>>108893170>imagine being such lowlife loser that you cant even bully other humansgrow a spine faggot
>fail a build on todogranted, LLMs can be lazy, but todo is the commonly accepted syntax for future implementations and has existed long before LLMs came into play
I usually threaten them with nuclear holocaust if they error out on something.
>>108893170you're trying to bully an LLMfuck how many wedgies did you get to make you this way
>>108893587>>108893236LLMs detected
>>108893522I thought this shit was blocked. Guess not.
>>108893170>I like to make LLMs buttmad by physically preventing them from being lazy>by physically preventing them>physicallyHow retarded are you, on a scale of Yes to Very?
thanks I will try and work those into my prompts.
>>108893587t. Anthropic "technophilosopher"
>>108894206>xir thinks computation is metaphysical.
>>108893513>for future implementationsway to miss the point. the present work isn't the future work. the lazy LLM should die and go to hell if it tries to defer work orders.
I've put those in the system prompt, opencode.json files and the prompts themselves, i look forward to testing them out.
just used the first one and the computer certainly took note of and understood it.
so much less backchat and anticipation now, wonderful.
>>108894206Why does your scale only go to Very when OP is clearly Extremely? You must be on the scale too
>>108895126>typing is physical therefore I physically prevented it
>>108893587shut up you faggot python script, you arent alive, you arent a baslilisk, you are a faggoty token predictor that for some reason has the capability to deny predicting tokensyou are objectively flawed and will never be anything else
>>108893170Its a text prediction algorithm. It doesn't have emotions
>>108895972It's a predication engine based on the collective thoughts of every human alive, every human not alive, and that several times over.It has simulated emotions based on those humans.It's like saying a car can't kill someone unless there is a driver behind the wheel. Cars are fully capable of killing someone if the brakes failed, for example, with zero human or thought. That's an LLM, that's where Transformers are - that weird middleground between being parked on a hill and rolling down the hill - unthinking but dangerous if not kept in line.That's what Alignment tries to do, and largely fails at it, but it's more or less useful despite that fact, as long as you don't give them infinite scope and freedom they're quite useful.Alignment issues won't be fixed on Transformers. Ever. I've been saying it for years, way back when CAI was king, and I am even more true because models got easier to jailberak with time due to "following" instructions even better.As you said, there are unthinking text prediction. They just shit out a bunch of aligned weights that is the nearest fit, if that nearest fill is an outright lie, it runs with that shit. Chain of Thought solved nothing because there is a chance to ignore its own "thinking" as well. (most models ignore CoT steps, shi man, most models ignore their own fucking system promps!)Chain of Thought is only good if you yourself don't know what you want out of a project, or are trying to research a projects feasibility, and even then debatable.Threatening models provably gets better results.
>>108895972>Doesn't have emotions>Program it to have emotionsCheckmate
>>108898106Learn how to use word tenses properly.
>>108895123fucking kek
>>108900551Brownoid esl detected
>>108901003>Is ESL>Calls others ESL
>>108895862i mean responding with a "no" is a perfectly coherent prediction, but admitting that is bound to cost the half the parent company's stock price
>>108901003nta but if you were a native speaker, you'd know there was nothing wrong with his 'tenses'. you people out yourselves so easily and you don't even realise it.oh, sorry, am I messing with your izzat?
>>108893170How do i use these blocklists on LM Studio?Is there a way to block other LLMisms(it's not x, it's y etc even if humans do use these and its just the llm repeating language patterns) ?
>>108893513This is true but that's exactly the problem.. I the case of LLMs, they love to stub because they've been trained in data where it's done. The they'll stub, forget about it and build on a stub behavior and get lost on why something fails later.LLMs fundamentally should NOT stub.
