Hacker News new | past | comments | ask | show | jobs | submit login

That is f'ing scary because I think I've read a letter or two just like that.



> I've read a letter or two just like that.

So did the LLM.

Likely thousands as non-fiction, perhaps magnitudes more as fiction, satirical, or speculative. And I bet I'm still missing zeroes — countless business leaders have been writing this drivel a long time.

It's hard for us to keep in mind the sheer scope of online copy. Less hard for LLMs, that's why they feel to me like magic in the Arthur C Clarke sense. I have to keep reminding myself nothing's new under the sun.

This has all happened before, it will all be generated again.


The difference is: I wouldn't be able to regurgitate it on a moment's notice adapted to the situation at hand.


In my eyes, the LLM took very generic high-level instructions, interpreted them correctly, and then followed the instructions to generate output text that was similar to what it had seen before.

What it didn't do was to just use the input prompt as the key into a gigantic lookup table and find the corresponding output text.

A different example of this is when an LLM can correctly interpret a sentence like "the trumpet did not fit in the suitcase because it was too big / small".

Here, "big" would refer to the trumpet, while "small" would refer to the suitcase. This becomes important when translating to German, where "the" comes in different flavors for "trumpet" and "suitcase".

ChatGPT interprets this correctly. It can also explain in detail why "too big" / "too small" makes sense or not for the different interpretations.

You could argue that this is still just a lookup response. But I have a hard time seeing how it can have stored so many different variants in the amount of bytes that the LLM is made up of?

--

ChatGPT explanation for trumpet / suitcase:

Let's delve into the logic and reasoning behind which noun the size descriptor applies to in each of your sentences:

"The trumpet did not fit in the suitcase because it was too big."

In this sentence, the logical interpretation is that the trumpet is too large to fit inside the suitcase. It's understood that if an object doesn't fit into a container, the issue is typically that the object is too large for the container's capacity.

Hence, "it was too big" naturally refers to the trumpet, not the suitcase. It wouldn't make much sense to say a suitcase is "too big" to contain something; if the suitcase were bigger, it would more likely accommodate the trumpet.

"The trumpet did not fit in the suitcase because it was too small."

Here, the logical understanding is that the suitcase is too small to contain the trumpet. When an object doesn't fit into a container due to size, and the descriptor is "too small," it implies that the container isn't large enough.

In this case, "it was too small" refers to the suitcase. If the sentence intended to say the trumpet was too small, it would contradict the idea of not fitting; a smaller trumpet would more likely fit into the suitcase, not less.

In both sentences, the key is understanding how size descriptors (big, small) interact with the concept of fitting something into something else. The size issue that prevents the fitting is logically attributed to the object that needs to be accommodated (the trumpet) in terms of being too big, or to the accommodating object (the suitcase) in terms of being too small.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: