|
Listen to this story
|
In a fair world, technology should be accessible to all and used for humanity’s betterment. However, there is a growing concern that AI could be monopolised for vested interests by large corporations. This is why open sourcing is paramount.
ChatGPT, the popular chatbot by OpenAI, has been one of the biggest breakthroughs in AI. In just five days since its launch, approximately one million individuals engaged with the bot and it is expected that the number will soon reach the billion mark.
Its popularity escalated to such an extent that it was featured on the cover of Times Magazine. However, ChatGPT is entirely under the control of OpenAI and no external parties are aware of the methodology or the construction behind it. Neither has OpenAI open sourced its GPT3 language model.
So far, only Google has launched a ChatGPT competitor in Bard. Amazon’s chief executive Andy Jassy also told the Financial Times that the e-commerce giant has been working on a ChatGPT competitor for some time now.
Besides, China-based ‘Baidu’ is also working on something similar; however, none of them are—or are likely to be—open source.
Open source ChatGPT competitors
Without doubt, ChatGPT has been one of the biggest breakthroughs in AI. Today, everyone—from individuals to multiple enterprises in diverse fields—is looking to leverage the technology.
However, OpenAI has not released the code for ChatGPT. This makes it difficult for the people outside OpenAI to recreate such models.
To tackle this, Emad Mostaque revealed on Twitter earlier this month that his firm ‘Stability AI’, is working on an open-source version of ChatGPT.
Similarly, Hugging Face—the company that bootstrapped the BigScience collaboration—is also working on an open source ChatGPT rival. The company has also partnered with Amazon Web Services (AWS) for the next iteration of their BLOOM Large Language Model as well as its open-source ChatGPT rival.
Further, AI startup Colossal-AI has also found a way to build their own ChatGPT with less computing resources.
(Source: Colossal AI blog)
To achieve this objective, the organisation has utilised a PyTorch-based implementation that encompasses all three stages, including pre-training, reward model training, and reinforcement learning.
They offer a demo version of the training process that requires only 1.62 GB of GPU memory and can be done on a single consumer-grade GPU, with 10.3x growth on one GPU model capacity.
The importance of open source
Over the years, large corporations have developed many Large Language Models (LLM); however, most of them were not open source.
To counter this, around 1000 researchers from different parts of the globe came together to launch BLOOM, an open source LLM trained in complete transparency.
Similarly, when OpenAI released DALL-E 2, the AI text-to-image generator, the internet fell in love with it. However, neither DALL-E 2 or other similar models such as Midjourney and Imagen by Google were open source.
Enter Stable Diffusion, a similar open source text-to-image model by Stability AI. Founder Emad Mostaque, a firm believer in open source, wants to remove many of the barriers, such as compute and funding for independent and academic researchers to build some of these new AI models.
Now, we are in the age of ChatGPT, which despite being an impressive tool, has its shortcomings. Apart from blurting out hallucinatory responses and convincingly suggesting incorrect answers as correct from time to time, it is also politically biased.
Open sourcing promotes technological agility and would allow the community to collaborate and address the drawbacks of ChatGPT, thus resulting in improvements at a lower cost.
Besides increased collaboration among the community, open sourcing allows users to inspect the source code and understand how the algorithms and models work. This transparency promotes accountability and trust in AI.
“Remember the explosion when DALL-E levelled up via the open source Stable Diffusion? AI will level up yet again when open source ChatGPT hits the streets,” Jeff Garzik, co-founder at Bloq, tweeted.
Open sourcing also encourages innovation by allowing users to build upon existing code and contribute to the development of new features and functionalities.
“I believe that while ChatGPT-like models are probably inaccessible for people to train and develop right now, this will change very shortly with the combination of more label-efficient approaches and open-source initiatives,” Tanishq Mathew Abraham, Founder and CEO at MedARC AI, said.
Besides, OpenAI asserts that it is developing this technology for the betterment of humanity. Therefore, it would be logical for them to open source ChatGPT, allowing the entire community to collaborate and help the technology scale. But, will OpenAI make ChatGPT open source? Unlikely.
