Welcome to read the Open Source Daily produced by OSCHINA editorial department, which is updated once a day.
# 2024.5.17
Today's key points
Stability AI, a well-known Wenshengtu model company, has discussed the sale with at least one potential buyer in recent weeks due to the shortage of funds. It is reported that in the first quarter of 2024, the revenue of Stability AI will be less than $5 million and the loss will be more than $30 million. The company said last October that it had nearly 200 employees worldwide.
In addition, Stability AI currently owes cloud computing providers and other companies nearly $100 million in bills , the company has not yet responded to this message.
Alibaba Cloud and ZTE jointly announced that they have reached an open source database cooperation. ZTE announced that it has joined the PolarDB open source community and was elected as a member of the first board of directors. In the future, the two sides will jointly promote the development of domestic databases based on the PolarDB open source community and EBASE database.
NetBSD Foundation announce A new development policy has been introduced, focusing on code generated by AI technology. The new policy points out that the code generated by the big language model or similar technologies (such as ChatGPT, GitHub Copilot) will be assumed to be contaminated (that is, the copyright is unknown, and does not meet the licensing objectives of NetBSD), so it cannot be submitted to NetBSD.
Observation today
Social observation
OpenAI is not so popular in the industry. One of the main reasons is that it is not open source but under the name of OpenAI. But as a startup company, open source is really a double-edged sword, and Stability AI is a typical example.
1. Open source is hard to make money
Stability is very generous in open source, but it is difficult for users to pay for the products provided for free, even with the attraction of additional functions. So investment and return are not proportional. Although it was announced not long ago that the Stable Diffusion 3 and the Stable Diffusion 3 Turbo would no longer be open source after being updated, it was too late.
2. Calculation cost is expensive
Stability has privately calculated that its revenue last year was $8 million (although higher than $1.5 million in 2022), and its revenue in the first quarter of this year was less than $5 million. However, because every token is invested, the loss in the same period was up to more than 30 million dollars. At present, cloud computing providers and other companies still owe nearly 100 million dollars in bills.
3. It's dangerous to be on the same track with giants
Of course, it is not terrible not to make money. What is terrible is that the giant still has a track with you. In the past two days, OpenAI's GPT-4o and Google's Imagen3 model's Wensheng map capabilities have been strengthened, and the Matthew effect will make it more and more difficult for Stability AI to make money. In March this year, Inflection AI also chose to commit to Microsoft, which hired most of the company's employees and paid 650 million dollars.
Of course, you may want to open source Meta's Llama model, which seems to be very smooth. However, the business model of Meta is different from that of Stability AI. Meta is not a model, but a social one. In other words, the model is not the final product of Meta, but a middleware. You can't imagine Zuckerberg opening up the software code of Facebook and Ins, just as you can't imagine that Google will open up the data capture and search algorithms completely.
-Weibo Goofy
My overall feeling is that there is not much time left for Google, I suggest Google open source all its products, and then become the leader in the open source world again through open source , leading everyone to work together to deal with GPT and OpenAI in multimodality. I think it is possible to win.
-Weibo Zhou Hongyi
Hugging Face, the world's largest open source AI community, recently announced that it will provide 10 million dollars of free shared GPUs to help developers create new AI technologies.
Specifically, the purpose of Hugging Face's move is to help small developers, researchers and start-ups fight against large AI companies and avoid AI progress falling into "centralization".
Clem Delague, CEO of Hugging Face, said in an interview with The Verge that he was very lucky to be able to invest in the community. The reason why he was able to invest this time was that the company "has made profits or is on the way to making profits". Some time ago, Hugging Face also raised $235 million, and the company's valuation reached $4.5 billion.
-Weibo Scientific exploration
Media observation
On May 17, it was reported that Zhiyuan Research Institute held a big model evaluation conference, launched a scientific, authoritative, fair and open evaluation system for Zhiyuan, and released and interpreted the results of comprehensive ability evaluation of more than 140 open source and commercial closed source languages and multimodal big models at home and abroad.
