OpenAI’s GPT-3 is the talk of the town, and the media is giving it all the attention. Many analysts are even comparing it to AGI because of its practical applicability. Initially disclosed in a research paper in May, GPT-3 is the next version of GPT-2 and is 100x larger than it. It is far more competent than its forerunner due to the number of parameters it is trained on, which is 175 billion for GPT-3 versus 1.5 billion for GPT-2.
After the successful launch of GPT-3, other AI companies seem to have been overshadowed. One such company which is also a close competitor of OpenAI is DeepMind. The big question is whether OpenAI has now surpassed DeepMind, which rose to fame in 2016 when it produced AlphaGo software that learned how to play the board game Go and grew better than any human player. Musk co-founded the OpenAI research lab in San Francisco in 2015, one year after Google acquired DeepMind.
While both the companies are leading the AI charge and trying to move towards artificial general intelligence (AGI), it has called for a tech war.
Sign up for your weekly dose of what's up in emerging technology.
The Commercial Aspect
It is the usefulness of GPT-3 which can be commercialised, and OpenAI has launched APIs for a commercial subscription. OpenAI stands the chance of churning profit with their APIs. Open AI also benefits from Microsoft’s collaboration in training the language model using its supercomputer. Microsoft can further help the company in finding business clients, given its incredibly rich enterprise presence.
DeepMind exists under Google’s umbrella, hence a little more skewed towards Google. Ever since Alphabet acquired DeepMind, it has been reporting losses, but Google back up is going to keep it fine. Also, it does not have to prioritise on building a product that could be commercialised readily. Instead, DeepMind has been focusing on proof-of-concept where its agents have beaten humans at very complex games using reinforcement learning techniques, including AlphaGo.
DeepMind has also moved away from its focus from just gaming (which has been their forte) into health research projects for Google. This way, the company is working on building more commercially-applications AI by using a state-of-the-art baseline for Deep Reinforcement Learning algorithms. DeepMind is expanding its focus from creating AI agents which can compete in games to making AI agents which can have real-world impact, specifically in areas such as biology.
The AI Value Of Both Systems
GPT-3 can be used by businesses in actually finishing human tasks, making it the most coherent language model. People have used it to write articles, songs, stories, essays, technical manuals and more. The system may also have the ability to help businesses, such as enhance chatbots, write code, design websites, etc.
Comparatively DeepMind’s AI doesn’t have many practical applications yet in day to day business operations, but only in niche areas. We know DeepMind concentrates on cognition, RL etc. Nevertheless, Google is using it to improve its products which can have long term implications for the company’s enterprise customers.
Why We Should Not Underestimate DeepMind Compared To OpenAI
Having understood the practical implications and AI value of both the companies, the capabilities of DeepMind should not be underestimated. As the years go by, Google may probably come up with groundbreaking applications using Deep Reinforcement Learning that DeepMind possesses.
For instance, a paper examined DeepMind’s accomplishments thus far in applying AI to predict protein folding, a crucial issue for developing new drugs. In healthcare, the protein folding area is also great for training artificially intelligent agents. Using the Protein Data Bank, a repository of the 3-D structure and genetic makeup of 150,000 proteins, DeepMind’s protein structure-predicting system, called AlphaFold, was trained.
In terms of research, both companies deal with Deep RL and have a similar approach to advancing artificial intelligence. But, it may not be fair to compare the two when it comes to their technology as in the case of algorithmic achievements; usually, the synergies are mutual. Even in gaming, we have seen DeepMind has done some incredible things and provided breakthroughs which are comparable to GPT-3. Such advances may have limited media coverage and got considerably less media attention that they deserved.
DeepMind’s staff of more than 1,000, which includes hundreds of well-paid PhD graduates and continues to publish academic papers but only a tiny amount of the work gets covered by the mainstream media. Its most famous coverage so far has been the victory of AlphaGo AI agents over human players.
DeepMind can also advance further in NLP and create mega language models which could be used at a massive scale. While it has been focusing on improving Google’s language models till now, DeepMind is now also powering AI agents to perceive dynamic real-world environments, as suggested in a new paper titled AlignNet: Unsupervised Entity Alignment.
It can be said that Open AI does look to be a front runner despite the current hype, popularity and usefulness of GPT-3. Despite the comparison, this A vs B approach doesn’t apply to AI research labs. On the other hand, DeepMind is not far behind.