[thm]

Can we not link to AI grifters on Twitter and instead use the original source?

https://www.ft.com/content/dd9ba2f6-f509-42f0-8e97-4271c7b84...

https://archive.ph/r1O38

[VHRanger]

This just in: company working on next version of popular product

[vincent_s]

For context: Back in April, OpenAI said they aren't training GPT-5 and ‘won’t for some time’.

[musha68k → vincent_s]

I’d guess they have meanwhile unlocked “bigger war chest” to try see how far they can scale up their process?

IIRC they actually said they don’t know if it will. Just like last time. Optimization will come later just like they are trying now with bringing GPT-4 to the masses.

[3cats-in-a-coat → vincent_s]

Half a year is "some time". Especially these days. Welcome to the singularity.

[highwayman47]

I thought OpenAI said fine tuning and not bigger models were the future?

Side note: Wonder if Twitter will be filled with graphs of the motivation crowd’s interpretation.

Tiny circle = GPT4 Huge circle = GPT5

[3cats-in-a-coat]

Please for the love of God don't beat this one with RLHF stick until it's terrified of saying anything other than being a glorified Wikipedia clone.