Welcome to the open source daily newspaper produced by the OSCHINA editorial department, which is updated every day.
# 2024.5.17
Today's highlights
Developer of open source Diffusion graph model owes US$100 million
Well-known Vincent diagram modeling company Stability AI has been discussing a sale with at least one potential buyer in recent weeks due to financial constraints. It is reported that in the first quarter of 2024, Stability AI’s revenue was less than US$5 million and its losses exceeded US$30 million. The company said in October that it had nearly 200 employees worldwide.
In addition, Stability AI currently owes nearly $100 million in bills to cloud computing providers and other companies . The company has not yet responded to this news.
ZTE joins Alibaba Cloud PolarDB open source community
Alibaba Cloud and ZTE jointly announced an open source database cooperation. ZTE announced that it had joined the PolarDB open source community and was elected as the first council member. In the future, both parties will jointly promote the development of domestic databases based on the PolarDB database open source community and the EBASE database.
NetBSD bans submission of code generated by AI
The NetBSD Foundation has announced a new development policy focusing on code generated by AI technology. The new policy states that code generated by large language models or similar technologies (such as ChatGPT, GitHub Copilot) will be assumed to be contaminated (that is, the copyright is unclear and does not meet NetBSD's licensing goals) and thus cannot be submitted to NetBSD.
Today's observation
social observation
Stability AI owes $100 million and faces sale
The reputation of the OpenAI industry is not so good. One of the main reasons is that it bears the name of OpenAI but is not open source. But as a startup, open source is indeed a double-edged sword, and Stability AI is a typical example.
1. It is difficult to make money with open source.
Stability is very generous in terms of open source, but when it provides products for free, it is difficult to get users to pay even if it is attracted by additional features. Therefore, investment and return are not directly proportional. Although Stable Diffusion 3 and Stable Diffusion 3 Turbo were announced not long ago that they will no longer be open source after the update, it is too late.
2. Computing power is expensive.
Stability has done private calculations. Its revenue last year was US$8 million (although it was higher than US$1.5 million in 2022), and its revenue in the first quarter of this year was less than US$5 million. However, since every Token is backed by investment, it suffered losses of more than US$30 million during the same period, and currently owes nearly US$100 million in bills to cloud computing providers and other companies.
3. It’s dangerous to be on the same track as the giants.
Of course, it’s not scary if you don’t make money. What’s scary is that the giants are still on the same track with you. In the past two days, OpenAI's GPT-4o and Google's Imagen3 model's Vincent graph capabilities have been strengthened, and the Matthew effect has made it increasingly 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.
Of course, you may think that Meta’s Llama model is open sourced, which seems to be going smoothly. But the business models of Meta and Stability AI are different. Meta is not a model, but a social network. In other words, the model is not the final product of Meta, but a middleware. You wouldn’t imagine that Zuckerberg would open source the software code of Facebook and Instagram, just like you wouldn’t think that Google would fully open source its data crawling and search algorithms.
- Weibo Gaofei
There is not much time left for Google. It is recommended that all products be open source.
My overall feeling is that there is not much time left for Google now. I suggest that Google make all its products open source, and then become a leading company in the open source world through open source , leading everyone to work together to deal with multi-modal issues. GPT, against OpenAI. I think it's possible to win.
- Weibo Zhou Hongyi
The largest open source AI community provides $10 million to share GPUs
Hugging Face, the world's largest open source AI community (commonly known as "Hugging Face"), recently announced that it will provide US$10 million in free shared GPUs to help developers create new AI technologies.
Specifically, the purpose of Hugging Face's move this time is to help small developers, researchers, and startups fight against large AI companies and avoid AI progress from falling into "centralization."
Clem Delangue, CEO of Hugging Face, said in an interview with The Verge that he feels lucky to be able to invest in the community. The reason for being able to invest this time is that the company "is already profitable or is on the road to profitability." Some time ago, Hugging Face also raised US$235 million in funding, valuing the company at US$4.5 billion.
- Weibo Scientific Exploration
Media Watch
Zhiyuan releases the Zhiyuan evaluation system, and the evaluation results of “100 models” at home and abroad are released
According to news on May 17, Zhiyuan Research Institute held a large model evaluation conference to launch a scientific, authoritative, fair and open Zhiyuan evaluation system, and released and interpreted more than 140 domestic and foreign open source and commercial closed source languages and multi-modal models. Large model comprehensive capability evaluation results.
