3 ms·

Surprising lack of foresight on his part. Although I'm not even sure you need foresight now, everything is transforming already.

blamestross 3y ago
It feels surreal seeing so many comments like this here and in public discourse in general. The advertising/hype machine convincing people a low quality continuation model is the next coming of AI must be amazing. LLMs have lots of cool advancements involved in them but isn't what you are being sold. Anybody who knows how these models work and isn't making money off them can explain just how limited they are and how they can't safely be applied at scale. Yet they are doing so anyways. Invest and lie while you can, and get out before it all falls apart.
- ETH_start 3y ago
  ChatGPT is already providing demonstrable boosts to productivity: https://www.searchenginejournal.com/chatgpt-boosts-productivity-40-for-writing-tasks-study/491845/ https://www.searchenginejournal.com/chatgpt-boosts-productiv... >>A new randomized controlled trial by Harvard and Stanford economists finds that ChatGPT can substantially boost productivity for professionals completing everyday writing tasks. >>Participants were randomly assigned realistic work simulations with or without access to ChatGPT. >>Those with access to the AI tool completed assignments 40% faster while producing 18% higher quality output as judged by expert evaluators. Any assessment of LLMs that overlooks the profound positive impact they are already having on society is completely inaccurate.
  - blamestross 3y ago
    Yes, they do have use. They are also very oversold and literally being sold for jobs they cannot do. It does help write things that were either formulaic or entirely without real content. Allowing them to do analysis or summary of anything that isn't already in a Wikipedia entry is radically dangerous because of how good they are at hallucinating and lying (and we can't detect that at scale). It is like an automated "yes man". This is the most dangerous bubble in living memory and the cost of it popping will be in the failure of any industry that is foolish enough to rely on them at scale. And to attack that study in particular, "help people who couldn't write well write better" is complicated, increasing accessibility is good, but there is a BIG difference in an increase in writing quality versus an increase in productivity AND correctness.
    - ETH_start 3y ago
      Everything new is going to be oversold. That is not an indictment of the technology, although it's also fair to point that out to counteract the exaggerated perceptions that the overselling creates. Putting aside all of the hype, LLM are already providing enormous productivity boosts in some activities, and those boosts can be fairly characterized as revolutionary when you consider how significant they are and how quickly they came about since GPT2.

blamestross 3y ago

It feels surreal seeing so many comments like this here and in public discourse in general. The advertising/hype machine convincing people a low quality continuation model is the next coming of AI must be amazing. LLMs have lots of cool advancements involved in them but isn't what you are being sold. Anybody who knows how these models work and isn't making money off them can explain just how limited they are and how they can't safely be applied at scale. Yet they are doing so anyways. Invest and lie while you can, and get out before it all falls apart.

ETH_start 3y ago
ChatGPT is already providing demonstrable boosts to productivity: https://www.searchenginejournal.com/chatgpt-boosts-productivity-40-for-writing-tasks-study/491845/ https://www.searchenginejournal.com/chatgpt-boosts-productiv... >>A new randomized controlled trial by Harvard and Stanford economists finds that ChatGPT can substantially boost productivity for professionals completing everyday writing tasks. >>Participants were randomly assigned realistic work simulations with or without access to ChatGPT. >>Those with access to the AI tool completed assignments 40% faster while producing 18% higher quality output as judged by expert evaluators. Any assessment of LLMs that overlooks the profound positive impact they are already having on society is completely inaccurate.
- blamestross 3y ago
  Yes, they do have use. They are also very oversold and literally being sold for jobs they cannot do. It does help write things that were either formulaic or entirely without real content. Allowing them to do analysis or summary of anything that isn't already in a Wikipedia entry is radically dangerous because of how good they are at hallucinating and lying (and we can't detect that at scale). It is like an automated "yes man". This is the most dangerous bubble in living memory and the cost of it popping will be in the failure of any industry that is foolish enough to rely on them at scale. And to attack that study in particular, "help people who couldn't write well write better" is complicated, increasing accessibility is good, but there is a BIG difference in an increase in writing quality versus an increase in productivity AND correctness.
  - ETH_start 3y ago
    Everything new is going to be oversold. That is not an indictment of the technology, although it's also fair to point that out to counteract the exaggerated perceptions that the overselling creates. Putting aside all of the hype, LLM are already providing enormous productivity boosts in some activities, and those boosts can be fairly characterized as revolutionary when you consider how significant they are and how quickly they came about since GPT2.