4 ms·

Give those metrics a year, and we’ll see how it turns out. This reasoning completely disregards the long term vision for a clear quick gain. We’re in a reality

Give those metrics a year, and we’ll see how it turns out. This reasoning completely disregards the long term vision for a clear quick gain. We’re in a reality distortion field where, for some reason, everyone thinks that the circus trick performed by GPTs is an absolute, it’s here to last forever and it will just get better. I can’t think of any other technology that has fooled business people so quickly into adopting it without ever worrying about the alpha nature of the tech itself. Feels like mass hypnosis

sebzim4500 4y ago
> We’re in a reality distortion field where, for some reason, everyone thinks that the circus trick performed by GPTs is an absolute, it’s here to last forever and it will just get better Why would it not last forever and improve? Unless OpenAI is currently selling api/chat access at a loss, but I see no reason to believe this.
- camillomiller 4y ago
  Because infinite grow does not exist except in the deceptive narrative of neoliberal economics? Everyone is SO focused on the immediate byproduct of this new technology, and there's so little questioning if applying it to business can actually make sense in the mid to long term. There are regulation pitfalls waiting to happen. We have no idea how web search and content consumption will adapt to this avalanche of cheap irrelevant content. Yet we give for granted this is the future of everything, based on the fascination with what these machines can do right now. I'm saying this as someone who's also extremely fascinated and interested in the technology. This is the most excited I've been about tech in quite some time, but as I said, from a business perspective I've never seen so many business minds dive lemmings-style into an alpha-state technology before, and it's worrying how FOMO is pushing everyone to just take GPTs for granted!
  - sebzim4500 4y ago
    You mean because in future it will be impossible to obtain a large high quality dataset due to AI contamination? Worst case you pre-train a model on pre 2022 data and finetune on a small, curated dataset of recent events. And there are some important breakthroughs that have happened since GPT-4 started training, so it will almost certainly be possible to make a significantly better model with the same dataset and inference cost.
  - cultureswitch 4y ago
    The LLMs are useful tools in their current state, right now. They produce text and images, which you can use freely after generation. Nothing about their success is predicated on assumptions about the future. There's no risk associated with using an LLM today.

sebzim4500 4y ago

> We’re in a reality distortion field where, for some reason, everyone thinks that the circus trick performed by GPTs is an absolute, it’s here to last forever and it will just get better Why would it not last forever and improve? Unless OpenAI is currently selling api/chat access at a loss, but I see no reason to believe this.

camillomiller 4y ago
Because infinite grow does not exist except in the deceptive narrative of neoliberal economics? Everyone is SO focused on the immediate byproduct of this new technology, and there's so little questioning if applying it to business can actually make sense in the mid to long term. There are regulation pitfalls waiting to happen. We have no idea how web search and content consumption will adapt to this avalanche of cheap irrelevant content. Yet we give for granted this is the future of everything, based on the fascination with what these machines can do right now. I'm saying this as someone who's also extremely fascinated and interested in the technology. This is the most excited I've been about tech in quite some time, but as I said, from a business perspective I've never seen so many business minds dive lemmings-style into an alpha-state technology before, and it's worrying how FOMO is pushing everyone to just take GPTs for granted!
- sebzim4500 4y ago
  You mean because in future it will be impossible to obtain a large high quality dataset due to AI contamination? Worst case you pre-train a model on pre 2022 data and finetune on a small, curated dataset of recent events. And there are some important breakthroughs that have happened since GPT-4 started training, so it will almost certainly be possible to make a significantly better model with the same dataset and inference cost.
- cultureswitch 4y ago
  The LLMs are useful tools in their current state, right now. They produce text and images, which you can use freely after generation. Nothing about their success is predicated on assumptions about the future. There's no risk associated with using an LLM today.