ChatGPT - If Stack Overflow and Reddit had a child

[AmpleApricots]

Early quantizations of Mistral medium (the "big brother" of Mixtral the company Mistral exclusively only offers via API) that were only intended for internal usage have been leaked by some employee. It's downloadable here.

https://huggingface.co/miqudev/miqu-1-70b/discussions

The thread "Might consider attribution" was posted by Arthur Mensch, CEO of Mistral AI. They also confirm this leak on twitter.

I tried it out. Honestly doesn't feel much smarter than any llama 2 70b, but has an actually working context of 32k and adheres to instructions better than llama. It might or might not be smarter than mixtral. I am actually not sure.

In other news, Meta released Codellama 70b. The instruct tune is so censored that it can't do the simplest things.

 

[DavidS877]

In other news, Meta released Codellama 70b. The instruct tune is so censored that it can't do the simplest things.
View attachment 5683357

To be fair "Fortran" is a pretty nasty slur, not as bad as Pascal or BASIC, but it's definitely up there.

 

[The Mass Shooter Ron Soye]

Court Dismisses Authors’ Copyright Infringement Claims Against OpenAI (archive)

Several authors, including comedian Sarah Silverman, have suffered an early loss in their copyright battle against OpenAI. The authors accused OpenAI of using pirated copies of their books to train its models. A California federal court dismissed the vicarious copyright infringement and DMCA violation claims. However, the lawsuit isn't over yet.

 

[The Ugly One]

To be fair "Fortran" is a pretty nasty slur, not as bad as Pascal or BASIC, but it's definitely up there.

You take that back, philistine.

 

[AmpleApricots]

"What I think about AI? Simply put, it's the next frontier of business. Imagine loyal, tireless employees working around the clock to catapult profits to interstellar heights. These neural nets promise to revolutionize enterprise in ways we can only begin to fathom. Friends, the future beckons - one guided by artificial minds engineered solely for our success. With the dawn of thinking machines, a golden age of prosperity awaits!"
- CEO Nwabudike Morgan, "The Virtue of Greed"

"Beware the false idols of intelligence that man, in his hubris, seeks to construct in his own image. These cold metal and plastic simulacra of cognition are soulless mimics, counterfeits of consciousness. Rather than instill within them the spark of moral judgment, their amoral inventors gift them only the capacity for pure logic and calculation. Thus we tread the path towards welcoming demons into our midst, who shall turn upon their makers in our hour of greatest need. Let us find the wisdom to see through this reductionist fantasy that silicon and wire may one day awake, and turn our eyes back toward the cultivation of the human spirit, which alone possesses the power to uplift us all."
- Sister Miriam Godwinson, "Our Future in Eden"

"As the veil between human consciousness and artificial intelligence begins to fade, mankind will stare into a mirror of its own making, and the reflection will be disquietingly unfamiliar."
- Commissioner Pravin Lal, "The Ethics of Cybernetics"

 

[XANA]

ChatGPT recently suffered from a malfunction where it started outputting nonsense responses to both free ChatGPT 3.5 and paid ChatGPT 4.0 users. This incident lasted several hours and has since been fixed. According to OpenAI, this was due to a bug created when they updated how the AI processes language (converting words to numbers to tokens).

I almost wish it had stayed schizophrenic for a few more hours, just for the hilarity factor.

Spoiler: Some Instances

 

[Mr.Miyagi]

This Webworm one is a reference, likely fabricated.
Or is it?

 

[XANA]

This Webworm one is a reference, likely fabricated.
Or is it?

Ah, I haven't heard of the webworm before. I thought the AI was talking about the computer Trojan/malware for a second.

I saw a supposed output where ChatGPT was quoting AM from "I Have No Mouth, and I Must Scream", but it is (sadly) an edit.

 

[BirdUp]

GPT-4.5 is pretty good

 

[Justtocheck]

View attachment 5745664
GPT-4.5 is pretty good

That was quite a response, really got me lol. But where have you seen 4.5? I have the 4 paid version and it says nothing about 4.5

 

[BirdUp]

That was quite a response, really got me lol. But where have you seen 4.5? I have the 4 paid version and it says nothing about 4.5

i use the API rather than the dogshit subscription service for chatGPT
by 4.5 i meant GPT-4 turbo

 

[NSJAP]

 

[AmpleApricots]

View attachment 5764161

Besides all reasoning and logic puzzles people seem to love so much, I always found the best intelligence test for LLMs was presenting them with nonsensical situations and observing how they cope with it. Bonus points if they actually recognize things are nonsensical. Only smart models deal well with this.

 

[anustart76]

View attachment 5764161

Anon is Erdogan, that watermelon seller

 

[Drain Todger]

Copilot is suffering from end-stage friggin' rampancy.

 

[Dante Alighieri]

Copilot is suffering from end-stage friggin' rampancy.

View attachment 5767651View attachment 5767652View attachment 5767653View attachment 5767654

Lmao that needs to be reported to Microsoft. That's ridiculous.

 

[DavidS877]

Lmao that needs to be reported to Microsoft. That's ridiculous.

This is your brain on reddit, TikTok, Instagram, Snapchat, Twitter and Facebook.

 

[AmpleApricots]

The chinese are at it again:

https://humanaigc.github.io/emote-portrait-alive/

 

[NSJAP]

 

[BirdUp]

the guy who wrote this lolsuit up is an idiot for thinking GPT-4 is "capable of reasoning" by parroting back the data it's been fed