OpenAI, once a beacon of hope for the open-source community, has evolved into a closed-source, profit-driven behemoth, leaving many to question its trustworthiness. Despite this transformation, LunaSec has been making great use of their products, including the recent release of ChatGPT, which has ignited a chatbot arms race among tech giants. However, due to OpenAI's departure from its original vision and the influence of multi-billion-dollar deals like Microsoft's massive investment, we hope to eventually transition to an open-source alternative.
The Transformation of OpenAI
In the beginning, OpenAI was established as a nonprofit research organization committed to advancing digital intelligence for the benefit of humanity. The founding members, including Sam Altman, Elon Musk, and Peter Thiel, promised to focus on positive human impact and transparency. Researchers were encouraged to share their findings, code, and patents with the world.
Fast-forward to today, and OpenAI has pivoted to a profit-driven model. Critics, including co-founder Musk, argue that the company is now focused on speed and profit instead of positive human impact.
Elon Musk's Worst Nightmare: Close Source, Corporate Control of AI
Musk has long considered AI to be "fundamental existential risk for human civilization". He has spent years calling for strict government regulation and caution around AI research, and received some interest from regulators. One of Musk's main suggestions was to learn as much as possible to better understand the problem.
OpenAI was created as an open source (which is why I named it “Open” AI), non-profit company to serve as a counterweight to Google, but now it has become a closed source, maximum-profit company effectively controlled by Microsoft. Not what I intended at all.— Elon Musk (@elonmusk) February 17, 2023
Originally when Musk co-founded OpenAI he envisioned it as a nonprofit organization that would share its research and technology. Musk lost any hope of control over the company in 2018, in a power struggle with the other co-founders. This recent report explores the complex reasons for the breakup. Following Musk's departure, OpenAI became increasingly entangled with Microsoft and other corporate partners, trading transparency and openness for financial resources.
Vendor lock is really inconvenient
Mission statements and ideologies aside, there are a lot of day-to-day struggles with using something controlled by just one company.
API Outages, Reliability Issues, and Unannounced Changes
OpenAI's ChatGPT has experienced several API outages and reliability issues, leaving developers and businesses in the lurch. Furthermore, subtle changes to models behind the scenes have the potential to break existing use cases without warning. Put simply, developers do not have control of a huge and often critical component of their own software.
Inability to Train Additional Data
The inability to train additional data into ChatGPT limits developers and businesses from customizing the AI to their specific needs. There are a thousand companies out there that would love to train an LLM as powerful as ChatGPT on their own data sets, us included. The fine-tuning API that OpenAI does offer isn't well suited to additional pre-training. This means that what the model "knows" about is up to OpenAI.
This stifles innovation and increases dependence on OpenAI's proprietary technology.
Alternatives: A Glimmer of Hope
In spite of the concerns surrounding OpenAI, there are alternatives. If you're curious to see what more open competitors are like, trying them out isn't as difficult as you might think.
These next two models have restrictive, non-commercial licenses.
Another non-commercial LLM, ChatGLM, is even easier to try. You can launch ChatGLM for free on Google Colab with just one click. With far fewer neurons than ChatGPT, a smaller training dataset, and presumably less advanced technology, it falls noticeably short, at least in English.
Open-source alternatives: The open-source community is working on alternatives to ChatGPT, like the ones listed on this awesome totally open chatgpt list. As shown by the categories in the list, some of these models can be trained, customized, and self-hosted, with permissive licenses that allow for more control and customization without vendor lock-in.
OpenAI's transformation from an open-source champion to a closed-source, profit-driven company is a cautionary tale for the AI industry. Although the company has made significant strides in AI development, its increasing secrecy, lack of transparency, and limited customization options have alienated the very community it once aimed to serve. The future of AI depends on stakeholders, including researchers, developers, and businesses, actively working to prevent monopolization and ensure transparency, openness, and collaboration remain at the forefront of AI development.