>OpenAI announces GPT-4.5 Turbo, a new model that surpasses GPT-4 Turbo in speed, accuracy and scalability. Learn how GPT-4.5 Turbo can generate natural language or code with a 256k context window and a knowledge cutoff of June 2024.
No, you fricking moron. Internal development has sped up; OpenAI has only gotten skittish about actually releasing things because their AI is getting too powerful, but they've been skittish for a long time. Sam Altman regretted releasing GPT-3.5 'too early' and Illya was reluctant to even release GPT-2. The release of SOTA models from other companies has sped up with GPT-4 going from 1st place to 3rd or 4th since the beginning of this year.
People hyped AI advancement through media like a stock chart that goes up quickly and steep. It’s going to happen in waves, until AI can think up greater things than itself that it can also make or tell us to make. Even then that can taje time and effort to do. Which will slow down the speed of advancement.
does anyone know how to get the cache ID for a given web page? for example, you can get the cached version of https://openai.com/blog/new-models-and-developer-products-announced-at-devday by going to https://cc.bingj.com/cache.aspx?d=5013756218379730&w=6iSDS4Uc_6nn_mbXwGgcadpZkIsAN28f
on google it's as easy as typing cache:$URL, but doing that for https://openai.com/blog/gpt-4-5-turbo/ gets you a 404
>any progress?
none, I didn't keep looking for info
>how did you get the cache ID for that new-models-and-developer-products-announced-at-devday page?
bing shows the cached page in the results. I removed the search statement from the URL and the language info and got that. if you remove either parameters, d or w, it will show an error.
I guess d is the ID and w is some sort of signature. I have no idea how to get the ID for some random page, though.
On most websites the main bing.com/search HTTP response contains the cache d and w parameters in the URL, like
<div class="b_attribution" u="4|5092|4925035076274715|W4SHxnK-vTq_pb6EMVCxL8FrmgXrnUiR" tabindex="0">
But for the gpt-4-5 URL, none shows up in the HTML
There are no AJAX requests done later either
I don't think any public docs exist regarding the generation system either, and from comparing a few of them, there aren't any obvious patterns
On most websites the main bing.com/search HTTP response contains the cache d and w parameters in the URL, like
<div class="b_attribution" u="4|5092|4925035076274715|W4SHxnK-vTq_pb6EMVCxL8FrmgXrnUiR" tabindex="0">
But for the gpt-4-5 URL, none shows up in the HTML
There are no AJAX requests done later either
I don't think any public docs exist regarding the generation system either, and from comparing a few of them, there aren't any obvious patterns
oh, I just checked and realized d isn't an ID, you can change d to 1 or whatever and bing will still show the page.
so, the actual ID is the w parameter.
btw, w is base64-encoded: >>> import base64 >>> base64.b64decode("6iSDS4Uc_6nn_mbXwGgcadpZkIsAN28f"+"==")
b'xea$x83Kx85x1cxeayxe6m|x06x81xc6x9dxa5x99x08xb0x03vxf1'
I guess the | separates the ID from some kind of signature
oh, nice, I assumed since d was required (not supplying a value gives you a 404), the actual value mattered too
Are you sure w is base64 though? I don't think base64 is supposed to have underscores or dashes
You can interpret any string as base64 in the context of b64decode
But someone got a Bing API response to the blog post URL and it looks like it was cached a while ago
https://twitter.com/stimfilled/status/1767617991980589209
So even if we saw the page, the info probably wouldn't be very useful
good, I'm glad competition is finally lighting a fire under their asses
claude has a 200k context window and the public GPT has been dogshit for a long time >we might give you 64k if you're lucky, but probably not
Frick off nobody cares
will i have to pay more is the real question
why would you pay at all
>OpenAI announces GPT-4.5 Turbo, a new model that surpasses GPT-4 Turbo in speed, accuracy and scalability. Learn how GPT-4.5 Turbo can generate natural language or code with a 256k context window and a knowledge cutoff of June 2024.
