Has OpenAI Lost the "Open Competition"?

2023-05-18

Will OpenAI return to its original "openness" and propose a plan to release open-source models? If the reports are credible, OpenAI is preparing to release its open-source AI models. However, the new models are likely to be unable to compete with their flagship GPT. But why is OpenAI reconsidering "openness"? Given the closed nature of OpenAI's flagship GPT, announcing the creation of an open-source model would allow them to enter the open-source competition, which they have been avoiding. Open sourcing allows for continuous innovation by allowing people to work with the code and provide feedback, harnessing collective intelligence and open contributors to unleash their potential, resulting in exponential technological growth. Open-source encourages mutual benefits for both manufacturers and users. To continue keeping secrets, OpenAI has been cautious about releasing a new model that may be unrelated to their GPT model. However, this is not the first time OpenAI has attempted to engage with the open-source community after pursuing profit. Last month, the company announced a bug bounty program, inviting researchers to report vulnerabilities, bugs, or security flaws in their systems and offering rewards ranging from $200 to $20,000. Is open-source the right path? Although OpenAI has been closed-source for many years, many companies, such as Google (PaLM), Meta (LLaMA), and Hugging Face, have kept their models open-source. Companies adopt an open-source approach believing that they can receive donations from developers. Meta's LLaMA has become a topic of interest for researchers due to open-sourcing their models. Their latest models are also open to everyone depending on the circumstances. With the code leaked on GitHub, people can access LLM at GPT levels for free. With models like Alpaca and Vicuna released by Stanford University, as well as collaborations with UC Berkeley and UC San Diego, LLM has become popularized. Through these open-source models, developers can have public access to LLM to further enhance its capabilities. Google's PaLM is also following suit as it is open-source. Benefits and threats of open-source Recently, Google discussed the significant threat open-source poses to the company in an internal document. It emphasized how the modern internet is "built on open-source" and compared recent open-source AI developments to the internet. The open-source community has played a significant role in the internet technology revolution, which cannot be replicated by any institution. According to an internal document shared by Google senior software engineer Luke Sernau, open-source technology has been "quietly advancing" and surpassing OpenAI and Google in the AI competition. Some argue that the open-source community poses a threat to Google and large tech companies as people can build competing models at a faster pace and lower cost. It may also be detrimental when people are not ready to pay for high-quality technology freely provided by the open-source community. OpenAI's openness and closedness OpenAI has faced criticism for keeping its models closed-source. The company initially started as a nonprofit research organization dedicated to developing digital intelligence but later transformed into a profit-driven closed-source model institution. When asked about this change, co-founder Ilya Sutskever stated that "open-source doesn't make sense" if AI or AGI becomes incredibly powerful. He also predicted that in the coming years, everyone will realize that "open-source AI is not wise." Elon Musk, one of the original founding members of OpenAI, expressed concerns about the company. Initially, the company was established as a counterbalance to Google's open-source dominance, but it has now become a closed-source "maximizing profit" company controlled by Microsoft, which is not what he hoped for. Today, with OpenAI planning to release an open-source model, the company seems to be returning to its original goal to cater to the needs of the open-source community. Ironically, OpenAI uses open-source to train and expand their GPT models, but their code remains closed to the community. In March of this year, OpenAI faced a security vulnerability due to the exposure of user search history, attributed to a vulnerability in the Redis client library, indicating that the company is still using open-source code. OpenAI's development path is now challenging and full of mysteries. On one hand, they are monetizing through special subscriptions like ChatGPT Plus and ChatGPT Business, and on the other hand, they plan to open-source their code. They recently announced over 70 ChatGPT plugins and web browsing capabilities, where ChatGPT will know when and how to browse recent topics and events on the internet, which will be rolled out to all ChatGPT Plus users in the coming days. Looking at how OpenAI strives to maintain its leading position in the AI competition, the open model format may be the only factor hindering its progress. What their open-source models will offer remains to be seen.

RECENT AI NEWS

RECENT AI TOOLS