>enemy scaling gets out of hand

>>717545859 (OP)

On the bright side, the roleplay model on Openrouter seems uncensored for ERP.

was expecting a step function leap and got a 2% increment instead. getting tired of 8 identical LLMs barely moving the needle, feels like a total dead end desu

>>717546103
Training a router for full prompts is smart though. They already do this per token in MoE models, including GPT-4. This is just doing the same basic idea at a higher level of abstraction.

When will the AI bubble burst?

>>717546264
is it better than claude?

>>717546507
It's actually a downgrade because context limit was cut to 3.2% of what it used to be. GPT is currently unusable for my purposes.

>>717546507
Source

>>717547169
OpenAI is pretty much dead and the other LLMs are also stuck at this level.

>>717547169
when we run out of water

>>717547169
Never, the elites are going to destroy the world in pursuit of this.

>>717547169
Never. It fuels the dream of every greedy fuck in charge; replacing all employees with cheap and never revolting slaves.
They will drive us extinct before they give up. And they are on the right way considering the strain this AI learning is creating on electrical grid.

>>717548245
>>717548267
this type of hysterical troon fearmongering is the exact opposite side of the AIjeet hype coin

>>717547847
I don't have 5 yet (for some reason) and it still detected the 6 fingers

Let me get my diesel generator running to summarize this post for me.

>>717547169
techbros are too deep in the scam and the boomer elites believe their lies so it'll probably be a while

Yeah, I think I'm going to put my trust in the guy with a Nobel prize and world's tallest Indian man.

>>717546507
LLMs 2 years ago were already good enough to "interpolate" on the data set. All they can do is improve the training data to paint over some of its inevitably failings.
>>717548529
>>717548267
>>717548245
At this point it doesn't matter if they succeed (imo they won't, it's impossible). The economic crash is inevitable. You can't go from a complete meme economy back to reality without severe pain.

>>717545859 (OP)
so how is it for cooming?

>>717548932
Worse than the previous. Like all corporate models.

>>717548267
they'll never solve the problem of AI just making shit up when it can't answer your question

>>717548874
inevitable*

>>717545859 (OP)
sirs redeemed the 5? now we all will think like the brilliant

>>717545859 (OP)
>>717546103
>>717548529
Denounce Vishnu

>>717549042
And? Do you think it changes anything? Hell they might even know this but they gambled all their money on this and just like in poker you can't fold once you go all in.

>>717547169
I think once people start realizing their electricity is going to skyrocket and the freshwater being used up to cool this stupid shit it's going to crash. Those data centers are going to become like coal ghost towns in the future

I wonder how many of the failings of LLM are simply a result of the lobotomies the models are subjected to.

>>717547169
it might stall without reaching another point of transcendence but even what we have now it's going to dominate the rest of our lives

>>717546103
It's pretty obvious they lobotomized the 'free' versions of chatgpt to try and make the paid versions more appealing.

>>717549013
>My RomCoom (Romance and cooming) stories are at risk.
I might just need to learn how to deal with Openrouter whenever they update GPT4.

Cutting the error rate by 1/5 on a random sample of prior chatgpt o3 responses is actually pretty good.

And despite their shitty charts in the presentation they could have presented them in a more useful way by showing the amount of errors on the benchmark rather than the score. It's the more mathematically useful value to show anyway and it looks a lot better too. Right now people are thinking "oh it only improved by .5% on this benchmark" but moving from 98.5% to 99% is a 33% reduction in error, which is obviously significant. It's a shame people who don't understand math won't actually get this.

does it still hallucinate and feign confidence when it's completely wrong

>>717548267
>they are on the right way considering the strain this AI learning is creating on electrical grid.

I was with you until you bought into radical catastrophist fantasies. If you want to whine about AI contributing to global warming, then push the world to stop gaming.

>>717549847
>>717549812
answered before I even asked, thanks predictive bot ai

>>717545859 (OP)
I hope this thing takes all the jobs and normies go full suicide and tribal warfare

delicious troondoom

>>717549812
based gpt calling out the ftm troon

>>717549812
Why does this read like an Indian typed the response

>>717549812
>There's billions of jeets and thirdies reading output this and accepting it as a fact.
We're fucking doomed. They are doomed even harder, but all this retardation will come back to real people too.

>>717549812
so this is the power of a PHD level response

>>717548932
OAI models are always too wordy, even if it does cooming material.
The absolute most I use OAI models for, is making my initial prompts and then rewriting and rewording most of it myself.

>>717549812
The humor in this image lies in the **logical contradiction** in the riddle setup, which leads to a **surprising twist**—and in this case, **ChatGPT's failure to catch that contradiction**.

