The controversy that surrounds training AI models with copyrighted material is certainly a debatable one.
Recently, tech billionaire Elon Musk spoke about how he strongly agreed with the views that AI experts had about negligible real-world data left for training these models. This was stated by Musk during his livestream chat that he held with the chairman of Stagwell recently.
The SpaceX and Tesla CEO added how the point has come where we’ve all basically exhausted the aggregate of human data in training our models, adding that it took place in 2024.
The statement is not exactly too surprising because we’ve been hearing the same thing on repeat from AI scientists for a while now. This is because the World Wide Web keeps being highlighted as an ocean of useful human knowledge. Sadly the resources are nearly depleted as it was never enough to begin with.
Now, training has sucked up all that was left and what we’ve ended up with is a dry pool near to extinction. Ilya Sutskever who served as the former leading AI scientist on OpenAI did warn about the matter during a boardroom meeting before exiting the firm.
He shared how the AI world reached its limit for peak data while putting more light on his predictions about models. These were created and trained in a unique manner that solves this problem.
Meanwhile, another leading virtual training company called Epoch AI shared how it was sure about us running out of training information in nearly four years. The fact that models keep growing in size and their power is another issue altogether, they added.
For now, we must mention that researchers do like to surround themselves with better ideas and means to tackle hurdles. Just because no more real data exists does not AI training comes to a halt.
So many leading tech giants of today like Meta, Microsoft, and Anthropic, not to mention OpenAI publicly agreed there was an issue. They suggested how they happened to be producing new content in different ways.
Today, up to 60% of the information is up for grabs in different AI products that use material that’s made synthetically. For instance, Google’s Gemma and Meta’s Llama happened to be trained using synthetic content along with the usual real-world material.
OpenAI shared during their interview with Nature recently that they use an abundance of different sources such as public data and collaborations for those that aren’t public. This gives rise to synthetic material generation and also material that originates from AI trainers.
Musk did agree that synthetic content or that coming from AI models alone is definitely where the future of AI training lies. This ends up grading itself and also adds more to processes such as self-learning. Did we mention how cost-effective it can be? The only drawback seems to be less diversity and creativity with material that’s more leaning towards being biased.
Image: DIW-Aigen
Read next: Shorter Stories, Bigger Impact: YouTube’s Three-Minute Rule Changes the Game for Creators
Recently, tech billionaire Elon Musk spoke about how he strongly agreed with the views that AI experts had about negligible real-world data left for training these models. This was stated by Musk during his livestream chat that he held with the chairman of Stagwell recently.
The SpaceX and Tesla CEO added how the point has come where we’ve all basically exhausted the aggregate of human data in training our models, adding that it took place in 2024.
The statement is not exactly too surprising because we’ve been hearing the same thing on repeat from AI scientists for a while now. This is because the World Wide Web keeps being highlighted as an ocean of useful human knowledge. Sadly the resources are nearly depleted as it was never enough to begin with.
Now, training has sucked up all that was left and what we’ve ended up with is a dry pool near to extinction. Ilya Sutskever who served as the former leading AI scientist on OpenAI did warn about the matter during a boardroom meeting before exiting the firm.
He shared how the AI world reached its limit for peak data while putting more light on his predictions about models. These were created and trained in a unique manner that solves this problem.
Meanwhile, another leading virtual training company called Epoch AI shared how it was sure about us running out of training information in nearly four years. The fact that models keep growing in size and their power is another issue altogether, they added.
For now, we must mention that researchers do like to surround themselves with better ideas and means to tackle hurdles. Just because no more real data exists does not AI training comes to a halt.
So many leading tech giants of today like Meta, Microsoft, and Anthropic, not to mention OpenAI publicly agreed there was an issue. They suggested how they happened to be producing new content in different ways.
Today, up to 60% of the information is up for grabs in different AI products that use material that’s made synthetically. For instance, Google’s Gemma and Meta’s Llama happened to be trained using synthetic content along with the usual real-world material.
OpenAI shared during their interview with Nature recently that they use an abundance of different sources such as public data and collaborations for those that aren’t public. This gives rise to synthetic material generation and also material that originates from AI trainers.
Musk did agree that synthetic content or that coming from AI models alone is definitely where the future of AI training lies. This ends up grading itself and also adds more to processes such as self-learning. Did we mention how cost-effective it can be? The only drawback seems to be less diversity and creativity with material that’s more leaning towards being biased.
Read next: Shorter Stories, Bigger Impact: YouTube’s Three-Minute Rule Changes the Game for Creators
