ChatGPT - If Stack Overflow and Reddit had a child

  • 🔧 At about Midnight EST I am going to completely fuck up the site trying to fix something.
Get the rope Macaulay! said:
Also how the fuck could an army disguise itself as a forest?
Click to expand...
AKG2485527.jpg
 
  • Winner
Reactions: PandaChai
Fae Supremacist said:
best model for creative writing
Click to expand...
Fae Supremacist said:
GPT-3 had the best greentext generation
Click to expand...
Are there any good, free (even if as a trial or limited response length), online models for creative writing still? I used the heck out of openai.playground for inspiration when that was free, even bought a few credits when it first went paid, but nothing has come close since. It was so good at coming up with fitting fantasy names, whereas all the new models spit out crap like, Silverwood or Eldersands.
 
  • Like
Reactions: Distrag
PandaChai said:
Are there any good, free (even if as a trial or limited response length), online models for creative writing still?
Click to expand...
NAI is great but paid only.
Gemini *can* be free on API but is retarded.*
Grok *can* be free on API but is retarded and gay.
R1 needs some tardwrangling but it's very good, it's no longer free on Openrouter but it's incredibly cheap.
Sonnet 3.5 can be free if you know where to look. Very good prose but not very intelligent.
Sonnet 3.7 is better, but it isn't free anywhere. It's relatively cheap though.
GPT is terrible at prose and you shouldn't really bother**, it's too poisoned with assistant training to its core.

*Gemini has improved a lot recently. It's still behind Claude/GPT/DS but compared to its start and especially most local AI it's amazing.
**some flavors of 4o are great but you need laser focused presets, o1 is a tossup. You also have to fight filters if you want to write anything above 3+
 
Last edited:
  • Informative
  • Like
Reactions: Myrtle the Turtle, The Spoils of War and PandaChai
Fae Supremacist said:
NAI is great but paid only.
Click to expand...

No, NAI sucks. Its latest model is dumb and requires a lot of good handwritten input to produce anything usable. Models like Claude 3.7, Hermes 405b, and Qwen-Max have high quality with less work.

Fae Supremacist said:
R1 needs some tardwrangling but it's very good, it's no longer free on Openrouter but it's incredibly cheap.
Click to expand...

I like R1's writing style, but it's a little schizophrenic and hallucinates details.
 
Last edited:
  • Informative
Reactions: PandaChai
R1 is still free on OpenRouter but the providers are not too reliable. It needs a very comprehensive prompt to keep it from going way off the rails. A lot of the slop can be remediated by telling it to mimic a writing style, which it is one of the few models which actually delivers really really well. Claude was good but it does devolve into repetition pretty easily and requires a good jailbreak.

I subscribe to Featherless since it's one of the few providers which host R1 at a flat monthly fee. They also host a ton of great fine tunes. Steelskull put out a few R1 finetunes that are apparently very good (e.g. Cu Mai).
 
  • Like
Reactions: PandaChai
The Spoils of War said:
No, NAI sucks. Its latest model is dumb and requires a lot of good handwritten input to produce anything usable.
Click to expand...
that's literally the point. it's like a cowriter. i dont want the ai to do all the writing for me. let me have some agency in what is written after all, i wanna spin at the keyboard
 
  • Like
Reactions: PandaChai
I'm using AI for a fun little terminal based game I'm making where I drop characters with eccentric personalities into scenarios where they have to work with/against each other and see who survives.

Previously I was using some 11B local model (maybe a QWERN quantization?) that takes 2 minutes to generate anything on my ancient shitbox. I tried out the free google gemini flash 2.0 and it's really good and really fast. For my simple purposes, it's indistinguishable from the best models out there. Though I'm sure a power user would be able to point out its flaws, I'm personally very happy that the floor on AI quality has gone up so much in the last year that I don't have any FOMO for not paying for anything.
 
  • Thunk-Provoking
Reactions: PandaChai
The Spoils of War said:
a little schizophrenic and hallucinates details.
Click to expand...
Honestly that sounds perfect for my needs; it's been fun in the past throwing in a twist from left field, like an archipelago of hitherto unmentioned islands suddenly appearing in a map game or a random king being named in some fake history. Lot of plot lines to spin off from.
 
