Click the card below to follow the " CVer " official account
AI/CV heavy dry goods, delivered in the first time
Click to enter —>【Computer Vision】WeChat Technology Exchange Group
Meng Chen Feng Se from Aufei Temple
Reprinted from: Qubit (QbitAI)
Musk entered the AIGC war and urgently snapped up about 10,000 GPUs !
What concept?
Tesla's strongest supercomputer Dojo doesn't take advantage of so much.
Dojo’s data released after the upgrade in 2022 is 7,360 A100s. Although it’s unclear what model they bought this time, the quantity alone is scary enough.
According to Business Insider, the 10,000 graphics cards were purchased for Twitter .
The whistleblower said that the Twitter version of GPT is already in preparation, and that Twitter itself has a huge amount of data, which has certain advantages.
Wait a moment? Didn’t Musk sign an open letter a while ago, calling for a 6-month suspension of AI models that are more powerful than GPT-4?
Is it really in line with what netizens said, "You guys stop and wait for me to catch up"?
For Musk's big move this time, some netizens think that hoarding GPU itself is a good investment.
Focus on large models and poach two DeepMind engineers
According to people familiar with the matter, Musk's artificial intelligence project is still in the early stages, and buying so much additional computing power is enough to show his determination.
As mentioned at the beginning, these 10,000 GPUs are currently mainly involved in the development of a large language model——
Of all Musk's companies, Twitter is probably the best suited for this, because it can provide a lot of training data.
According to a netizen, Twitter, which was established in 2006, has accumulated a database of nearly 20 years, which is an excellent weapon for Twitter to enter the LLM field .
In addition to spending a lot of money to buy GPUs to build large models, Musk is also vigorously robbing people.
Currently, two DeepMind engineers have been successfully hired:
One is Igor Babuschkin, who is engaged in AI system research and is a senior scientific research engineer of DeepMind. He has worked in DeepMind for about 5 years, and jumped to OpenAI for a year and a half in the middle.
The other is Manuel Kroiss, a software engineer who has worked for DeepMind for six years and was a solution engineer at Google before that. After Musk recruited him, he arranged for him to be the senior director of software engineering at Twitter.
According to BusinessInsider, in fact, since February , Musk has frequently "closed" with people in the AI field, including Igor Babuschkin.
As for where this large model works, we don't know for the time being, but there are two guesses:
One is to improve search.
Because Lao Ma complained about this feature, he even hired famous iPhone hacker George Hotz for a three-month fix.
However, it seems that he and Musk's "discord" not only rejected Tesla's offer invitation, but also stayed on Twitter for three months and left after only one month.
In addition to search, another use may be advertising marketing .
Advertising is the backbone of social media. It’s a pity that Musk introduced a lot of new policies after the acquisition of Twitter. Many advertisers directly cut spending or simply stopped cooperating. As a result, Twitter’s revenue in December last year alone fell by about 40% compared to the same period last year.
And if there is a large model for generative AI creation, whether it is designing materials or providing ideas, it may improve the efficiency and benefits of Twitter in advertising and marketing.
emmm... Speaking of Twitter's current (always) not-so-optimistic financial situation, people familiar with the matter also said this time that Musk may have spent tens of millions of dollars buying GPUs this time .
After all, it is very likely to buy Nvidia, because it occupies 95% of the market share, and a product for AI large models costs about $10,000.
So, the reason why Twitter has been short of money has been found? (manual dog head)
Failed to take over OpenAI 5 years ago, Musk's counterattack came
Since the birth of ChatGPT, Musk has expressed his views on OpenAI.
On the page of the open letter that suspended AI development for 6 months, his name is still listed.
And often expressed dissatisfaction with the status quo of the company he co-founded.
How did I donate 100 million US dollars to establish a non-profit organization and become a 30 billion for-profit company? If it's legal, why isn't everyone else doing it?
As the world continues to pay attention to ChatGPT and OpenAI, more inside stories about Musk and OpenAI have also been unearthed.
According to Semafor, Musk tried to take over the CEO position in 2018, but failed.
At the time, Musk believed that OpenAI had lagged far behind Google technically and needed to make major changes.
After a bitter internal power struggle, the board voted down Musk's meaning. In response, Musk quit OpenAI entirely and poached Andrej Kaparthy to become Tesla's head of AI.
Sam Altman, the current CEO of OpenAI, also said in an interview that "I have always regarded Musk as a hero, although he has been a jerk on Twitter recently (being a jerk)".
In addition to making offensive remarks, Musk has also taken practical actions against OpenAI.
In the past, OpenAI could use Twitter data for AI training, but on the fifth day after ChatGPT was released, Musk ordered the suspension of this permission.
Three months later, Musk announced the establishment of his own AI laboratory and chose BasedAI as the team name.
Now, Musk has mysteriously posted a new opinion on AI, perhaps revealing the development direction of the Twitter version of GPT:
A curious AI might be the best.
The cat might be against it (curiosity killed the cat), but George the Curious Monkey (from the popular primer) is in full support.
Reference link:
[1]https://www.businessinsider.com/elon-musk-twitter-investment-generative-ai-project-2023-4?op=1
[2]https://twitter.com/artificialguybr/ status/1645852415600611333
Click to enter —>【Computer Vision】WeChat Technology Exchange Group
The latest CVPP 2023 papers and code download
Background reply: CVPR2023, you can download the collection of CVPR 2023 papers and code open source papers
Background reply: Transformer review, you can download the latest 3 Transformer review PDFs
目标检测和Transformer交流群成立
扫描下方二维码，或者添加微信：CVer333，即可添加CVer小助手微信，便可申请加入CVer-目标检测或者Transformer 微信交流群。另外其他垂直方向已涵盖：目标检测、图像分割、目标跟踪、人脸检测&识别、OCR、姿态估计、超分辨率、SLAM、医疗影像、Re-ID、GAN、NAS、深度估计、自动驾驶、强化学习、车道线检测、模型剪枝&压缩、去噪、去雾、去雨、风格迁移、遥感图像、行为识别、视频理解、图像融合、图像检索、论文投稿&交流、PyTorch、TensorFlow和Transformer等。
一定要备注：研究方向+地点+学校/公司+昵称（如目标检测或者Transformer+上海+上交+卡卡），根据格式备注，可更快被通过且邀请进群
▲扫码或加微信号: CVer333，进交流群
CVer计算机视觉（知识星球）来了！想要了解最新最快最好的CV/DL/AI论文速递、优质实战项目、AI行业前沿、从入门到精通学习教程等资料，欢迎扫描下方二维码，加入CVer计算机视觉，已汇集数千人！
▲扫码进星球
▲点击上方卡片，关注CVer公众号
整理不易，请点赞和在看