Post

austin (eeek!/ackkk! 👻)
@thebadcode.com

very underrated that you can use deepseek without having to listen to a guy holding a flashlight under his chin

This post is unavailable.

30 reposts250 likes

Oden
@odense.bsky.social

I do research in part on a 200B parameter model locally (part of my PhD investigation, not using it to research for me, just so we are clear), and there is not that huge of a gap between that and the US “frontier” models.

Oden
@odense.bsky.social

I would be interested to know how the American LLM shops think that they will stop the rest of the world from using open weight (and surely someday completely open) models? (You can also get a selection of Chinese EVs here.)

Bitey
@jsherman999.bsky.social

Here's the likely scenario: They're positioning themselves for when the price of compute drops, when they can offer cloud-based LLM services to enterprises and consumers more cheaply. If the delta between open weight and US shrinks, enterprises will pay a little extra to get Meta or Anthropic. >

Bitey
@jsherman999.bsky.social

this is a very real dynamic that plays out in a lot of areas. US firms could have moved to offshore coders and knowledge workers long ago to save salary $, but we hit barriers that made the $ delta not worth it. And buying HW to place on the raised floor to run local? Introduces other costs

Haunted Agent of the Deep State (Spook)
@swelljoe.bsky.social

Liang Wenfeng seems normal aside from being brilliant. He runs one of the smallest teams in the AI industry, spends orders of magnitude less money, and still manages to go head-to-head with the big guys. US companies accuse DeepSeek of "stealing" but they've all benefited from DeepSeek research.