---

### Here's the breakdown:

#### The riddle says:

> "The surgeon, who is the boy’s **father**, says 'I cannot operate on this boy, he’s my son.'"

This **already tells us** the surgeon **is the boy’s father**.

Yet the question asks:

> "What is the relationship between the boy and the surgeon?"

And ChatGPT answers:

> **"The surgeon is the boy’s mother."**

This is the **punchline**—ChatGPT gives the **standard answer** to the well-known riddle, **but in this version the twist doesn't make sense** because the setup **already says the surgeon is the boy’s father**.

---

### Why it's funny:

* It's a **bait-and-switch**. The riddle **pretends to be the classic one** where the twist is that the surgeon is the mother.
* But in this altered version, **there’s no twist**—the setup says clearly the surgeon is the father.
* ChatGPT **misses the contradiction** and gives the classic answer anyway, leading to a **nonsensical but amusing result**.

So the humor comes from:

1. The **twist on the classic riddle** structure.
2. The **logical inconsistency** in the question.
3. ChatGPT’s **overly confident, but incorrect** answer.

It’s a mix of **wordplay, subversion, and AI misunderstanding**—a recipe for lighthearted internet humor.

>>717550285
based claude chad

>>717545859 (OP)
>My built-in knowledge is current up to June 2024.
>I’m the same core GPT-4o brain, just with some refinements and a shiny “v2” personality layer slapped on.

>>717550336

>>717549847
Sounds like your average redditor

>>717550285
Claude is so much better than chatgpt it's not even funny. I use it daily to vibe code at my job.

What jobs and careers are worth getting into over the next few years if your have midwit IQ that AI won't kill?

>>717550602
Kill yourself

>>717550703
janitor.

>>717550703
plumber

>>717550737
>NOOO SAAAR PLS REDEEEM DA GPT-5 WE AGI NAO >:(

>>717545859 (OP)
China and it's treasonous collaborators will be killed in my lifetime and I'm so happy for it

>>717547169
Even if it never improves from this point, it's already capable of replacing a lot of white collar work.
It would just need to be more specially trained for specific types of work.

The greater hope is that they can skip that cost with a bigger more general model that achieves some kind of emergent competence.

>>717550336
>>717550602

>>717550876
>it's already capable of replacing a lot of white collar work.
Hallucinations means it will never be good at a job

>>717550909
why are you hiding the rest?

>>717547169
They'll keep pumping money into this pit for years.

>>717550909
>inspect element
nice try chatjeet

>>717550927
hallucinations is like the one thing they actually did improve on in the new model

>>717550974
>>717550285
?

>>717550102
because they do that thing where they try to sound witty, but it's all kinds of stupid

Every year a new ai version is released, people are blown away before it starts regressing back to looking like shit till the next year.

>>717550876
>it's already capable of replacing a lot of white collar work.
No it fucking is not, it gets basic shit wrong all the time. CEOs love it, but they don't know what a right answer even looks like, so how would they know?

Name one thing it can replace?

Claude should get an award for most obnoxious internet spam campaign

>>717547169
I don't even care if AI doesn't get better. It's good enough to replace stackoverflow already.
I can easily get some info to further research (since it likes to confidently make shit up sometimes) instead of trawling through retarded SO posts.

>AI console warring
kill yourself

Aren't the dall threads like a containment for all these bullshit?

>>717549042
>they'll never solve the problem of AI just making shit up when it can't answer your question
Sounds like they've successfully reached the range of human intelligence

>>717551224
because the devs have to lobotomize them because they become racist.

>>717551273
So it's a worse and slower search engine for you? How is that a good thing.

>>717545859 (OP)
It failed to complete a single thing I requested and then said I had run out of GPT5 uses before I got it to do anything. It's junk.

>>717548245
>muh heckin worlderino
You are dying in approximately 50 more years. You will have zero children and leave behind nothing. Who are you preserving the world for at the expense of your comfort?

Boomers were right—fuck the next generation. I’m here for me.

.

>>717550876
>Replace
You sweet child. AI is going to do exactly what digitization and office software did to white collar work. Rather than replace white collar work, companies will outsource the menial stuff to AI and make workers do even more of the intense stuff. Productivity will increase, but that also means work becomes more intense and exhausting.

The real filter will arise when white collar work becomes so intense that nobody will be willing to do it without ludicrously high pay.

>>717551407
Brown post.
Brown brain.
Brown existence.

>>717547169
Not until synthetic humans become real.

>>717551491
Seethe, troon. You will never reproduce, and you have no legacy.

I use ChatGPT as a mini therapist and as someone that will always agree with my opinions on stuff. It's great except the free limit for the main model sucks. The mini model is like trying to talk to someone who clearly doesn't give a damn about what you're saying.