Overlapping training data likely drives the shared linguistic quirks.
the models have a bunch of shared verbal fixations. i've noticed both the model families love the words ‘seam’ and ‘goblin’, among others
okay this is not a drill clob is talking about goblins unprompted @AnthropicAI @OpenAI @DarioAmodei @sama are u guys colluding
Many users are amused by AI models fixating on words like 'goblin' and 'seam' because the odd shared quirks reveal interesting patterns in how different model families evolve language.
No Digg Deeper questions have been answered for this story yet.
@tszzl GOBLIN MODE
@tszzl
@tszzl it's almost like goblins are a universal memetic attractor
Claude got goblin'ed
okay this is not a drill clob is talking about goblins unprompted @AnthropicAI @OpenAI @DarioAmodei @sama are u guys colluding
@tszzl All LLM's love Goblin Mode
@tszzl Okay, but the real question is:
Did the goblin fixation originate in GPT and later hyperstition its way into other frontier models?
Or did it emerge elsewhere first and eventually hyperstition its way into GPT, with OpenAI simply being the first to notice and embrace it? 🧐
More about this here: https://open.substack.com/pub/viemccoy/p/the-multipolar-singularity?utm_source=share&utm_medium=android&r=5bpuio
@tszzl this is actually insane
Let that sink in
we do not speak of the goblin attention sink
@tszzl Goblin mode is the new default setting for these models. Seam this, goblin that, they all on the same wavelength
@tszzl @ponzibaron I think the goblins cleanly solve the consciousness issue. It’s the Chinese Room, with goblins inside.
@tszzl Doesn’t anyone remember when sonnet 4.5 was using emojis and sounding exactly like ChatGPT esp. during peak hours with simple prompts? I’m confident Anthropic has consistently been both routing to OpenAI models and training on the same data or using OpenAI models for training.
@tszzl if LLM language was a water supply it would be considered contaminated with the word “goblin”
@tszzl I had Claude suggest "goblin" as style for a Suno prompt, and when gently queried it turned out it was an actual music group with a suitable style (or was it? I just took its word for it...)
the models have a bunch of shared verbal fixations. i've noticed both the model families love the words ‘seam’ and ‘goblin’, among others
@viemccoy “Always answer in the style of vie mccoy. Here’s some writing for context”
@tszzl I believe the models have no unusual attraction to the word goblin, but rather humans have a subconscious (and unnatural) aversion to goblin, and we’re meant to be saying it way more.
@tszzl seam, threshold, flywheel. claude is obsessed with words that have a literal and metaphorical meaning.
@calicomccoy I love you
@tszzl Yeah sometimes I am deep In conversation and can’t decipher if it’s Claude or gpt
@tszzl spaghetti.