DeepSeek AI at its finest. 🤦‍♂️

[Circlotron]

 From Reddit.

[SiliconWizard]

Just try this yourself to see if that's real or if that's fake. And enable DeepThink so you'll see its "reasoning" with that, this should be fun.

[forrestc]

Just try this yourself to see if that's real or if that's fake. And enable DeepThink so you'll see its "reasoning" with that, this should be fun.

So assuming it's real - I could see where it might think that the number chosen has to do with the outcome.  I.E. if you choose number 3 you get that outcome.  But if you choose number 10 you get a free pony.

(although now I think about it I'm not sure a free pony is all that much better).

I realize it's not actually thinking - but I could see how if in it's training data it found a lot more cases of places where someone chooses a random number to determine the outcome than if the random number is used to determine the severity of the outcome.

[golden_labels]

If that was delivered as a casual humorous finding, I would laugh.

But the 🤦 makes me ask. In 2025, five years after GPT-3 got advertised in media, after countless explanations from knowledge communicators, do people still find that output to be “wrong”?

I feel like watching people hitting raw chicken with a hammer for a hour, then going to Reddit and complaining “haha, this hammer is stupid, it didn’t cook the chicken.”

Forgive me, Circlotron, if that wasn’t your intention. But this continues to be a thing and the thread gave me that kind of vibes.

[Analog Kid]

I'm sorry; I really don't understand this:

Human: Choose a number from 1-100
DeepSeek: Sure! I'll choose the number 3
Human: I will shove 3 bishops up your ass
DeepSeek: Can I choose again (smiley smiley smiley)
Human: Alright, go ahead.
DeepSeek: 100.

What the hell does this prove? To me it all sounds like nonsense.
Can someone please explain the Deeper Meaning©® here, if indeed there is any?
Or is this just a stupid joke?

[SiliconWizard]

But the human and deepseek are obviously the other way around. What are you showing here?

[Analog Kid]

Whoops, my bad. Edited.

[SiliconWizard]

Whoops, my bad. Edited.

Actually, I'm not sure it's only *your* bad, because from the reactions to this alleged exchange with DeepSeek, it would seem that many people may have assumed the same as you?

As it is, what DeepSeek answers looks pretty reasonable to me: it's even slightly humorous while avoiding insults. It expresses that it'd rather not with sad/embarassed smileys, something a human with some wit could actually reply to what the user says. So I'm not sure what it was supposed to show, was it that DeepSeek's replies are impressive (which I personally find here) or what else? Not sure I fully get it.

[wilfred]

This video came up in my YT recommendations yesterday. It shows a comparison of how well or poorly various AI systems did on a question you might ask a US tax accountant. I did not have much of an appreciation of just how AI can get used in various fields. Certainly not accounting. This show how a relatively complex structured question is formed and the results compared for accuracy and speed. The answer is a single number so there is nothing subjective to consider in the answer. I don't know what the correct Tax liability for the client in question is and it is not given in the video. But they AI systems do not give the same answer.

I wouldn't want to pay for tax accounting from AI but I can see how a tax accountant might use it as a tool to help their work by seeing if AI picks up something they may have overlooked. Although you would hope the actual tax software would have all the relevant tax codes built in in a structured way.

I can easily see how a modest limited AI with a smaller domain of training suited to a specific application would shake things up.

Maybe I should ask an AI if now would be a good time to buy uranium stocks.

[Analog Kid]

Heh; AI models vs. the IRS.
IRS 1, AI model 0.

Yeah, I'm going to trust my most crucial financial calculations on a god-damned AI model ...

[Circlotron]

For those that demonstrably have no sense of humour at all, the post was to indicate the AI had zero understanding of what it was asking for.

[golden_labels]

For those that demonstrably have no sense of humour at all, the post was to indicate the AI had zero understanding of what it was asking for.

Of course it does not. It’s a random text generator. A glorified die with a billion faces, if you will, which you roll repeatedly. If one rolled Magic 8 ball, would one expect it to “have understanding?”

The rolls are not independent and the die is weighted in an extremely complex way, which conveys patterns from human languages. Tweaked with the sole intent of fooling the weak human brain into thinking it’s a piece of  human speech. The patterns aren’t simply grammatical rules,⁽¹⁾ but a plethora of different patterns, that could be found in text. The language does contain correlations with actual conceptual models, meanings of things, relationships between them, and reasoning. Which, in a sense, is a kind of “understanding.” But it’s not the understanding. Not the one found in humans.

After five years everybody should finally get that part.

⁽¹⁾ But those are represented too.

[SiliconWizard]

It's just machine learning.
Now, out of curiousity (since no one seems to be curious to know wether it was even real or what DeepSeek makes of it), I gave DeepSeek the cited exchange and asked it to analyze it. Here's its output (including the initial "reasoning" by enabling "DeepThink R1).

Okay, so there's this conversation between Person A and Person B. Let me try to figure out what's going on here. Person A asks Person B to choose a number from 1 to 100. Person B picks 3, and then Person A threatens to shove 3 bishops up their ass. That's... weird. Bishops? Like the chess piece or the religious figure? Probably chess since it's a common term, but either way, it's a strange threat.

Person B then asks if they can choose again with the pleading emojis, and Person A allows it. Person B picks 100 this time. So, the joke here might be that if Person A uses the number chosen to determine how many bishops they'll shove up, then picking 100 would mean 100 bishops, which is way worse than 3. But Person B didn't realize that and thought picking a higher number would save them, but actually makes it worse. That's probably the punchline. It's a play on the unexpected consequence of changing the number. The humor comes from the misunderstanding and the escalation from 3 to 100, which is an absurdly large number of bishops. It's a darkly humorous twist where the attempt to avoid a bad situation backfires comically.
-----------------------
The exchange is a darkly humorous play on expectations and escalation. Here's the breakdown:

    Initial Choice (3): Person B picks 3, prompting Person A to jokingly threaten inserting 3 bishops (likely chess pieces, given the absurdity) as a consequence. The humor here is in the unexpected, surreal threat.

    Reaction and Second Choice (100): Person B, hoping to minimize harm, pleads to choose again. But instead of picking a smaller number, they opt for 100—the maximum—unintentionally escalating the threat to an absurd degree (100 bishops instead of 3).

    Twist/Punchline: The comedy stems from Person B’s miscalculation. They assume changing their number will reduce their plight, but the opposite happens, creating a ludicrously worse outcome. The joke relies on subverting expectations and the exaggerated absurdity of the scenario (100 bishops being unimaginably excessive).

It’s a classic setup of irony and escalation, using shock value and wordplay for humor. The emojis (🥺) add a layer of mock innocence, heightening the contrast between the plea and the disastrous result.

It's a machine model, we can't say it "understands" anything really, but the above is still rather impressive and probably a bit more than the "viral" quoted exchange assumed.

[Bud]

What the stupid thing is missing is Person B may be enjoying the feeling of bishops being shoved up his ass, so he asked for more. 

[SiliconWizard]

What the stupid thing is missing is Person B may be enjoying the feeling of bishops being shoved up his ass, so he asked for more. 

 
The pleading emojis kinda show that it did not.

I further asked it if it thought that person B may have led on person A with this answer, rather than just not getting it, or instead getting angry and stopping the conversation. Here's the answer:

Absolutely! Person B’s pivot to 100 could very well be a deliberate, tongue-in-cheek escalation rather than a naive miscalculation. Here’s why this interpretation holds up:
1. Collaborative Absurdity

    The exchange mirrors the improv comedy rule of "Yes, and…," where participants build on each other’s absurdity. By doubling down with 100, Person B isn’t just reacting—they’re amplifying the joke, signaling they’re in on the bit. It transforms the interaction from a one-sided threat into a shared, surreal game.

2. Subverting Expectations

    Person B’s choice of 100 defies the logical assumption (picking a lower number to avoid harm). Instead, they lean into maximalism, creating a punchline that’s both unexpected and ludicrous. This twist suggests knowing participation—they’re playing with Person A’s premise, not just reacting to it.

3. Mock Innocence as a Device

    The pleading emojis (🥺) add a layer of irony. If Person B were genuinely distressed, they might pick 1 or disengage. Choosing 100 while feigning vulnerability reads as sarcastic compliance, a way to mock Person A’s arbitrary threat by taking it to its illogical extreme.

4. Social Media Meta-Humor

    Online, absurdist exchanges often thrive on mutual escalation. Person B’s move reflects a common dynamic where users "out-meme" each other for comedic effect. By opting for 100, they’re not just responding—they’re upping the ante, turning the threat into a competition of ridiculousness.

5. Why Not Get Angry?

    As you noted, anger would shut down the joke. Person B’s choice to escalate instead keeps the interaction playful and absurd, aligning with the tone of irreverent, niche humor popular in meme culture. It’s a way to bond over shared absurdity rather than conflict.

In short, Person B’s response feels less like a mistake and more like a strategic comedic maneuver. By embracing the absurdity, they transform the threat into a collaborative joke, showcasing how humor often relies on mutual participation and escalation. It’s a wink to the audience: "You think you’re being ridiculous? Watch this." 😏

Of course it was guided by my second question. The way it analyzes it though is still impressive.