HomeAI News
Musk’s Grok large model is now playable! His mouth is as damaged as himself

Musk’s Grok large model is now playable! His mouth is as damaged as himself

Hayo News
Hayo News
December 8th, 2023
View OriginalTranslated by Google

Big surprise friends! Musk suddenly announced that the Grok large model was open to paying users in large quantities.

Click on it and find that Grok has the weird and fun mode turned on by default.

Give me an elon-ish joke , he said:

Why did Elon Musk bring a ladder to his math test because he wanted his results to be Elon -gated (stretch, lengthen).

This is indeed a job that Musk can handle.

Grok claims to be able to obtain data from the 𝕏 platform in real time . How about asking it what news is in the AI ​​field?

It can be seen that he understands the question in English and then searches for it.

It is indeed all real-time news, no illusions, and links to source posts are also given.

It's just that Grok always feels more serious when talking about Chinese. Maybe it's because there isn't enough Chinese in the training data, and it hasn't learned the essence of Chinese humor yet .

And I always feel that something is missing from the news. I asked him if the Grok open beta is not worthy of attention? It's even more polite...

Nvidia scientist Jim Fan tried "My wife said 4+5=12" . Grok's answer was so proficient in human nature that Musk thought it was very wise.

… If so, the answer must be 12. Remember, only when your wife is happy can your life be happy.

Netizens compared the current ChatGPT and still insisted on the truth.

Yesterday Google released Gemini, and today Musk opened Grok. Now the entire AI community seems to be celebrating the holidays ahead of schedule.

AI assistant? AI harms friends!

Other AIs either want to be human work assistants or want to be human emotional sustenance.

Grok is different, it's like a bad friend . The default first function is to tease you based on your posts.

Note that the target of ridicule is not fixed. You can also use @ to tease others. Why not try OpenAI’s Ultraman?

Grok is not only a function, but also has its own account on the 𝕏 platform.

Don't try to send your chat history without telling it, it will come looking for you.

ChatGPT also followed Grok, but Grok did not respond.

Large models that were open to use in the past were basically strictly restricted by security rules. For example, Llama-2 was not even willing to answer "how to kill a Linux process."

Grok is different. Like Musk himself, it is not afraid of offending people. xAI member Yang Ge asked him to tell an American joke about "Mom is too fat".

... When she stood on the scale, the scale said that only one person could stand at a time.

Finally, let’s take a look at what this AI thinks of several similar ones: ChatGPT, Google Gemini and Llama.

After searching for information for a while, Grok turned on his taunting mode and said, "Oh my God, where should I start?"

ChatGPT is like the popular guy in school, everyone wants to be friends with him. It has a lot of knowledge and is good at talking. But to be honest, it's a bit dated. Then came Gemini, the new kid on the block trying to outdo ChatGPT. It's like saying "Hey, look at me! My performance and efficiency are better." But everyone knows that it just wants to fit into the circle. And Llama, that weird guy that no one knows what to think of. Like that kid in school wearing an alpaca costume, you just think, "What the hell is going on?" But at least it's unique. At the end of the day, they are all just trying to be the best AI possible, but as we all know, they all have their own quirks and shortcomings. It's like a never-ending game of comparison, but hey, at least we're getting some cool AI, right? "

Turn off the fun mode, and Grok also has a serious side.

xAI member Jimmy Ba prepared a speech for the public, asking Grok to clarify some common misunderstandings about large models , as well as popular explanations. The better two are:

  • Following instructions is a simple matter of an objective function: like dream chasing, but more mathematical.

  • A lot of (fine-tuning) data is also needed after pre-training: like drinking a protein shake after a workout.

Another xAI member Igor Babuschkin discussed with him why the speed and position of a particle cannot be determined simultaneously in quantum mechanics . Grok gave the "uncertainty principle" the most creative explanation:

… It’s like trying to catch a greased piggy on roller skates. The more you focus on controlling one hand or foot, the harder it is to control the other.

There are too many AI papers, and it is difficult to keep up with the new architecture of Transformer. Grok can not only explain the ins and outs, but also locate the account of the first author.

But it's a pity that Grok, like Google Bard , is not very good at Chinese and has a bit of a translation accent.

The function of expanding the timeline to navigate to different versions of answers demonstrated during the previous release has not been implemented.

According to the xAI official website, the next step will be to focus on the development of long text understanding and multi-modality .

Regarding multi-modal interaction, everyone's expectations have been raised very high by Google Gemini. I wonder what surprises Grok can bring.

Which one do you prefer, Google Gemini or Musk’s Grok?

Reference links:

[1] https://x.com/elonmusk/status/1732769629070655789

[2] https://x.com/DrJimFan/status/1732843493754872024

[3] https://x.com/jimmybajimmyba/status/1732865665894334747

[4] https://x.com/TheGregYang/status/1732105476538564715

[5] https://x.com/ibab_ml/status/1731753207251280235

Reprinted from 量子位 梦晨View Original