Sabitlenmiş Tweet
Angel 🌼
23.1K posts
Angel 🌼
@Angaisb_
Waiting for AGI Give the world some color Now that you're here, follow me. It's free
Katılım Eylül 2022
1.5K Takip Edilen11.2K Takipçiler
I've said this a few times already, as long as the models aren't open source I don't care if they're 500b or 10T
Can we run it ourselves? No? Then don't excuse the poor performance of your models on the param size
Also, I don't really think 5.5 or Sonnet 4.6 are really much bigger than that
English
@Angaisb_ @Chrisgpt i dont like elon either but i would almost bet money frontier models from like openai are bigger than 500B parameters a 3x increase im willing to give the benefit of doubt might actually be pretty good
for example scoring ~same as these 2 1T param models on AA-II at 500B is good
English
Elon Musk just said that the latest Grok foundation model has finished pre-training.
The model, V9-Medium (1.5T) was trained with additional data from Cursor.
And that he estimates the model will be publicly released in 2-3 weeks.
And that “evals look good”
Is Grok back?
Elon Musk@elonmusk
Grok foundation model V9-Medium (1.5T) has finished training. Evals look good. A lot of Cursor data was added in supplementary training and there is more to come. Fine-tuning is underway and reinforcement learning begins in a few days. 2 to 3 weeks to public release. This will be a major improvement over the 0.5T v8-small that currently serves all Grok production traffic, especially for difficult coding tasks.
English
@pigeon__s @Chrisgpt Closed source labs don't share the parameter count because no one cares, only them will run it
Elon shares it because he has to find an excuse as to why their models are so terribly bad
English
@Angaisb_ @Chrisgpt unlike every other closed source lab at least he shares the parameter size to give an actually objective indicator the new model is 3x larger the only thing im worried about is the main reason to ever use grok was how cheap it was if its 3x more expensive now it better be GOOD
English
JFC. ChatGPT can't even read the content of a single X post because OpenAI is too cheap to pay for the API. Embarrassing. 😡
@OpenAI this needs fixing urgently!
English
@0x506c61746f I actually just asked Codex to get the API docs of the supermarket I go to
English
@Angaisb_ That's a very long time to get the results.
During my time at uni we got them usually the same or next day or after a few days tops.
Is there a reason for such a prolonged delay, or is it just a professor decision?
English
Thank you OpenAI, can we get GPT-5.6 next week now? Hahaha
Angel 🌼@Angaisb_
Hi OpenAI, I have a very important exam on May 23rd so could you please release GPT-5.6 after that date? I would like to be able to study but if you release it before my exam I'll be so excited that I won't be able to Thanks
English
@emollick Agree
Angel 🌼@Angaisb_
As much as I like GPT-5.5 for coding it can be super annoying sometimes when asking it stuff For example, if I ask him "So A is this and B is that, right?" it says "Mostly, but I'd say that C is that, D is other thing, E isn't there, A is this and B is that" Why can't it simply say "Yes, it's right"
English
GPT-5.5 Pro is a very solid fact checker. I can throw entire chapters at it and it will hunt down every key reference accurately. The only real annoyance is that it loves nuance, so returns a lot of “the general idea is right, but you are not taking into account tiny detail X”
English