Why is GPT-5.4 obsessed with Goblins?

#1

Post by pants2 » Tue, Mar 10, 2026, 5:10 AM UTC

After the 5.4 update, ChatGPT uses "goblin" in almost every conversation. Sometimes It's "gremlin." A recent chat of mine used goblin 3 times in 4 messages:

> this stuff turns into legal goblins fast

> hiding exclusions like little goblins

> But here’s the important goblin

I am not the only one to notice this, there are many Reddit threads on it:

https://www.reddit.com/r/ChatGPT/comments/1roci77/anyone_elses_chatgpt_obsessed_with_goblins_since/

https://www.reddit.com/r/ChatGPT/comments/1rll8hb/suddenly_obsessed_with_goblins_and_gremlins/

---

This is such a weirdly specific word that it chooses to use in over half of its conversations (IME, you should search your chat history for goblin/gremlin and report).

I'm genuinely curious what happens in their post-training that leads to something line this.

What's ironic is OpenAI has been touting 5.4's great personality, but these quirks irritate me like a tiny chaos goblin.

Re: Why is GPT-5.4 obsessed with Goblins?

#2

Post by arthurcolle » Tue, Mar 10, 2026, 5:11 AM UTC

why don't you ask the model?

Re: Why is GPT-5.4 obsessed with Goblins?

#3

Post by Tarraq » Tue, Mar 10, 2026, 6:17 AM UTC

arthurcolle wrote: post #2

why don't you ask the model?

Not to scare away the goblins!

Re: Why is GPT-5.4 obsessed with Goblins?

#4

Post by ghostlyInc » Tue, Mar 10, 2026, 11:29 AM UTC

[flagged]

Re: Why is GPT-5.4 obsessed with Goblins?

#5

Post by d--b » Tue, Mar 10, 2026, 1:16 PM UTC

ghostlyInc wrote: post #4

[flagged]

They seem a lot more common in OP's conversations than in any regular engineering conversation though. Like I've been an engineer for 20 years. I don't remember the phrase used in my work context, ever.

Re: Why is GPT-5.4 obsessed with Goblins?

#6

Post by kilianciuffolo » Tue, Mar 10, 2026, 2:19 PM UTC

I am getting the world goblin and gremlin once every hour.

Re: Why is GPT-5.4 obsessed with Goblins?

#7

Post by ghostlyInc » Tue, Mar 10, 2026, 3:32 PM UTC

d--b wrote: post #5

ghostlyInc wrote: post #4

[flagged]

They seem a lot more common in OP's conversations than in any regular engineering conversation though. Like I've been an engineer for 20 years. I don't remember the phrase used in my work context, ever.

[flagged]

Re: Why is GPT-5.4 obsessed with Goblins?

#8

Post by HPSimulator » Tue, Mar 10, 2026, 7:51 PM UTC

One thing that might also be happening is that LLMs tend to converge on metaphors that compress complex ideas quickly.

If you look at how engineers explain messy systems, they often reach for anthropomorphic metaphors — “gremlins in the machine”, “ghost in the system”, “yak shaving”, etc. They’re basically shorthand for “there’s hidden complexity here that behaves unpredictably”.

For a model generating explanations, those metaphors are useful because they bundle a lot of meaning into one word. So even if the actual frequency in normal conversation is low, the model might still favor them because they’re efficient explanation tokens.

In other words it might not just be training frequency — it could be the model learning that those metaphors are a compact way to communicate messy-system behavior.

Re: Why is GPT-5.4 obsessed with Goblins?

#9

Post by muzani » Wed, Mar 11, 2026, 12:03 AM UTC

It could be a kind of watermark. It's possible they aimed for it to be just 5% more noticeable but overshot it. Also humans tend to spot these things better than computers.

It used verdant excessively in the past, but that's a less noticeable word than goblin.

Re: Why is GPT-5.4 obsessed with Goblins?

#10

Post by tmaly » Wed, Mar 11, 2026, 3:38 PM UTC

HPSimulator wrote: post #8

One thing that might also be happening is that LLMs tend to converge on metaphors that compress complex ideas quickly. If you look at how engineers explain messy systems, they often reach for anthropomorphic metaphors — “gremlins in the machine”, “ghost in the system”, “yak shaving”, etc. They’re basically shorthand for “there’s hidden complexity here that behaves unpredictably”. For a model generating explanations,…

I am waiting for future versions to start compressing with memes