Hacker News Reader

Comment by Capricorn2481 | Hacker News Reader

That's not an LLM, though. We haven't seen evidence yet that an LLM can combine existing language in novel ways. In fact, we've seen over and over that LLMs are quite generic.

Compared to text-to-image, which is seemingly impossible to use without getting something weird

View on HN

There was a post here on hacker news not that long ago where GPT4 came up with "the confetti has left the cannon" when asked for an original phrase similar to "the cat is out of the bag".

Other users confirmed Google could not identify any other use of that phrase.

People saying GPT4 is unoriginal have an uphill battle, it is not the default assumption of anyone who has worked with it.

Confetti cannon is a thing. So there is an obvious link between confetti and cannon. Furthermore, These ("cannon" and "conffetis) are two very common concepts used in a single sentence.

Excuse me for being genuine but I totally fail to understand how original the phrase is.

Why don't you come up with an example of a new phrase with the same meaning that is actually original so we can compare your "real" originality to this "fake" originality?

The dazzling ribbons, stripped from the corpses of trees, shot out of the pressurized tube

ChatGpt is as original as us but cannot ever surpass us because it needs our feedback loop. It is a mirroring of sorts after all.

If it could become extremely creative, push the limits and make ideas so advanced that we can’t understand then it failed at the task so to speak. It always remix idea into larger idea sallads and get our upvotes downvotes. It is a great tool but a tool nonetheless.

I have absolutely seen an LLM combine language in novel ways. With prompting they will create outputs that are entirely novel with concepts intermingled in new ways. It's nonsense when compared to observed reality, but it is novel nonetheless.

That distinction is so pedantic I don’t think it’s worth mentioning. LLM architecture can be extended to other contexts besides language.

> We haven't seen evidence yet that an LLM can combine existing language in novel ways.

How about BubbleSort written in Shakespeare style?

I would say coming up with BubbleSort or Shakespeare's writings from scratch is an example of novelty/creativity.

Most humans can do neither, so it's setting the bar far too high.

I think we should compare what best of LLMs can do with what the best of humans can do for a fair comparison.

Else, we will be comparing what the best LLM can do with what the media/mean human can do.

I can also point to tons of language models (both large and small) which don't do anything remotely useful.

What is fair about that when judging whether or not they can be creative? Most people whom other people would judge to be creative are nowhere near the best of us.

Is there something about the nature of language and linguistic meaning that would make it difficult to combine language in novel ways?

I would posit the opposite, language exists to be combined in any number of ways easily and still communicate well. A language that cannot do this does not I think exist.

Same. I find it hard to makes sense of why anyone would claim that computers can innovate visually but not with language. It would make more sense if someone at least felt that they both could, or both couldn't.

This is mostly a property of the RLHF/instruction-tuned chatbot models most people use. Base models tend to be more varied.

Reply on news.ycombinator.com