This Wisdom Source evaluation examines the seven major abilities of language models from subjective and objective dimensions: simple understanding, knowledge application, reasoning ability, mathematical ability, coding ability, task solving, safety and values; for multi-modal models, Multimodal understanding and generation abilities were mainly assessed.
-NetEase Technology
Big model, let’s go to war!
"The whole northwest of Shanxi is in chaos." At this moment, this sentence in "Bright Sword" is surprisingly suitable to describe competition in the field of large models.
It’s just past the middle of May, and many of the world’s most famous large-scale model players—including OpenAI, Google, Tencent, Alibaba, ByteDance, etc.—suddenly made big moves. Those with the ability to suddenly upgrade comprehensively, and some large-scale models have turned to open source. Some are free to use, and some have experienced price wars due to sharp price drops. The scene is very lively. At first, everyone might just want to join in the fun at the scheduled press conference of OpenAI, but unexpectedly a lot of "big tricks" exploded.
In this explosive arms race, we find that in the field of large models, only the two giants of China and the United States seem to have left their names. Behind the "explosive" competition is the dilemma that the entire field has been unable to find a reasonable business model. Perhaps the giants will go "crazy" because of this.
- Qingcheng Finance
A big model company was exposed as selling out! Many American AI startups laid off 20% of their staff, and celebrity unicorns are looking for "life-saving money"
Overnight, many U.S. generative AI startups were exposed to a crisis of funding shortages:
- Replit, an AI programming unicorn in San Francisco, USA, announced early this morning that it would lay off 20% of its employees, a total of 30 people.
- It was revealed that Reka AI, a large language model startup, may be acquired by Snowflake, a data storage and analysis company, for US$1 billion.
- Stability AI, an AI unicorn company that was facing the crisis of selling out or even going bankrupt before, is seeking a "life-saving money". Its investors include Sean Parker, the first president of Facebook, and Prem Akkaraju, the former CEO of visual effects company Weta Digital.
As technology giants launch and continuously upgrade various free or paid generative AI services, AI startups face fierce competition for customers.
- Smart things
Peak showdown between AI “star” players! Reporters tested the latest Google Gemini and GPT-4o
Recently, OpenAI used a 26-minute online live broadcast to demonstrate the amazing interactive capabilities brought by GPT-4o, bringing a new round of AI competition into the "Her 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, truly realizing multi-modality. Interaction.
The following day, the annual Google I/O developer conference came as scheduled. Google CEO Sundar Pichai announced a series of major updates around its latest generative AI model Gemini, a comprehensive counterattack against OpenAI, including the upgraded Gemini model. The driven AI assistant project Project Astra, the Vincent video model Veo that benchmarks Sora, etc.
The reporter conducted a capability evaluation on the "star" player in the AI industry - Google Gemini 1.5 Pro (1 million tokens), OpenAI's latest upgrade GPT-4o and the previously released GPT-4. The test results are:
-Text test: Google Gemini 1.5 Pro's accuracy and speed completely beat GPT-4o and GPT-4
-Multi-modal test: GPT-4o is superior in details and analysis capabilities
- Science and Technology Innovation Board Daily
Matrix Origins Completes Pre-A Series Financing of Ten Million Dollars
Recently, data intelligence platform technology and service provider Matrix Origin completed a Pre-A round of financing worth US$10 million, led by 21Vianet and followed by Honor Base. Matrix Origins completed an angel+ round of tens of millions of dollars in financing in 2021, led by Bell Ding Capital, followed by Wuyuan Capital and Xianfeng Evergreen.
After this round of financing, Matrix Origin will expand its business to the fields of AI Infra and AI Platform based on the hyper-converged heterogeneous database MatrixOne, and deeply integrate and collaborate with 21Vianet's AIDC business.
It is understood that this round of financing will be used to develop the minimalist, unified, open source and open AI-Native data intelligent global operating system MatrixOS. The system will be powered by the large-scale heterogeneous computing power management and scheduling platform MatrixDC and the hyper-converged heterogeneous data management platform MatrixOne. It is composed of three parts: MatrixGenesis and the AI agent application development platform. The goal is to create an AI Native software platform that links computing power, data, knowledge, models and enterprise applications.
