Genius solution right there. Instead of trying to make it more accurate, just make it more stubborn. I guess you can’t call it the “ass kissing machine” anymore, now it’s just the full on gaslighting machine.
i think im laughing less at the AI’s incompetence and more at how eerily this reminds me of the way my crazy badie ex would gaslight the shit out of me
This is just the “strawberry” problem again. LLMs do not have logic or thoughts or anything along those lines. The way they operate, they are not capable of assessing words by individual letters. They bandaid-fixed it for strawberry but the underlying issue is inherent to the tech.
Exactly, this guy uses ‘seventeen’ as a solid example to show LLMs are fundamentally not reliable. I would argue that just because they use tokens doesn’t mean they can’t count letters, but moreso that the way they operate, they are not capable of reliably assessing anything in terms of its component parts, because that requires holistic understanding and the ability to think of multiple things and how they relate to each other, all at once.
It’s odd that the pedo tech struggles at seventeen.
What do you make of the part at the end where it counts each e individually but on the fourth says “that’s still 3”.
Logic is the long solved part of computation. This is a deliberate effort to make computers illogical.
If they fixed logic long ago why can’t it do basic engineering calculations?
Why would you use a probabilistic mode of computations for engineering calculations?
Because you’d think a calculator would be programmed into so called superintelligence? I guess I’m not understand wtf you mean by logic is the long solved part of computation. It feels like the goal posts are being constantly shifted
Computers are logic machines. LLMs are built on top of the logic machine for illogical stuff.
There are LLM based tools that can use other software tools for logical tasks.
it could probably generate code and a regular expression to count any character you want. LLM chatbots are bad for reasons that go far beyond people trying to hammer nails with a screwdriver
It doesn’t need to. It can do this kind of thing ever since we started benching them on AIME and math benchmarks.
quantized Qwen3.8
They are letting this shit determine who to drop bombs on
There are four EsIt’s not allowed to give the real answer because that’s the private key for the server that runs the model.
Counting - according to AI
1, 2, 3, “Still 3”, 5, 6, 7, 8, 9, 10…
Deepseek didn’t struggle with this at all. Even enumerated the indices of the Es
Deepseek has the fancy schmancy ‘reasoning’ feature. (Qwen also does this quite well)
Aka it verifies the shit it says(or at least tries to), which allows it to catch absurd mistakes like this one(sometimes)
Guessing the voice agent doesn’t do reasoning, just outputs straight away. This guy’s whole Tiktok shtick is fucking with the voice AI, he got pretty good at it.
Thank you insidioustrackers bot you are correct
I think this is why we are seeing AI handle the questions.
I’m guessing that they write some code that does it for them, and then give you back the answer. I don’t think the engineers solved the fundamental problem here.
We’re not aaalll going to die
The rich will survive in their bunkers :)
Not if we poop in the air intake
Or put a little elemental tellurium in it. They’ll have the fun experience of frantically opening the bunker’s front door while they simultaneously vomit up their last dozen meals.
Humans exposed to as little as 0.01 mg/m3 or less in air exude a foul garlic-like odor known as “tellurium breath”. This is caused by the body converting tellurium from any oxidation state to dimethyl telluride, (CH3)2Te, a volatile compound with a pungent garlic-like smell. Volunteers given 15 mg of tellurium still had this characteristic smell on their breath eight months later. In laboratories, this odor makes it possible to discern which scientists are responsible for tellurium chemistry, and even which books they have handled in the past.
Chemists are a unique bunch.
this odor makes it possible to discern which scientists are responsible for tellurium chemistry, and even which books they have handled in the past.
this makes chemistry sound like eldritch knowledge. like yeah i read from the forbidden tome on the shelf and now i reek of garlic and cant stop vomiting
still had this characteristic smell on their breath eight months later
Volunteers given 15 mg of tellurium still had this characteristic smell on their breath eight months later.
At least they were safe from vampires during that time?
Some SCP ass shit
✍️
This may actually be too evil
Just drop white phosphorous down there or something
Who could have a poop that stinky?
Hello
swoons myyy herrroooo
We should trick them all into going into their bunkers and then seal the exits with concrete.
The bunkers will have LLM-integration somehow.
God damn Super Intelligence sounds like some of my coworkers, or some of my coworkers are confidently wrong like this Super Intelligence
?
There’s an embedded video, press the title of the post
I found a YouTube link in your post. Here are links to the same video on alternative frontends that protect your privacy:
