4 ms·

I know LLMs and genAI are trendy and they seem espectacular and work like magic, but maybe they are not the tool for every problem. I can see them working fine

by 101008 2y ago

I know LLMs and genAI are trendy and they seem espectacular and work like magic, but maybe they are not the tool for every problem. I can see them working fine in a lot of aspects/problem, but that doesn't mean they will work fine in every aspect that involves text, even if they are refined, etc. I guess once the fashion passes (probably not soon, unfortunately), people will see the real value of LLMs and will start applying them to problem that are solvable by them, and not absolutely everything.

Workaccount2 2y ago
No one knows what the real value of LLMs is though. Its a nascent technology. We could be discussing the internet circa 1994 or we could be discussing bitcoin circa 2011.
- skydhash 2y ago
  > the internet circa 1994 or we could be discussing bitcoin circa 2011. People were clear about the value of both. SMTP by itself was already a clear value for business. The WWW itself was invented in 1989. Same for Bitcoin as the title of the original paper, "Bitcoin: A Peer-to-Peer Electronic Cash System ", explains it all. Some capabilities were not there yet, but it was mostly infrastructure and engineering issues (Acceptance was another issue for bitcoin). LLMs generate tokens. The generation is governed by the models, but the model itself has no concept of usefulness or truth. Books are not enough to train them. The WWW has a lot of garbage informations. And natural language is insufficient to describe exact processes without formalizing it. The fact is that the human mind can already do what the LLMs fans hope them to achieve. And we can transfer knowledge through media. Getting something done is mostly assembling the right people in a room, recursively. And we built tools to enhance productivity. I believe that Microsoft and others are hoping that the general public will be ok with wrong results. But the fact is that all the capabilities exposed so far have an alternative and more deterministic process to achieve the same result. But it often requires learning or hiring a specialist and it's sad to see how many people are balking at that.
__loam 2y ago
Problems like auto-generating spam emails for Nigerian princes.

by 101008 2y ago

Workaccount2 2y ago

No one knows what the real value of LLMs is though. Its a nascent technology. We could be discussing the internet circa 1994 or we could be discussing bitcoin circa 2011.

skydhash 2y ago
> the internet circa 1994 or we could be discussing bitcoin circa 2011. People were clear about the value of both. SMTP by itself was already a clear value for business. The WWW itself was invented in 1989. Same for Bitcoin as the title of the original paper, "Bitcoin: A Peer-to-Peer Electronic Cash System ", explains it all. Some capabilities were not there yet, but it was mostly infrastructure and engineering issues (Acceptance was another issue for bitcoin). LLMs generate tokens. The generation is governed by the models, but the model itself has no concept of usefulness or truth. Books are not enough to train them. The WWW has a lot of garbage informations. And natural language is insufficient to describe exact processes without formalizing it. The fact is that the human mind can already do what the LLMs fans hope them to achieve. And we can transfer knowledge through media. Getting something done is mostly assembling the right people in a room, recursively. And we built tools to enhance productivity. I believe that Microsoft and others are hoping that the general public will be ok with wrong results. But the fact is that all the capabilities exposed so far have an alternative and more deterministic process to achieve the same result. But it often requires learning or hiring a specialist and it's sad to see how many people are balking at that.

__loam 2y ago

Problems like auto-generating spam emails for Nigerian princes.