In this intelligence assessment, the seven abilities of language model, namely, simple understanding, knowledge application, reasoning ability, mathematical ability, code ability, task solving, safety and values, were examined from the subjective and objective dimensions; For multimodal models, the ability of multimodal understanding and generation is mainly evaluated.
- Netease Technology
"The whole northwest of Shanxi is in a mess." At this moment, the phrase in the "Sword" used to describe the competition in the field of large models seems unexpectedly appropriate.
Just half of May, the world's most famous big model players - including OpenAI, Google, Tencent, Ali, Byte Jump, etc. - suddenly made a big move. Those who were able to suddenly upgrade in an all-round way, those who were able to convert big models to open source and free use, and those whose prices went down sharply and started a price war were very lively. At the beginning, we may just want to join in the event scheduled by OpenAI, but we didn't expect to have a bunch of "big moves".
In this explosive arms race, we found that only the giants of China and the United States remained in the big model arena. Behind the "explosive" competition is the dilemma that the whole field has been unable to find a reasonable business model. Perhaps the giants will become "crazy".
-Qingcheng Finance
Overnight, many American generative AI start-ups were exposed to the crisis of capital shortage:
- Replit, an AI programming unicorn in San Francisco, USA, announced 20% layoffs early this morning, totaling 30 people.
- Big language model founder Reka AI is exposed to be acquired by data storage and analysis company Snowflake for US $1 billion.
- Stability AI, an AI unicorn enterprise that was faced with the danger of selling itself or even going bankrupt, is striving for a "life-saving money". Investors include Sean Parker, the first president of Facebook, and Prem Akkaraju, the former CEO of Weta Digital, a visual effects company.
As technology giants have launched and continuously upgraded various free or paid generative AI services, AI start-ups are facing fierce competition for customers.
-Intellectual things
Recently, OpenAI used a 26 minute live online broadcast to show the amazing interaction ability brought by GPT-4o, bringing a new round of AI hegemony into the "His Era". The "o" of GPT-4o stands for "omni", which means "omnipotent". This model can realize seamless text, video and audio input, and generate corresponding modal output, which truly realizes multimodal interaction.
The next day, the annual Google I/O Developer Conference came as scheduled. Google CEO Sundar Pichai announced a series of major updates around his latest generative AI model, Gemini, to comprehensively counter OpenAI, including the AI assistant project Project Astra driven by the upgraded Gemini model, and the cultural video model Veo benchmarking Sora.
The reporter conducted an ability evaluation on the "star" players in the AI world - Google Gemini 1.5 Pro (1 million tokens), OpenAI's newly upgraded GPT-4o and the previously released GPT-4. The test results are as follows:
-Text test: accuracy and speed of Google Gemini 1.5 Pro completely beat GPT-4o and GPT-4
-Multimodal testing: GPT-4o is better at details and analysis
-Science and Technology Innovation Board Daily
Recently, Matrix Origin, a data intelligence platform technology and service provider, completed a $10 million Pre-A round of financing, led by 21Vianet and followed by Honour Base. Matrix Origin completed angel+round financing of tens of millions of dollars in 2021, led by Zhongding Capital, followed by Wuyuan Capital and Qianfeng Evergreen.
After this round of financing, Matrix Origin will expand its business to AI Infra and AI Platform fields on the basis of hyper integrated heterogeneous database MatrixOne, and deeply integrate and cooperate with 21Vianet's AIDC business.
It is understood that this round of financing will be used to develop the extremely simple, unified, open source and open AI Live data intelligence global operating system MatrixOS, which will be composed of a large-scale heterogeneous computing management scheduling platform MatrixDC, a super integrated heterogeneous data management platform MatrixOne and an AI agent application development platform MatrixGenesis. The goal is to create a link between computing power, data, knowledge AI Native software platform for model and enterprise applications.
-Cloud hunting net
In the middle of March 2023, Baidu's big language model Wenxin Yiyan launched the test invitation. In the past year, the competition for big models has been in full swing, and Baidu has recently launched several lightweight big language models. On May 16, Baidu released its financial report for the first quarter of 2024, showing a revenue of 31.5 billion yuan, up 1% year on year, and a net profit of 7.011 billion yuan under non international accounting standards, up 22% year on year. According to the information provided by Baidu the day before, Wenxin big model processes 249 billion Tokens on a daily basis. Which is more important, open source or closed source, price and comprehensive effect, is the focus of the current discussion in the industry, which is actually a commercial competition.
