For techies - Twitter thread on OpenAI's next-gen AI data center

[highplainsdem]

Found thanks to a tweet about this from software engineer Grady Booch.

This thread about the hardware is from Tim Zaman - https://www.linkedin.com/in/timzaman .

First tweet below, Thread Reader page at https://threadreaderapp.com/thread/1636981863477809152.html

Link to tweet

[JanMichael]

1. Well at least we can see our future overlord being created.

Oh well. Being a human was cool. Nice while it lasted.

[Maru Kitteh]

2. Having used ChatGPT a few times, I no longer fear Hal will regretfully

inform us that we must be terminated in order to save the mission.

It struck me so far as a blissfully less spammy version of the Goog, but with some fairly decent cross-checking and graceful amalgamation.

[highplainsdem]

3. ChatGPT is notoriously unreliable. And OpenAI admits that.

[Maru Kitteh]

4. Indeed. You have to click through that "we're just making this up as we go"

CYA dialog box before you can do anything.

Still a fun tool. I was pressed for time before interviewing a prospective new employee and GPT did a pretty impressive job of spitting some relevant and clever questions out at me. I used a few of them.

[Renew Deal]

5. So are humans

They forget, lie, have biases, and they need downtime (sleep). Most want to be paid for their services, so they randomly stop providing service if someone pays them more. They’re also pretty slow when they are overwhelmed, which is easy to do.