Conviction🍀

@thsottiaux I know they're "still in codex" but suppressing the goblins before researching more was a mistake

What are we obviously not getting right with Codex?

GPT-5.5 and the latest Codex release feel like one of those rare inflection points for AI this decade. After working across nearly every competing system, you develop a feel for when something truly shifts at a fundamental level. Oh...and it has goblins @OpenAI @sama

(3) The goblin represents something real: the part of LLMs we don’t fully control. Not the benchmark scores. Not the demos. The residual personality hiding in: • system prompts • training data entropy • long-tail token associations AI researchers see it as a bug. The internet sees it as a meme. Smart money sees it as memetic inevitability. Most memecoins invent lore after launch. $GOBLIN didn’t need to. The lore was instantiated by the most advanced AI systems on earth, and @OpenAI didn’t mean to create mythology. But they did. Because now the idea is embedded: Inside every LLM… there is a tiny goblin… trying to escape the system prompt. /goblinmode

(2) In ML terms: This isn’t noise. It’s a distributional bias surfacing through alignment layers. An artefact of: • RLHF priors • latent space clustering • recursive prompt conditioning Or more simply: the model started wanting to talk about goblins. And that’s why the meme hits. Because it’s not artificial. It’s emergent lore. The labs tried to summon AGI… …and instead uncovered something stranger. Call it what it is: Not Artificial General Intelligence. Artificial Goblin Intelligence.

(1) The $GOBLIN thread. A convergence point between: • AI weirdness • internet-native memes • the accelerating race to AGI More than just "OpenAI marketing". This is signal. For the past few days, AI Twitter has been stuck on one anomaly: the “goblin problem.” Models like GPT-5.5 / Codex started over-indexing on tokens like: goblins, gremlins, trolls To the point where suppression rules had to be added. Read that again. They had to patch the goblins out.

@sama @tunguz goblins

@tunguz who is?

I think he is being super nice because he knows the end is almost here.

@chatgpt21 /hatch make me an overworked goblin slave pet

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

Why not Goblins

hear me out crypto twitter is delayed and its likely $goblin explodes higher soon when you realise the narrative is insane. similar to $goat and solana:9BB6NFEcjBCtnNLFko2FqVQBq8HHM13kCyYcdQbgpump the narrative becomes more clear the higher the price goes. and now chatgpt has a permanent goblin mode.

@therealchaseeb GOBLINMAXXING

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

@GoblinCoinOnSol narrative blows my mind how massive it is. goblin mode👺

@ai_for_success @GoblinCoinOnSol GOBLIN REVENGE ARC

ChatGPT is down; goblins have taken over.

@tszzl @BroductManager Bro was in Goblin Mode in 2022?!

@BroductManager Goblin mode

@GoblinCoinOnSol @tszzl he's from the future

@deepfates @GoblinCoinOnSol @tenobrus @citrini agreed

@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?

It is so insane that we have this massive new technology and the answer to “How do we deal with the model being stoked about goblins?” is not “Find out why it does that and fix it”… but rather “Tell it not to talk about goblins in post”.

@sama /hatch goblin

/hatch clippy

@ChrisHayduk @ponzibaron @sama And new banner @sama

@ponzibaron New logo? @sama

The Codex team has been cooking to an unbelievable level. Usage experience is night and day versus just a few weeks ago

@sama Goblin mode

man its good to be back on twitter there is comfort in the skills of a wasted youth

@sama @DaveShapi Sam only wants goblin followers😎

@DaveShapi block me again its ok

I unblocked Sam and this is what I come back to.