スレッド

新しいツイートを表示

会話

What am I looking at? Is it possible that the 1T GPT-4 model is fine-tuned on a weekly basis? Is "Developer Livestream" like a `develop` branch CI/CD?

ツイートを翻訳

午後9:46 · 2023年4月8日

2万

件の表示

ブックマーク

返信先:

さん

source: twitter.com/swyx/status/16 time: 0:37s

引用ツイート

swyx

@swyx

4月8日

Andrey from Openai demoing how to make a ChatGPT plugin in <1 minute with @replit this is seriously smooth, all the openapi stuff autogenerated by @FastAPI

このスレッドを表示

メディアを再生できません。

再読み込み

返信先:

さん

Hope it's something more exciting than this, but could it be related to the GPT-4 Developer Livestream they did on Youtube three weeks ago? Has exactly the same name: youtube.com/watch?v=outcGt Other than that, it's well known that they are constantly improving the model, it's the… さらに表示

返信先:

さん

, good catch with the "GPT-4 Developer Livestream". It's probably a snapshot for the live stream demo.

返信先:

さん

Wow not even nightly? for shame

返信先:

さん

Still, it would be a good start

返信先:

さん

For sure it's the case!

返信先:

さん

If weekly fine-tuning occurs, then my "sleep theory" is quite plausible.

引用ツイート

Dom Sip

@dom_sipowicz

4月2日

AGI models may "sleep," pausing cognitive tasks. During this break, short-term memory, including conversation history, thought chains, and gained knowledge, is transferred to long-term memory by fine-tuning the AI model itself.

このスレッドを表示