Never has the debate between “open” or “closed” systems been more hot than when it comes to large language models of AI.
One of the biggest fears that AI insiders have is the outsized control that big tech companies like Microsoft, Google, and Amazon will have over the generative AI companies.
Big tech has invested heavily in startups that have led the way in generative AI engines.
OpenAI’s GPT has received over $13B in funding from Microsoft.
Google and Amazon both have stakes in start-up Anthropic.
However, one organization has a much different approach to the concept of openness when it comes to the models used to train generative AI tools.
The Allen Institute for AI is a non-profit research institution founded in 2014 by Paul Allen, the co-founder of Microsoft and longtime philanthropist.
Ali Farhadi, the CEO of the Allen Institute for AI (AI2), advocates for a glass-box approach to the training models and decision criteria that gen AI models use.
AI2 released the first of these open training models containing 3 trillion tokens that companies can use to train their own generative AI tools.
Advocates for this believe that it allows for a crowdsourcing approach to managing the data set to remove possible hallucinations and other negative responses by the generative tools that use them.
But OpenAI and others prefer to keep their decision-making private in a so-called “black box”, fearful that open models can be manipulated by bad actors.
The topic will clearly continue to be hotly debated by AI insiders for years to come, but AI2’s approach is not unlike Linux’s open-source OS and Mozilla’s open-source browser.