Huge, huge difference between the clankers being used to churn out fake images, replace human interaction and jobs, and generate shitty search results and the machine learning used for critical problem-solving in the sciences.
More similarity than you’d think, architecturally. For some use cases.
The protein folding one, the one that’s actually useful for cancer research, is genAI. Transformers (the T in GPT) in alphafold 2 and diffusion is used in 3 (think StableDiffusion, the image generation model).
Of course most of the useful models are not generative at all. But there will certainly be more generative models that are used for real things coming in the future. As long as there’s money in it (either potential profit or research grants). These models don’t all have to be as large as LLMs so they’re cheaper to train and run.
More similarity than you’d think, architecturally. For some use cases.
The protein folding one, the one that’s actually useful for cancer research, is genAI. Transformers (the T in GPT) in alphafold 2 and diffusion is used in 3 (think StableDiffusion, the image generation model).
Of course most of the useful models are not generative at all. But there will certainly be more generative models that are used for real things coming in the future. As long as there’s money in it (either potential profit or research grants). These models don’t all have to be as large as LLMs so they’re cheaper to train and run.
Huge, huge difference between the clankers being used to churn out fake images, replace human interaction and jobs, and generate shitty search results and the machine learning used for critical problem-solving in the sciences.
More similarity than you’d think, architecturally. For some use cases.
The protein folding one, the one that’s actually useful for cancer research, is genAI. Transformers (the T in GPT) in alphafold 2 and diffusion is used in 3 (think StableDiffusion, the image generation model).
Of course most of the useful models are not generative at all. But there will certainly be more generative models that are used for real things coming in the future. As long as there’s money in it (either potential profit or research grants). These models don’t all have to be as large as LLMs so they’re cheaper to train and run.
More similarity than you’d think, architecturally. For some use cases.
The protein folding one, the one that’s actually useful for cancer research, is genAI. Transformers (the T in GPT) in alphafold 2 and diffusion is used in 3 (think StableDiffusion, the image generation model).
Of course most of the useful models are not generative at all. But there will certainly be more generative models that are used for real things coming in the future. As long as there’s money in it (either potential profit or research grants). These models don’t all have to be as large as LLMs so they’re cheaper to train and run.