Fasciating. Language is a type of action evoloved for information exchaging, which maps latent "video", "audio" and "thoughts" into "sentences" and vice versa.
Asking models to do math is kind of an effecitve way to measure their capabilities, especially in reasoning and abstraction, which are quite important for problem solving.
You don't need to reason and abstraction to do basic calculation. ChatGPT will however happily give you some decent answers about not-too -hard math that requires reasoning. It just won't operate on digits.
I think the last paragraph quite makes sense. It seems "true" that some kind of reasoning capability emerges as LLMs get bigger, which makes those LLMs quite useful and blows a lot of people's minds at the beginning. But, I think, essentially, the fundamental training goal of LLMs--guessing what the next word should be--pushes the model into a kind of reasonable nonsense generator, and the reasoning capability emerges because it can help the model to make stuff up. Therefore, we should be cautious about the result generated by these LLMs. They might be reasonable, but to make up the next word is their real top priority.
Maybe in the future there might be some kind of "forbidden to use in commercial AI models" policy on websites in the near future, just like what the art community is doing now.
I can't wait to see the model's ability of saying "They don't know", which I think is an important feature if it serves as a search engine, because it can reduce the amount of generated ramblings which it's actually really good at.
So i think the ability of the search engine to say "I don't know" is very important, and most of current chatgpt like models in the market don't have this feature.
It's fascinating to think about the future landscape of the search and web.
Some assumptions:
1. Url-based web will not wither away.
2. Asking questions in the chat-like mode is more natural to people.
3. Generated answers cost more when longer.
4. Generated answers are some kind of distilled knowledge and can't be right all the time.
4. People don't like long answers and prefer the concise one.
5. Sources and citations make generated answers more credible.
6. Fully exploring a question needs a lot of information from different views.
7. Generated answers
some simple thoughts:
The search behavior would hugely be two main steps: 1.getting some concise answers from the AI model directly through a chat, which might be enough for 90% use cases. 2.some more extensive search just like how people are searching today, which might be a kind of niche.
For websites, being cited in the generated answers will be the new kind of SEO things, and it would be a good strategy to producing some newest, deep or long-tail knowledge and information, which leads to a more traditional way of search because AI model doesn't have enough data to generate a good answer.
>Asking questions in the chat-like mode is more natural to people.
It's not just that it's chat, its the ability to refine. Currently, I search something. It returns garbage. I search something new. What I dont do is tell the search what it did wrong the first time. I might sort of do that with -words, but its a fight every time.
The beauty of these new chat systems is that they have short term memory. Its bein able to work within the parameters and context of the conversation. I dont particularly care if it is "chat like" or has its own syntax, what I want is a short term state.
And at the same time, I want long term state. I want to be able to save instructions as simple commands and retrieve them later. Like if I am searching for product reviews, to only return articles where it is convinced the people actually bought and tried the products, not just assembled a list from an online search.
