Are they though? I think the adaptive part is the memcached containing your "profile" that is running server-side w/ OpenAI and Anthropic. The model itself is deterministic in nature, which is evidenced by OpenAI's "goblin-fix" in codex:
So by the time that OpenAI released GPT-5.5, it had added a system prompt to its Codex programming harness, instructing the model to “never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user’s query.”
Changing a system prompt means you're changing the runtime input, not the machine itself.
Thus, this is definitely a yoloneering problem. The culture that the bug exposes is not that of the LLM, but of the yolo culture where actual quality is subordinate to marketing bullshit, in this case it is degraded quality inherited all the way from (post-)training data.
Let's delve deeper into this: This is not an artifact of RLFHF, it is a feature of the human brain. It is quite an enigma how LLMs elucidate the use of goblins
Are they though? I think the adaptive part is the memcached containing your "profile" that is running server-side w/ OpenAI and Anthropic. The model itself is deterministic in nature, which is evidenced by OpenAI's "goblin-fix" in codex:
Changing a system prompt means you're changing the runtime input, not the machine itself.
Thus, this is definitely a
yoloneeringproblem. The culture that the bug exposes is not that of the LLM, but of theyoloculture where actual quality is subordinate to marketing bullshit, in this case it is degraded quality inherited all the way from (post-)training data.Let's delve deeper into this: This is not an artifact of RLFHF, it is a feature of the human brain. It is quite an enigma how LLMs elucidate the use of goblins