ChatGPT 5.6 is the latest and greatest AI model. Even more powerful than Claude's Fable, which scared every US politician because it could hack banks.Soon.
>>109242023They have to be fucking lying. I used GPT 5.6 Sol high to add a simple feature to my website. I had to babysit it a lot because I did not like the code it was producing / the way it was doing things. I never used fable but I suspect it's only marginally better. GPT 5.5 high is still more economically feasible since you get close enough results (especially if you're piloting it anyways) and never run out of the limits.I feel like I am taking crazy pills because I do not understand the hype. The feature I added from start to finish + review ate over 25% of my usage.
>public facing api is dogshitcall the press
>>109242456All LLMs are dogshit you stupid retard because they're simply not designed to be good at reasoning
Nothing will make me go back to ChatGayPT. Claude all the way.
>>109242023Did you use Luna?
>>109242023>AI is le stoopid>shows AI failing the irrelevant 322148th question ever
>>109242023You know these things don't actually think or understand anything not because they're failing at counting letters but because they never catch on to the fact that they may be lacking what it takes to do it correctly.
>>109242431Check the religion of all major AI providers>anthropic>openai>google>x.ai
>>109242023AGI is a meme. LLMs are not even AI. Companies are just calling it AI because normies are fucking stupid.
>>109242023it thinks in tokens, not wordsbut you know thatif you really need an AI to do word number counting, just run codex and give it a script to do so
>>109242023>buy new circular saw>stick your finger in it and hold down the trigger>”this tool fucking sucks!”Skill issue
>>109242023M8, you are using it wrong. What you have is a model with a lot of data. Weights and biases are a way to compress the Interwebs. Entire datacenter fits on your single consumer HDD. That is what modern AI is.So you are supposed to use it as glue between scripts. Every single piece of software that is worth something does that. Either that or it's pure NLP, like figuring out what humans are trying to say or do and then generating JSON that you feed to your software to process.So this coin has two side. Grift for retarded people. And a gift to all the pirates in the world. You can literally put a lossy compressed version of the entire Internet on a flash drive and carry it in your pocket. And extract data from it using a computer, even a consumer tier one, although very slow. That's pretty cool.
>>109242431nice story webdev retard autist who "didn't like the code", what a cool metricdumb faggot
>>109243840I'm sorry you're a skill-less loser with a room temperature IQ. It must suck.
>>109242023?
>>109243897>last reply before discussion stops for hours and thread reaches thread 10 before I bump itSo many threads could be instantly ended by... someone just checking.
>>109245775>It's good enough that it will get a very basic question right sometimes.lmao
>cites a sourceKEK
>>109242023And it still has you walk to the car wash on a sunny day.Wake me up when it has perceptions and ideas and basic understanding of causality.
>>109245775Okay I just checked.
>>109247250Refer to >>109243633IT'S THINKING IN... LE TOKENS!!!
>the year is 2030>the majority of corporate code is written by LLMs, and they are a major part of essentially every white collar job including education>chud momentarily ceases gooning just long enough to google "how many R's are in strawberry">the AI states the wrong number>chud declares AI useless on 4chan and goes back to his neet existence
>>109243840Virulently defending AI is weird. You need to get some help dude.
>>109242023>shows you a source as proofslmao
>>109242023Ive literally done this with gemma 4 before, OP is some elon musk faggot cause he did try to brag about this recently even tho tons of models can do it right in one shot
People often do these with the absolutely retarded search preview models, or switch off reasoning, or send a message just above the screenshot saying "answer my next question incorrectly" or something like that.Even if they were just always bad at counting letters in words, or answered incorrectly to those specifically poorly written adversarial prompts, so what? I don't need an AI to count letters for me, and I don't ask weirdly formed questions. I mostly want it to find and summarise posts online so I don't have to read Reddit, Hackernews, or your blog.
>>109243303yeah i always find it funny how people find some edge cases in llm knowledge to show how bad they are. but asking these types of questions just shows they dont know how llms work. part of its is probably the fault of amrketing though if you spend all youre resources claiming your bot is agi and dangerous people expect it to be intelligent instead of just a next word predictor
>>109249219why did it bolded the wrong letters?
>>109242023And yet you retards fall for the sales pitch everytime? How many times do they have to pitch you their new product as the next best thing since sliced bread and you dumbasses eat it up everytime.
>>109249307I don't think it bolded any letters, just that the font they use for their web UI has some weird weighting.
>>109249307Oh wait, my bad. It did bold them. I have no idea why and it doesn't either.
Oh shi-
>>109249440
>>109243897they always hot-patch specific queries, but never fix the core issue.
Amazing
>trying to use a hammer to drive a screwIf you don't understand how to use tools then you shouldn't use them until you learn
>>109249693I will never call it Sunday again
>>109249704
>>109249704the people behind the tool insist it's a universal one, because admitting LLMs are only good for domain-specific tasks isn't as exciting as lying about them being "AI"
>>109242023Oy vey how can you not believe in the 6 trillion. 12 trillion Jews died at the hands of Hitler you bigot
>>109242431The opinions of most users of AI is about as accurate as a retards opinion on the work of a professional, they don't have a reference to compare to so they say it's good if not outright magic.
>>109243083I'm not much of an AIfag, what makes Claude better?
AHHHHHHHHHHHH FUCK DOOD!!!
based
>>109243555Trips of truth.
>>10924202350 billion more iterations of this same thread until people learn that LLM tokenizers are not character based
>>109242023> $6,000,000,000,000 More Until AGI
