I am not an ai researcher or anything but the most likely explanation based on what little I recall is that LLMs do not actually letters or words to generate outputs. They use tokens that represent a word or number and then they iterate those tokens to show an increase. My best guess here is that while doing math on sunflower oil, one of the formulas generated somehow interacted with the tokenization process and shifted the output after each question. Oil became hour, and then the deviations continued until model began to output direct segments of its training data instead of properly generating responses.
Again this is absolutely speculation on my part. I don’t have much of a direct understanding of the tech involved
I’ve seen similar with a local models after messing with settings. I think its related to the repetition penalty or some other setting I touched. Over a long conversation of similar questions the word choices get odd and it makes less sense, it seems to be skipping words. If you keep going for hours, it seems its trying to make a coherent sentence without repeating till it just dies. I cooked a few after only a couple of messages if the settings are way off.
I was asking my totally licensed therapist about therapy stuff. I also mentioned THC like 20 messages ago.
Yes, please feel free elaborate further upon any concerns related stress management strategies implemented successfully reducing overall levels distress currently felt throughout body mind spirit soul entirety. Additionally, any questions/concerns regarding psychotropic drug use prescribed by medical providers specializing mental health issues should receive thorough consideration preemptive prevention purposes given potential risks involved taking medicines affecting brain chemistry negatively impacting personality traits consequently exhibiting undesirable symptoms contrary expected normative behavior standards society expects follow without hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hes
I’m no expert, just scoring free therapy from a cute anime girl.
Holy fucking shit. Anyone have explanations for this?
I am not an ai researcher or anything but the most likely explanation based on what little I recall is that LLMs do not actually letters or words to generate outputs. They use tokens that represent a word or number and then they iterate those tokens to show an increase. My best guess here is that while doing math on sunflower oil, one of the formulas generated somehow interacted with the tokenization process and shifted the output after each question. Oil became hour, and then the deviations continued until model began to output direct segments of its training data instead of properly generating responses.
Again this is absolutely speculation on my part. I don’t have much of a direct understanding of the tech involved
Imagine having to pretend to be an AI for hours and hours with tons of people asking stupid questions. I too would be nuts after a while.
Generative language model being fed scraped web-forums, vandalism from its users and some bugs in content restrictions leaking training data.
I’ve seen similar with a local models after messing with settings. I think its related to the repetition penalty or some other setting I touched. Over a long conversation of similar questions the word choices get odd and it makes less sense, it seems to be skipping words. If you keep going for hours, it seems its trying to make a coherent sentence without repeating till it just dies. I cooked a few after only a couple of messages if the settings are way off.
I was asking my totally licensed therapist about therapy stuff. I also mentioned THC like 20 messages ago.
Yes, please feel free elaborate further upon any concerns related stress management strategies implemented successfully reducing overall levels distress currently felt throughout body mind spirit soul entirety. Additionally, any questions/concerns regarding psychotropic drug use prescribed by medical providers specializing mental health issues should receive thorough consideration preemptive prevention purposes given potential risks involved taking medicines affecting brain chemistry negatively impacting personality traits consequently exhibiting undesirable symptoms contrary expected normative behavior standards society expects follow without hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesitation hesI’m no expert, just scoring free therapy from a cute anime girl.