15 comments

[ 5.2 ms ] story [ 39.2 ms ] thread
I enjoyed Trump and Joe playing Minecraft a lot more.
For those that haven't seen it: https://www.youtube.com/@presidentialdiscord

Presidential Discord works amazingly well because the participants are acting how they frequently do: Trump is belligerent, Obama is calm, Biden is defensive.

The "Joe Rogan AI Experience" doesn't work well because "I'm your host Joe Rogan" isn't something Joe Rogan would naturally say, and there's not a lot of examples of Joe Rogan reading something someone else has forced him to say like this.

Ok, this is the start of something crazy.
Altmans pauses and ums feel super annoying - I've listened to him speak a few times and it's nowhere near this annoying IMO.

Joe feels very flat, and the entire conversation feels flat - no ups/downs, speed up/down.

Very good technology showcase but as a podcast this is annoying to listen to.

This comment reminds me of this: https://youtu.be/PdFB7q89_3U
I'm not saying the tech demo isn't impressive, I'm just saying the content on its own is kind of crap. Which is a compliment to the tech on it's own I'd say - that it can even be considered in the same category.
ChatGPT generated content is remarkably timid. It is boring and not edgy.

Text to speech feels real, gets monotonous if you listen it for a while.

All this will be fixed though. We are in a new era now

ChatGPT is more akin a demo of what GPT4 can do, not a real product. To really get good and interesting content out of it, you need to use the API where you have better control of the various parameters like the temperature and the system prompt. The ChatGPT system prompt is boring, so the replies you get from ChatGPT are boring as well.
Early content generation before the RLHF safety tuning was a lot more interesting and nuanced. It's significantly worse now, both dumber and less diverse output. [1]

We're seeing the enshittification of early AI products in real time, presumably so they can be sold to the lowest common denominator who can't distinguish AI output from human-embodied thought and opinion, and so that companies like micro$oft can safely sell advertising space in/alongside it, or whatever their plan is.

[1] https://www.youtube.com/watch?v=qbIk7-JPB2c - mentioned fairly early in the talk that the "sparks of AGI" are not reproducible with the release version of GPT4.

Impressive and creative. However, you can tell it's scripted with the rhythm of their conversation. If you're not paying attention though, you would be easily fooled that it's real.
Why would you even... I mean, WGASA? (Obviously I do, but, c'mon)
It's hilarious but created flat material. JR was plausibly believable though. :)

I don't buy Sama speaks flatly like a lifeless, bizspeak drone reading a script written by a press release writer.

I'm curious what The Yes Men could and will do with AI.