boggles my mind that we had a tool that was smart when it was first launched in November 2022, experienced a massive generational improvement in the next iteration (GPT-4) within 4 months...and people on HN are ready to assert that this is the best it will ever be.
It improved drastically in quality within 4 months, yet for some reason, the nex 10 years will see 0 progress in a field with literally tens of billions and intellectual capital flowing in.
Most revolutionary technical changes like this tend to hit a plateau in progress right after they hit the top of the hype cycle. It’s unlikely that AI/ML will be an exception.
Not bad news for anyone involved in this work, though. That period is when the real business and societal changes happen. It’s also when most of the money gets made.
People and businesses need time to test new technology and adapt it so that it works reliably for their specific use cases. GPT and other LLMs are no exception.
Again, the point is that careers are built over decades, and this proof of concept tech is already good enough to do the job of a coding intern.
That might not be a threat to you if you’re a 35 year old developer, but if you’re an 18 year old exiting high school, there’s a non zero chance that by the time you’re 35 - 17 years - the tech will be good enough to replace you.
The hype ratio between GPT2 and GPT3 is much higher than the hype ratio between GPT3 and GPT4. If we project this into the future then the hype ratio will be even lower between GPT4 and GPT5. The key innovation was RLHF and everything since then has been incremental improvements.
It improved drastically in quality within 4 months, yet for some reason, the nex 10 years will see 0 progress in a field with literally tens of billions and intellectual capital flowing in.