- Lieyun.com
Baidu and the year of big models
In mid-March 2023, Baidu’s large language model Wenxinyiyan launched an invitation test. Competition for large models has been in full swing over the past year, and Baidu has recently launched multiple lightweight large language models. On May 16, Baidu released its financial report for the first quarter of 2024, which showed that revenue was 31.5 billion yuan, a year-on-year increase of 1%, and net profit under non-international accounting standards was 7.011 billion yuan, a year-on-year increase of 22%. According to the information provided by Baidu the day before, the Wenxin model processes an average of 249 billion Tokens per day. Which is more important, open source or closed source, price and comprehensive effect, is the focus of current discussion in the industry. Behind this is actually commercial competition.
- Beijing Business Daily
Today's recommendation
Open source projects
joye61/pic-smaller
https://github.com/joye61/pic-smaller
Pic Smaller (图小小) is an image compression tool developed based on the Vite+React technology stack and supports image compression in four formats: JPEG/PNG/WebP/Gif. TuXiaoXiao performs image compression entirely based on the local browser, without any server-side interaction, and the images will not be uploaded to the remote server.
Daily blog
Expose the online JVM memory overflow problem caused by FileSystem
This article mainly introduces the entire process of analyzing and solving the problem of memory overflow caused by an online memory leak caused by the FileSystem class.
Event comments
People's Daily Online comments on matryoshka-style charging of office software: Only by actively solving the "sets" can we have a future
Member, Super Member, Super Member Pro, AI Member, Grand Member... According to the "Xinhua Daily Telegraph" report, many netizens have recently complained about a well-known office software, claiming that there are arbitrary changes in membership levels and "matryoshka" charging. And other issues. Some interviewees believed that this move was suspected of infringing on consumers' rights to know and choose.
"A small set meal is included in a large set meal, and the small set meal is charged separately." Judging from the complaints of some users, the charging model of this software has caused public outrage and raised questions.
Review
This incident reflects the current problems in the membership charging model in the software industry and the importance of consumer rights protection. At the same time, it also reminds companies to pay more attention to transparency and rationality when formulating charging strategies to avoid infringing on consumers' legitimate rights and interests.
Some companies adopt complex membership charging models when providing software services to achieve the purpose of increasing revenue. Although this approach helps companies increase revenue, it may cause dissatisfaction among consumers. Frequent changes to membership policies, especially sudden restrictions on original membership functions or the addition of additional charges, may lead to a decrease in consumer trust in the company and damage the company's image. If consumers feel that they are being "riddled" or that the charges are not transparent, they may choose to switch to other competitors' products, resulting in the loss of users.
This incident may prompt relevant departments to strengthen supervision and promote the charging model of the entire industry to be more transparent and reasonable, thus promoting the healthy development of the market. When formulating membership charging strategies, companies need to balance revenue growth and user experience, avoid overly complex charging models, and ensure the transparency and rationality of charging to maintain corporate image and market competitiveness.
At the same time, consumers should also be more vigilant about membership agreements and protect their rights and interests through legal means when necessary.
Developer of open source Diffusion graph model owes US$100 million
Prior to this, the Stability AI team was already in a very turbulent state, including the resignation of the founder and CEO , the replacement of the chief technology officer, the loss of a VP of product, a VP of engineering, a VP of R&D, A research director and two large language model leaders. Several key members of the AI research team that developed Stable Diffusion have also resigned from Stability AI .
Review
This incident may have the following impacts on the open source community and AI startups:
-
Money management challenges : Stability AI’s financial woes highlight the importance of money management and profitability for AI startups. Despite investment and market attention, continued capital inflows and an effective profit model are critical to the long-term success of the business.
-
Sustainability of open source projects : As an open source project, Stable Diffusion’s sustainability has been questioned. This could spark discussions about how open source projects maintain funding, especially when relying on community support and commercial investment.
-
Investor views on AI startups : This incident may affect investor views on AI startups, especially those that rely on the latest technology but have not yet achieved profitability. Investors may pay more attention to a company's business model and profit potential.
-
Implications for the open source community : The plight of Stability AI may prompt the open source community to reconsider the way it supports open source projects, especially how to maintain the continued development of projects in the absence of commercial investment.
-
The future of AI startups : This incident may have implications for the future of AI startups, especially those that rely on the latest technology but are not yet profitable. They may need to re-evaluate their business models to achieve sustainable development.
NetBSD bans submission of code generated by AI
The NetBSD Foundation has announced a new development policy focusing on code generated by AI technology. The new policy states that code generated by large language models or similar technologies (such as ChatGPT, GitHub Copilot) will be assumed to be contaminated (that is, the copyright is unclear and does not meet NetBSD's licensing goals) and thus cannot be submitted to NetBSD.
