Hacker News new | ask | show | jobs
by rishabhjain1198 1082 days ago
I wish this article was just 3 paragraphs. The verbose writing style was a little tiring, I found myself scrolling impatiently to find what the actual "Lone Banana Problem" was.
10 comments

> I wish this article was just 3 paragraphs.

AI might have problems with single bananas, but it can do that very well:

> In an experiment with the AI program Midjourney, the author found a peculiar issue: the program rendered images of monkeys holding bananas, but it consistently depicted two or more bananas even when asked to render a single banana.

The author suggests that the AI’s predilection for rendering multiple bananas might be due to biases in its training data or the lack of precise labeling. He points out that AI systems like Midjourney don’t understand objects in the human sense but rather recognize common patterns. These systems are only as good as the data fed to them and can inadvertently contain biases or incorrect representations.

The article further delves into existential questions about the nature of human intelligence, creativity, and morality compared to AI’s pattern recognition. It questions whether human cognition and morality are just advanced pattern matching with a better understanding of the physical world. The author also touches on the distinction between programming and prompt engineering, highlighting that the latter requires a more nuanced understanding of language and how AI models interpret it.

That's immediately inaccurate in a way that distorts the whole article.

The author was trying to produce a single banana on a grey background.

The monkey was a later attempt at prompt engineering.

That AI summary is completely misleading.

The lone summary problem
The problem with generated summaries as opposed to generated code for example, is that there is no way to verify them without reproducing the entire thing "manually".

Can't do better then brute force for verification.

This is the biggest communication problem most people have.

Say it plain and simple. No one wants to read your train of thought.

Thought process is fine, but lead with the plain and simple version. Offer context after you've introduced the topic and described what you're talking about.
I think when you're trying to be convincing of a counter intuitive or controversial point, it helps to start with less objectionable parts of the argument first. If an argument involves several steps, A to B to C ending in D, but D is wildly counter intuitive, some audiences, hearing D first will reduce their likelihood of actually integrating the consequences of A and so on. Maybe someone who's done more research on the psychology of priming would know?
I'm not so sure. A shining example is the essay "One Self: The Logic of Experience"[0]. The first paragraph is just 3 sentences, and the first sentence lays it out:

"I believe that we are all the same person. How could anybody come to think that? In this paper I shall try to explain how it happened to me." [https://www.researchgate.net/publication/233329805_One_Self_...]

Nah. It's either the 2nd biggest communication problem, or a side effect of a bigger problem. Depending on how you analyse it.

The biggest communication problem is that you always need to take into account that your reader might be braindead trash. That has two consequences:

1. Unless you expose your full train of thought, expect screeches like "I dun unrurrstand, SPOONFEED ME BASIC REASONING, REEE".

2. Unless you explicitly say something, expect some assumer to claim that you said the opposite. Bonus points if this is due to failure to take context into account, or even notice that the context is missing.

Both consequences have been training writers to idiot-proof their texts with big walls of unnecessary words. And that's the case here.

I used to write emails to my boss as summary, then clearly mark what follows as the (boring) details. I got a lot of promotions that way.
Apparently those kinds of people consider you to be the one with the communication disorder. I consider them incapable of writing an abstract.
It was probably 3 paragraphs before they ran it through a prose-generating LLM. Now we need to use an LLM to condense it back down.
Agree. I was a couple of pages in before I learned what the "Lone Banana Problem" is. That took two paras, and was followed by a lot of sophomoric philosophical noodling.
Maybe fittingly, it is a writing style I have associated with AI (ChatGPT in particular). Here, the article doesn't have the ChatGPT "signature", but it has the same tendency of having a lot of grammatically correct filler.
Someone needs to create a browser that parses and displays every website in a format that I request.

My preferred format would probably be:

- answer to the title / synopsis

- the article, without ads, using the site’s style

I skipped to Going Bananas, speed read the rest and it was fine.
The pictures did not immediately tell you?

They told me what the problem was in about 2 seconds, and frankly, the title and the first sentence alone were.

All right, I concede, a blind reader would not know, but most people are not blind.

I read almost the whole article thinking there might be something more in there. There was mention of TCAV but that was about it.
I am mystified.

The page is not long, the text is not complex. The message is obvious and it is contained in the title as shown here on HN.

The first pic and its caption make it plain.

The next group of 4 pics and their SINGLE SENTENCE caption spell it out clearly.

As for "almost the whole article" -- it is short! It took about 90sec to read the whole page, top to bottom.

"AI" chat bots can't count. This is well known. It is the stuff of memes.

https://osu.ppy.sh/community/forums/topics/1770930?n=1

> As for "almost the whole article" -- it is short! It took about 90sec to read the whole page, top to bottom.

Cool. The article has 2348 words, so that's around 26 words per second, or ~1560 words per minute.

According to some speed reading pages I randomly found via Google the consensus seems to be that 1000+ words per minute is quite exceptional. https://irisreading.com/what-is-the-average-reading-speed/ puts average adult reading speed at around 300 WPM, which you casually exceed by 400%.

Far be it from me to suggest that your estimate is inaccurate, but at least know that not everybody reads at that speed :-)

Yeah, I am a speedreader, and I did used to be able to do circa 3000wpm at a push. Good to know that even as a myopic 55YO I can still do half that without trying.

Points of comparison: I originally read Hal Clement's classic Mission of Gravity in about 25min, and Joseph Conrad's Victory in about 3 hours.

Did you forget you’re a speed reader when asserting how long it should take to read the whole article?
Right, so your 90 seconds is actually closer to 10 minutes for most people. That's a lot of time to communicate "AI tends to render one banana as two bananas", which was OP's point.

Yes, there were pictures that communicated the same thing, but people naively thought that those 2348 words may actually hold some additional information, as otherwise ca 2300 of them would be completely superfluous. But they were wrong; again, OP's point. Not sure what's so mystifying by this.

My point exactly. This article could have been written: "AI is biased against drawing just one banana" and I would have preferred it. And I said "almost" because I skimmed two of the paragraphs in the middle that seemed unlikely to contain information.
But that isn't the point.

The point is, LLMs look smart but they are not, and an easily-verifiable data point is that they can't count.

Ah, right, and that's not obvious to some people... I guess I'm just not the target audience.
Do you really not know how to skim effectively? Took me just a few seconds.
Right? I'm running a fever right now and I had no trouble with this article.
you're stumbled upon the Lone Writer problem.