Which intern fricked up?
Also
>June 2024.
Probably supposed to be January 2024.
4.0 sucks though so turbo is just fast suck?
Yes, sucks really fast.
>Which intern fricked up?
Maybe it was written by GPT itself
How did this happen? How can search engines index the page when the blog url isn't live yet? Did someone oopsie the robots.txt?
What happened to gpt-5
I thought things were accelerating, they seem to be going much slower than previously stated
No, you fricking moron. Internal development has sped up; OpenAI has only gotten skittish about actually releasing things because their AI is getting too powerful, but they've been skittish for a long time. Sam Altman regretted releasing GPT-3.5 'too early' and Illya was reluctant to even release GPT-2. The release of SOTA models from other companies has sped up with GPT-4 going from 1st place to 3rd or 4th since the beginning of this year.
People hyped AI advancement through media like a stock chart that goes up quickly and steep. It’s going to happen in waves, until AI can think up greater things than itself that it can also make or tell us to make. Even then that can taje time and effort to do. Which will slow down the speed of advancement.
>OpenAI blog gpt-4.5 turbo
o shit it's real. on bing too
does anyone know how to get the cache ID for a given web page? for example, you can get the cached version of https://openai.com/blog/new-models-and-developer-products-announced-at-devday by going to https://cc.bingj.com/cache.aspx?d=5013756218379730&w=6iSDS4Uc_6nn_mbXwGgcadpZkIsAN28f
on google it's as easy as typing cache:$URL, but doing that for https://openai.com/blog/gpt-4-5-turbo/ gets you a 404
>https://web.archive.org/web/20240312193449/https://openai.com/blog/gpt-4-5-turbo/
There's a capture on archive.org but it's empty. hmm
any progress? how did you get the cache ID for that new-models-and-developer-products-announced-at-devday page?
>any progress?
none, I didn't keep looking for info
>how did you get the cache ID for that new-models-and-developer-products-announced-at-devday page?
bing shows the cached page in the results. I removed the search statement from the URL and the language info and got that. if you remove either parameters, d or w, it will show an error.
I guess d is the ID and w is some sort of signature. I have no idea how to get the ID for some random page, though.
I looked around for a while but gave up
On most websites the main bing.com/search HTTP response contains the cache d and w parameters in the URL, like
<div class="b_attribution" u="4|5092|4925035076274715|W4SHxnK-vTq_pb6EMVCxL8FrmgXrnUiR" tabindex="0">
But for the gpt-4-5 URL, none shows up in the HTML
There are no AJAX requests done later either
I don't think any public docs exist regarding the generation system either, and from comparing a few of them, there aren't any obvious patterns
oh, I just checked and realized d isn't an ID, you can change d to 1 or whatever and bing will still show the page.
so, the actual ID is the w parameter.
btw, w is base64-encoded:
>>> import base64
>>> base64.b64decode("6iSDS4Uc_6nn_mbXwGgcadpZkIsAN28f"+"==")
b'xea$x83Kx85x1cxeayxe6m|x06x81xc6x9dxa5x99x08xb0x03vxf1'
I guess the | separates the ID from some kind of signature
oh, nice, I assumed since d was required (not supplying a value gives you a 404), the actual value mattered too
Are you sure w is base64 though? I don't think base64 is supposed to have underscores or dashes
You can interpret any string as base64 in the context of b64decode
But someone got a Bing API response to the blog post URL and it looks like it was cached a while ago
https://twitter.com/stimfilled/status/1767617991980589209
So even if we saw the page, the info probably wouldn't be very useful
>GPT-4/Initial release date
>March 14, 2023
Maybe they will release it in 2 days?
good, I'm glad competition is finally lighting a fire under their asses
claude has a 200k context window and the public GPT has been dogshit for a long time
>we might give you 64k if you're lucky, but probably not