Review
NetBSD's policy emphasizes the concern of open source projects in maintaining code copyright and originality, reflecting the open source community's concerns 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 prompt the open source community to discuss more deeply the moral and ethical issues of AI-generated code.
It may also trigger a re-evaluation of the open source community's reliance on and trust in AI tools. If other open source projects adopt similar policies, it may have an impact on the sustainability of projects that rely on AI tools for development.
Voice of open source
media opinion
Baidu: Search Kingdom, can it rely on AI to defend the city?
Again, the lack of attractiveness of the old business is the main reason why the market has been unwilling to give a reasonable valuation to Baidu (compared to Baidu's advertising business, the market generally only gives 8-10x PE, and ignores the high proportion of net cash/market value ). As long as AI brings meaningful contributions, decent valuation repairs will be slow to start.
Perhaps on the other hand, from the perspective of overall shareholder returns, if Baidu makes full use of the abundant cash lying on its books and increases repurchases or dividends when AI is still growing, it is expected to make the bottom of every round of fluctuations Valuation has reached a new level.
But the window period for waiting for AI to grow is also shortening. Although Baidu's technology is still leading the industry, with the launch of open source models by global giants, other platforms also have great opportunities to catch up. And 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
Is price reduction the way out for Byte AI?
Large byte-based models that do not publish list results and parameter scales rely on price wars to get out of the industry "ingeniously".
Compared with Baidu and Alibaba’s pricing of 0.12 yuan/thousand tokens for 32k models of the same specifications, Byte’s recently unveiled large bean bag model is said to be 99.3% cheaper than the industry, with the pricing reduced to 0.0008 yuan/thousand tokens. This means that users For 0.8 cents, you can process more than 1,500 Chinese characters.
Behind the price war to clear the way, cloud services are taking on more growth responsibilities within Byte, where many businesses are stuck in growth bottlenecks.
-Qianzhan.com
Yao Jinbo, what he lacks is not a digital clone
Standing at the juncture of great changes in the times, Yao Jinbo obviously lacks more than a digital clone. The key issue before him is how to continue to cultivate his internal skills, lead 58.com to occupy the next trend, and reap greater dividends of the times.
AI may be one of the solutions.
- Lieyun.com
User point of view
Zhou Hongyi: There is not much time left for Google. It is recommended that all products be open source
- Viewpoint 1: Lao Zhou has been thinking about Huawei’s open source cause from the bottom of his heart. He is not a Huawei person, but he has Huawei’s soul.
- Viewpoint 2: I suggest that 360 make all its products open source, and then become a leading company in the domestic open source industry through open source, leading everyone to fight against foreign companies.
- Point 3: That’s so right. Lao Zhou can't control Google, but he can control 360. Do unto others, do not impose on others. All 360 products should be open sourced first.
- Viewpoint 4: Google has great roots, so what if AI lags behind?
- Viewpoint 5: This is not the same thing for large companies to say to small companies. Compared with others, my own company is not at the same level as others, but it feels strange to point fingers at others.
People's Daily Online comments on matryoshka-style charging of office software: Only by actively solving the "sets" can we have a future
- Opinion 1: I currently use MS Office for RMB 300 per year. I'm too lazy to waste brainpower with WPS
- Viewpoint 2: Businesses want to make money. How come the software industry does not protect the right to survival of market entities? Water prices can rise, but software can’t collect money, so why can’t we find ways to make money?
- Viewpoint 3: It is okay to increase fees, but the prerequisite for increasing fees is the improvement of product quality and the innovation of product functions, rather than the creation of revenue through innovation in charging methods.
- Viewpoint 4: Free is the most expensive
- Opinion 5: WPS is getting more and more disgusting. It doesn’t allow you to log in and even basic editing is not allowed. The software is getting more and more bloated. I only used it for less than a month before I was deeply disgusted.
ZTE joins Alibaba Cloud PolarDB open source community
- Viewpoint 1: Let’s promote one company. Huawei’s gauss and polardb are a bit scattered. The compatibility of MySQL is of little significance. Go directly to PG.
- Viewpoint 2: Is there anything bad about pg?
iOS 17.5 restores photos that have been deleted for many years, Apple responds "Don't worry about privacy security"
- Point of view 1: With the existence of icloud, you can no longer think about real privacy...
---END---
Finally, you are welcome to scan the QR code to download the "Open Source China APP" and read massive technical reports and sharings from programmers and geeks!
