HACKER Q&A
📣 erlich

What if we discovered LLMs 10 years ago?


It feels like a technology that didn't have such a resource-intensive tech-tree leading up to its discovery, so someone could have just discovered Transformers and Attention back then.

Would anything be that different today?

  👤 Shekelphile Accepted Answer ✓
The gulf between the hardware of 2013 and 2023 is huge. Modern GPUs are probably close to ~100x faster for ML (thanks to the use of mixed precision math and hardware level optimizations for that usecase) than the Titan, which was the best consumer NVIDIA GPU you could buy at the time in 2013. Not to mention the huge change in the ML software landscape, tensorflow and pytorch didn't exist back then.
👤 not_your_vase
> Would anything be that different today?

You think today that as a mere mortal it is impossible to reach a human being at any BigTechCo. In 10 years you will remember to the state of today's customer support, and call it the golden age of human interaction - you will have the same garbage interaction with all companies, big or small.

If LLMs were there 10 years, we would have already arrived to that point.