-Beijing Business Daily
Recommended today
Open source project
https://github.com/joye61/pic-smaller
Pic Smaller is an image compression tool, which is developed based on the Vite+React technology stack and supports image compression in JPEG/PNG/WebP/Gif formats. The image is compressed locally based on the browser. Without any server interaction, the image will not be uploaded to the remote server.
A daily blog
This article mainly introduces the whole process of analyzing and solving the problem of memory overflow caused by an online memory leak caused by the FileSystem class.
Event comments
Members, super members, super member Pro, AI members, big members... According to the Xinhua Daily Telegraph, several netizens recently complained about a well-known office software, saying that it had problems such as modifying the membership level at will and charging "dolls". Some respondents believe that this is suspected of infringing consumers' right to know and choice.
"There are small packages in the big packages, and the small packages are charged separately". From the complaints of some users, the charging mode of the software has offended the public and raised questions.
comment
This event reflects the problems existing in the current software industry in the membership fee model, as well as the importance of consumer protection. At the same time, it also reminds enterprises to pay more attention to transparency and rationality when formulating charging strategies to avoid infringing on the legitimate rights and interests of consumers.
Some enterprises adopt a complex membership charging model when providing software services, so as to achieve the purpose of increasing revenue. Although this practice helps enterprises to increase their income, it may cause consumer dissatisfaction. Frequent changes in membership policies, especially sudden restrictions on the functions of existing members or additional charges, may lead to a decline in consumers' trust in the enterprise and damage the image of the enterprise. If consumers feel that they are "routine" or that the charges are opaque, they may choose to switch to other competitors' products, leading to the loss of users.
This event may prompt relevant departments to strengthen supervision and promote the charging mode of the whole industry to be more transparent and reasonable, thus promoting the healthy development of the market. When formulating member charging strategies, enterprises need to balance revenue growth and user experience, avoid overly complex charging models, and ensure the transparency and rationality of charges to maintain corporate image and market competitiveness.
At the same time, consumers should also be more alert to the membership agreement and safeguard their rights and interests through legal channels when necessary.
Before that, The stability AI team has been in a very turbulent state, including Founder and CEO resign , replaced the chief technology officer, lost a vice president of product, a vice president of engineering, a vice president of research and development, a research director and two heads of large-scale language models. Several key members of the AI research team who have developed Stable Diffusion have also long been from Stability AI quit 。
comment
This event may have the following impacts on the open source community and AI start-ups:
-
Challenges of fund management : The financial predicament of Stability AI highlights the importance of capital management and profitability for AI start-ups. Despite investment and market attention, continuous capital inflow and effective profit model are crucial to the long-term success of enterprises.
-
Sustainability of open source projects : As an open source project, the sustainability of Stable Diffusion is questioned. This may lead to discussions on how to maintain financial support for open source projects, especially when relying on community support and commercial investment.
-
Investors' views on AI start-ups : This event may affect investors' views on AI start-ups, especially those that rely on the latest technology but have not yet made profits. Investors may pay more attention to the business model and profit potential of enterprises.
-
Enlightenment to the open source community : The dilemma of Stability AI may prompt the open source community to reconsider its support for open source projects, especially how to maintain the sustainable development of projects in the absence of commercial investment.
-
Future of AI start-ups : This event may have an impact on the future of AI start-ups, especially those that rely on the latest technology but have not yet made a profit. They may need to reassess their business models to achieve sustainable development.
NetBSD Foundation announce A new development policy has been introduced, focusing on code generated by AI technology. The new policy points out that the code generated by the big language model or similar technologies (such as ChatGPT, GitHub Copilot) will be assumed to be contaminated (that is, the copyright is unknown, and does not meet the licensing objectives of NetBSD), so it cannot be submitted to NetBSD.
comment
NetBSD's policy emphasizes the concern of open source projects in maintaining code copyright and originality, reflecting the concern of the open source community about AI generated code. NetBSD's decision may become a reference for other open source projects when dealing with AI generated code, affect the norms and standards of the entire open source community, and urge the open source community to discuss the moral and ethical issues of AI generated code more deeply.
