you can see carefully on the top of this LM-3B booster it has a PARACHUTE attached.
https://t.co/MtGGP0uXFP
@manuakasam@girl_whocode You most likely don’t want to use this as implementations vary and your output may be different based on the browser you use. I recommend a tool like quill, ckeditor5, etc. If you need rich text editing
@cr8tive_mikky @laurieontech I like it, you should add pagination because rn there are so many posts it makes everything quite slow to load. Also, the page re-renders very often making It even slower.