Since OpenAI is deprecating old models, it should only make sense that they make them publicly available to download.
Nice idea but I suppose there is no way.
One example I can think of is when one of my previous companies retired the old versions of the tri-eccentric valves for a modernized version. Making the design publicly available would have been a huge mistake as the competition would have easily saved years of research and decades of experience.
2 Likes
By open sourcing the deprecated models, they might be giving out the secret sauce.
I wish they would open source them, but I can understand why they may not do this, at least not until the secret sauce is vastly different in whatever later generations we are on.
There is this war between proprietary and OS models right now if you read the headlines. So no incentive for proprietary to help OS. You have to put yourself in their shoes.
1 Like
Also, the magnitude of computational power required to execute these models is a considerable accomplishment, and the model itself would need to possess a capacity of at least a few dozen terabytes.
1 Like
I doubt that, especially considering how old they are.
There are actually great solutions for that. I agree! After training a model, then all sudden you have to switch again cab be a pain. it may be why ppl are wanting to switch to open source. They can know that their data and information is private. They don’t have to keep retraining models as well. *** Increased transparency and accountability**
*** Faster innovation and collaboration**
*** Better safety and trust in AI systems**
*** Gaining trust and improving public image**
*** Potentially generating new revenue streams**
*** Faster innovation and collaboration**
*** Improved data privacy and security**
*** Ability to customize and modify AI models to suit specific needs**
*** Contribution to the overall advancement of AI technology**
Closed Source AI Companies’ Concerns
*** Protecting intellectual property and maintaining a competitive edge**
*** Fear of giving away secrets or losing revenue**
Solutions:
*** Use open source licenses that allow for limited use or modification of the AI models**
*** Implement measures to protect intellectual property while still promoting transparency and accountability**
*** Offer incentives for companies to open source their older AI models, such as tax breaks or public recognition**
*** Create a platform for companies to share and collaborate on AI models, promoting innovation and progress**
*** Develop a framework for responsible AI development and deployment, ensuring that AI models are aligned with human values and ethics**
1 Like
Will I agree with most of the things here, I don’t follow the logic on “retrain models” because of their deprecation. From the business stand point I mean. Maybe I’m missing something.
I work with fine-tuning since it is available and what I see is you keep getting better data and that’s why you retrain your models anyway. Or when new models are available, their cost is lower and performance is generally better, so retaining using latest models reduces your costs of operation and that’s why you do it.
Actually, what I don’t get is how your app can go without retaining models for enough time that they get deprecated.
Maybe because your dataset is so vast that it is a costly endeavor to retain? If that’s the case, are you fine-tuning to “teach the model domain knowledge”? And if that’s the case, you’re most likely doing it wrong, as good RAG + workflows are way more efficient to get the results done.
If I’m wrong, so what are the reasons to keep fine-tuned models for so long?
1 Like