It may also lead to a reassessment of the open source community's reliance on and trust in AI tools. If other open source projects adopt similar policies, it may affect the sustainability of projects that rely on AI tools for development.
Voice of Open Source
Media opinion
Again, the lack of attractiveness of the old business is the main reason why the market has been reluctant to give Baidu a reasonable valuation (compared with Baidu's advertising business, the market generally only gives 8-10x PE, and ignores the high proportion of net cash/market value). Only AI brings meaningful contribution, and decent valuation repair will start slowly.
Perhaps on the other hand, from the perspective of the overall return of shareholders, if Baidu makes full use of the abundant cash lying on the ledger while AI is still growing, and increases the strength of repurchase or dividend distribution, it is expected to make the bottom valuation of each round of fluctuations rise to a higher level.
But the window period waiting for AI to grow up is also shortening. Although Baidu's technology is still leading the industry, with the launch of the global giant's open source model, other platforms also have great opportunities to catch up. If Baidu's leading gap does not further expand to the extent that users can clearly perceive it, other giants are likely to make up for the technical gap through ecological advantages.
- Dolphin investment research
The big byte system model, which does not publish the results of the list and the size of the parameters, has been "ingeniously" driven by the price war.
Compared with the pricing of Baidu and Alibaba 32k models of the same specification at 0.12 yuan/thousand tokens, recently, the price of the big model of beanbag, which was first unveiled in bytes, was claimed to be 99.3% cheaper than the industry, and the price was reduced to 0.0008 yuan/thousand tokens, which means that users can process more than 1500 Chinese characters for 0.8 percent.
Even behind the price war, cloud services are taking on more growth responsibilities in the bytes where multiple businesses are trapped under the growth bottleneck.
- Forward-looking network
Standing on the node of the great changes of the times, Yao Jinbo obviously does not lack a digital part. The key issue before him is how to continue to cultivate internal skills, lead the 58 city to occupy the next outlet, and enjoy greater dividends of the times.
AI may be one of the solutions.
- Cloud hunting net
User perspective
- Viewpoint 1: Old Zhou is deeply concerned about Huawei's great cause of open source. He is not a Huawei person, but has Huawei's soul.
- Viewpoint 2: I suggest that 360 open source all products, and then become the leading enterprise in the domestic open source industry through open source, leading everyone to fight against foreign enterprises
- Viewpoint 3: That's too right. Old Zhou can't control Google, but he can control 360. Do not do to others what you do not want. All 360 products should be opened first.
- Viewpoint 4: Google is big and AI is backward.
- Viewpoint 5: Big companies say the same thing to small companies. Compared with others, their own enterprises are not at the same level. It's strange to point out to others.
- Viewpoint 1: I use MS Office for 300 yuan every year. I'm too lazy to waste my brain with WPS
- Viewpoint 2: Enterprises want to make money. How come the software industry can't guarantee the survival right of market subjects? If the water price can rise and the software can't make money, can't we find a way to make money?
- Viewpoint 3: It is OK to raise the charge, but the premise of raising the charge is the improvement of product quality and the innovation of product functions, rather than the innovation of charging means to achieve income generation.
- Viewpoint 4: free is the most expensive
- Viewpoint 5: WPS is getting more and more disgusting. It doesn't log in, and even basic editors are not allowed to use it. The software is getting more and more bloated. I used to use it less than a month ago, and I was disgusted.
- Viewpoint 1: It's better to promote one company. Huawei gauss and polarb are a little scattered. MySQL compatibility is of little significance. It's directly related to PG
- Viewpoint 2: What's wrong with pg
- Viewpoint 1: With the existence of icloud, you can't think about real privacy
---END---
Finally, welcome to scan the code to download "Open Source China APP", read massive technical reports, and share with programmers and geeks!
