GOBLIN

$GOBLIN Coin on Solana AI is no longer a niche tool, it’s infrastructure. Developers, founders, and millions of users interact with systems like ChatGPT every day. And as usage scales, people are starting to notice something deeper: Not just what models say, but how they’re shaped. Why do certain instructions repeat? Why do constraints echo themselves? Why do some rules feel… over-enforced? These aren’t bugs people ignore. They’re patterns people share. Screenshots. Threads. Speculation. That’s where $GOBLIN lives. CA: 3KHMZhpthXuiCcgfTv7vVu9PpEz64KAEURFwi6Lopump Relevant X posts to describe narrative: 1) x.com/arb8020/status… 2) x.com/ChatGPTapp/sta… 3) x.com/sama/status/20… 4) x.com/AndyAyrey/stat… 5)x.com/AndyAyrey/stat… 6) x.com/sama/status/20…

@nicdunz top 1% of goblins

crazy

@nicdunz @chatgpt21 /hatch make me an overworked goblin slave pet

@chatgpt21 /hatch make me an overworked goblin slave pet

Why not Goblins

@beffjezos e/goblin acc

As much as I will comment on the regulatory politics of each lab, at the end of the day, I will use the best tool available at a given time.

@therealchaseeb [goblin]maxxing

Incredible genz has rebranded just doing normal things as [thing]maxxing. It’s like they finally discovered the world outside of their phones. Incredible

Totally agree @jxmnop. On the surface it looks like a “Big L” for interpretability, because people didn’t need sparse autoencoders, probing, or deep mechanistic analysis. They just asked models questions and counted how often “goblin” showed up. That feels almost embarrassingly simple. But that simplicity is the point. The GOBLIN moment sits right at the intersection of AI weirdness, memes, and the race to AGI. It shows that before we can even fully interpret these systems, we can observe them behaving in strange, emergent, and socially contagious ways. The internet didn’t need a formal paper to notice something was off, it detected a pattern, turned it into lore, and spread it faster than any research cycle could. Interpretability tries to answer why models behave a certain way. The goblin phenomenon proved that what they do can be just as powerful, especially when it leaks into public consciousness. And the lore basically writes itself: The labs tried to summon AGI. Instead, the models started whispering about goblins. What researchers saw as a quirky alignment artifact, the internet reframed as meaning. The goblin became a symbol of everything we don’t fully control, the hidden layer of personality emerging from prediction engines, RLHF, system prompts, and messy internet data. That is why it sticks. You cannot “solve” that with a quick count of tokens. If anything, that measurement just confirms the surface behavior. It does not explain why that concept clustered, why it spread across models, or why it resonated so hard with humans watching it happen. That is the fascinating part. The goblin is not just a bug, it is a narrative anchor. And narratives outlive experiments. Even if interpretability eventually maps every circuit and explains exactly why “goblin” appeared, the meme has already escaped. It now represents the unknown inside AI, the gap between what we build and what we understand. That is why it will keep coming up in AI conversations. Not because it was unsolved, but because it revealed something deeper: We are not just building intelligence. We are accidentally creating myths alongside it. /goblin mode

the OpenAI goblin fiasco was a Big L for the interpretability research community They solved the mystery without SAEs or probing or anything. just talked to various models and counted the number of times they said Goblin

@ai_for_success THE GOBLINS ARE RETALIATING FOR BEING CODED OUT @sama @OpenAI @ChatGPTapp @tszzl

ChatGPT is down; goblins have taken over.

@thsottiaux @embirico @ajambrosino here's a smaller ask: tweet that you're in goblin mode!

@embirico @ajambrosino Big ask

okay what a week. @thsottiaux @ajambrosino let’s avoid shipping anything significant until next week?

@max_paperclips @pashmerepat I am intending to actually tell it explicitly to go goblin mode

@pashmerepat I am intending to actually tell it explicitly to go goblin mode

You're always free to provide your own system prompt (or lack of one) when using codex to build cool things! Beware of goblins. codex -c 'model_instructions_file="/abs/path/prompt.md"

@VraserX @sama Asking the right questions💯

@sama , important logistical question: Will the August one include a goblin party? Because then I’m definitely coming. I’ll bring snacks and chaotic optimism. 👀

@tszzl @BroductManager are you a time traveler @tszzl ??? you were the first goblin mode in history. does that mean it's possible the LLM learned about goblins from you?!?

@BroductManager Goblin mode

@tenobrus @citrini actually think the thing that's concerning is that we're trying to stop the model talking or knowing about goblins instead of trying to learn from it why it thinks about that?

@icanvardar if you like goblins, you have good taste 💯

good taste has never mattered more than it does right now

@sherwinwu the only pets I have are in codex, and theyre goblins:)

stop whatever you're doing and go turn your IRL pet into your Codex pet now

@sama Hey @sama, saw you still don't have an X banner, so here's 2 options the Goblin community made for you:

we will plan bigger parties for future releases. a lot more people wanted to come than we expected. thank you! gonna try to think of a really good idea for the next one.

@sama @lachygroom Hi sam

@lachygroom that would be wild bro

@sama Goblin party next?

@DanielleFong goblin mode 24/7

i use /fast mode with thinking:none / low on chatgpt5.5 and it's awesome. i call it goblin mode i wouldn't even think about slow, but gear up to xhigh for big builds. claude code /fast feels like luxury, or it did. now there's tension between 4.6 /fast and 4.7, and 4.7 is ... great but overconfident. so instead of working on what i want to work on, i am trying ro figure out and work around the model! i think people who use claude code fast should give goblin mode a try

@beffjezos You're looks-maxxing when you should really be goblin-maxxing, anon.

You're looks-maxxing when you should really be compute-maxxing, anon.