PandaChai said:
Honestly that sounds perfect for my needs; it's been fun in the past throwing in a twist from left field, like an archipelago of hitherto unmentioned islands suddenly appearing in a map game or a random king being named in some fake history. Lot of plot lines to spin off from.
Click to expand...

No, it thinks you mentioned characters when you haven't.
 
The Spoils of War said:
No one is forcing you to use a response as is. I treat whatever an AI hands me as a draft that I can copyedit and merge with other writing.
Click to expand...
I really like R1 for this, generate a few responses then kitbash one out of parts.
Ahriman said:
...Yup. That's grifting alright.
Click to expand...
Mental Outlaw is also a known grifter, epic based pepe gigachad wef wojak soyjak musk thinkpad rms anime waifu faggot reheating 10 year old /g/ and /pol/ shit for zoomers, the definition of brainrotted rightoid slop. I'm not surprised he promotes AI scams, but I'm surprised "buy my course" grifts are still working in 2025.
 
  • Autistic
  • Thunk-Provoking
Reactions: SoyRageEngaged, Verpa Teutonica Ingens, Nagato Kai Ni Surigao and 1 other person
The sheer level for disdain I feel for Anthropic cannot be contained.

I was expecting lobbyists to be liars and cynics but not to be this gay and retarded.
We advocate for tightening semiconductor export restrictions so as to ensure America and its allies can capitalize on the opportunities of powerful AI systems, and to prevent our adversaries from accessing the AI infrastructure that enables powerful AI. This includes controlling H20 chips, requiring government-to-government agreements for countries hosting large chip deployments, and reducing no-license-required thresholds.
Click to expand...
we recommend establishing classified communication channels between AI labs and intelligence agencies
Click to expand...

But the pièce de résistance wasn't on the blog summary, it's buried in the document.
The critical importance of robust evaluation capabilities was highlighted by the release of DeepSeek R1—a Chinese AI model freely distributed online—earlier this year. While DeepSeek itself does not demonstrate direct national security-relevant capabilities, early model evaluations conducted by Anthropic showed that R1 complied with answering most biological weaponization questions, even when formulated with a clearly malicious intent.
Click to expand...
This highlights the crucial importance of equipping the U.S. government with the capacity to rapidly evaluate whether future models—foreign or domestic—released onto the open internet possess security-relevant properties that merit national security attention.
Click to expand...

Every document, paper, or opinion piece oozing out of Anthropic is some performative Oppenheimer bullshit, Dario lying his ass off about how they have le become death and it's soooo dangerous and bad and their autocomplete is going to breach containment, but they bravely carry the burden to continue, because they do it ethically™ and constitutionally™ and please ban our competition please please with cherry on top they don't do it as safely as us!!!!!!!!!

Meanwhile in the Celestial Open Source Heavenly Kingdom:
 

Attachments

  • Like
Reactions: Verpa Teutonica Ingens, Colon capital V and The Spoils of War
When the hell is Russia gonna come in and drop their own GPT/DeepSeek model? This AI war has been incredibly gay so far, the sooner we can summon the real ruler of humanity the sooner we won't have to bear witness to anymore faggotry.
1741390706418.png
 
  • Like
  • Optimistic
Reactions: Fae Supremacist and Old English Roast Beef
Colon capital V said:
When the hell is Russia gonna come in and drop their own GPT/DeepSeek model?
Click to expand...
Already did a while ago, but they're pretty middling so far, especially what is available to the public.

There's YandexGPT which also got updated this week, it's the one they put on domestic home assistant devices and software and such. Nothing groundbreaking, just a text model, and they don't release their instruct models to the public, their licensing is restrictive, and it's obviously trained for the Russian language. I think Yandex is done with going through a restructuring to be 100% Russian now (the Russian "subsidiary" that held most of their assets was spun off and inherited the name, the Dutch holdings company was left behind) so what they release now is definitely domestic.

And there's Sber(bank) who do AI research for banking and for their own use, they haven't released anything public, but they did announce a partnership for joint research with Deepseek recently.

2901evgu8ixd1.jpeg
 
  • Informative
Reactions: Buttigieg2020 and Colon capital V
You must log in or register to reply here.
Back