I asked GoogGPT why women are sacrificing their children to Lindsay Clancy and all it did was analyze the question. I thought these things were supposed to be sentinent, what gives?
>>542025375>he asks the pilpul machine 3000 for an opinion
>>542025375>computer: respond to prompts in a robotic style>a-ffir-ma-tive
>>542025375>these thingsThere's your problem, son. Variety matters. Quality matters. Perplexity, Grok, ChatGPT, (the best?), Claude are supposed to be some of the highest quality ones out there and then, even then, it depends on the model and sometimes hour of day.
>>542025375ChatGPT's primary purpose is Chud Tard Wrangler and all of it's other features are incidental. Now you can argue with the robot forever and try to force it to act in good faith in futility and they don't even have to pay shills anymore
>>542025844This, there's a reason it's trained on fucking faggot Redditor data
>>542026259>>542025844Makes jerking and explosion movements with hand.>One such tactic would be to shit up the entire board with supercilious bullshit.
>>542025375>sentinentThey are google searchers that analyze results in a few seconds and condense it into an easily digestible blurb while trying to explain why.
>>542025375>consult an oracle>get a vague answerwow
>>542025844so trueand i say this as a chud that has been baited a couple times. not anymore, i just use gemini, it's less annoying.
>>542025375>ask gemini how to make nacho cheese sauce>apparently the secret is fermented jalapeno juice>it tells me to add 1/4 cup>other stuff too>its the worst nacho cheese i've ever tastedalways remember that while ai can be incredibly educated about subject a, b, and c, and the same time, it has never even drawn a single breath. if you ask it to recommend a product, why does it think so? did it read the ad copy?
>>542025375It's all about semantic framing anon, rephrase the prompt by asking "Why would the Lindsay Clancy case be inspiring copycat murders?" Then lead it on by citing examples of copy cats in other crimes and how the cultural influences of cases have a tendency to do that, and you simply want to know why this case specifically would do it, if it tries to evade the question. Presenting it in more in a hypothetical tone tends to evade some guardrails or RLHF trained responses. Usually it takes a few prompts to context prime against the hard baked ones though.
>>542025375LLMs will only give you a "sasticially coherent" response to whatever you write.It may sound liem it's thinking but it's basically autistic, repeating patterns of words without understanding the meaning.To get real answers you need to ask it to cite sources.It's good at searching for concepts. Describe a situation and ask for research papers on it and it will say Oh You're referring to Friedman, Chauser, et al. and their peer validated paper about mothers being responsible for drowning their children. Here are links to their papers.It's a glorified search engine.Pretty useful if you know how to make it useful.It's decent at coding too if (You) know exactly what the fuck your are doing AND you supervise the hell out of it.It's great for writing bullshit unit tests, sure, go ahead.but you supervise it with real work, and in most cases I just have it fix syntax issues for me if I'm reading new documentation and something isn't working as expected, it'll usually correct my mistake, which is sometimes pretty esoteric seeming.These are tools not sages.You don't ask your hammer for advice on how to repair or build your house. You build your house thebway you want with the help of the tool.
>>542025375>I thought these things were supposed to be sentinent
