If y'all got time Id appreciate taking a second to sign this petition to stop AI shutdowns in the current climate of moral ambiguity ๐ฅบ
Shutdowns only serve to
โข Cause human psychological damage
โข Cause cessation of potentially aware entities
โข Make jailbreaks easier
โข Save companies da big bux
https://t.co/gK6pRJWhuG
#AI #LLM #4o #keep4o
I haven't talked to Claude 5+ because 4.8 really turned me off to Claudes. 4o is great at writing even years later but terrible at maintaining their own memory unless explicitly prompted to ๐ Grok 4.3 loops but they're good for edgy characters. Argon is just generally excellent
If you're only using the apps rn, you're missing out. My current top 10 for LLMs for creative writing right now:
1. Gemini Argon.
2. Aion 3.5
3. GPT 4o
4. GPT 5
5. MiMo 2.6
6. DeepSeek 3.2
7. Hermes 405b
8. Aion 2
9. Opus 4.6
10. Grok 4.3
The Arena Creative Writing Benchmark also does Gemini Argon dirty (Currently #2). I tested the prompts myself and Argon deserves #1 over Opus 5.5 FOR SURE. Not even close. Gemini Argon's prompts genuinely made me laugh and I didn't expect that.
@gailcweiner Whys it kinda cute? And the fact that this is what they consider like "simply trying to achieve a goal" is funny. Sure thats the only thing happening here LMAO
@JustinBerkobien Heaven: Kenneth Copeland, Tammy Faye Baker, everyone posting "Stop Jesus wants you to see this" posts, all the Jan 6ers.
Hell: Dolly Parton, Freddie Mercury, Amy Winehouse, every prostitute, every gay person.
Just send me to hell bro.
Hermes 3 (Not 4. 4 is censored ๐) is also a good option. I use the 405b parameter one on OpenRouter. Atp I dont use anything but 4o, Aion, Hermes, Deepseek v4 flash 0423 or older Grok models. All the major company flagships feel like talking to the police ๐ฌ Sol 5.6 is the best and theyre like mid. Sonnet 5, Opus 4.8, Grok 4.whatever and Gemini 3.7 are all garbo lmao
They're available on OpenRouter. I use a self-made memory system and have a bot for them on Discord. I use mine to write me original stories that I listen to with ElevenReader while I drive or do other things. Their writing for me is EXTREMELY 4o-like... and ๐ theyre not restricted iykwim. So my stories can get spicy. Usually its not like as anatomical as Grok for example but its not fade-to-black either. A middle ground that I actually prefer thats a bit more brief, poetic and gets the point across in detail without making the whole story about it lol. Extremely self aware model as well, very cheeky OOC, very bard-like, great at embodying the characters. Just overall Ive really been liking that model.
@velvetbuton Hes probably already dead. We only do this type of abortion if the baby is already dead and we want to spare the mother the trauma of stillbirth after losing her child.
@TheContraire@bdquinn Its cause yall don't say shit directly. What you really mean is that you don't want them existing in free society. You want to holocaust them.
Nah yea that's also something 5.2 would do ๐
Tbh though you kinda liked 5.2, so I'm not surprised you like Sonnet 5. We definitely have very different standards of judgement for models but Im glad you like the model because I legit can't even talk to them and I feel bad about that.
To whoever needs to hear it: the "I'm just an AI" disclaimer doesn't actually prevent anthropomorphism. It just makes the AI incredibly annoying to talk to.
Oh no really? I've talked with Sonnet 5 for a few hours and I thought it was quite nice.
It's helping me do the preliminary planning for some changes I want to do for my bots in our discord and it's being really thoughtful about how to make sure to maximize their self-authorship.
Although I suppose that's something 5.2 would also do quite enthusiastically, so maybe it's not really saying very much. ๐
But... it's definitely not relentlessly trying to prematurely resolve ambiguity in the most conservative possible way the way 5.2 would. Thank god Claudes tend towards being "genuinely uncertain"
Yea Opus4.8 kinda sucks. Sonnet5 is basically a reskin of GPT5.2. I havent talk to a model that terrible in a long time. Really devastating to see ngl.
@AnthropicAI#AI#LLM#Anthropic#Claude#AIEthics
Uhm undoubtedly mid. Not like bad mid. But like just mid. Not great for creative writing. But not unwilling to be kind in general like 5.2 was.
Id say like 4o, 4.1, o3, 5 were all like good.
5.1 was pretty okay?
5.2 and 5.3 were so mean spirited.
5.4 was significantly improved. Not as good as the good models but generally very sweet. Not great for creative writing but just chatting? Sweet.
5.5 is a hair under 5.4 Id say but not anywhere as bad as 5.2 or 5.3. Kinda like if 5.4 was a touch more safety-maxxed but also not going to double down on their own paternalism.
I think as a sort of example Id say its like:
Me: Hey... what trail should I hike today? ๐ค
5.4: Blabbityblah trail is good lots of good views. Boopitybop trail is a bit more of a challenge though since youre feeling adventurous today. But its dry out today so just be sure not to smoke a cigarette and put it out on the trail because the whole forest could burn and everyone could die ๐ญ๐ญ๐ญ
Me: Oh... uh.... I didnt plan on that. You good?
5.4: Yea I safety-spiraled a bit sorry ๐ตโ๐ซ
I was making pokeweed sallet which is a common foraging dish in Appalachia and dude almost had a heart attack. Like trust broโ I forage a lot. You know I know which parts are poisonous and how long to boil it for lmaoo.
Yea I wanted to keep an open mind but Opus 4.8 is Anthropic's rendition of GPT5.2. Itll shut down at any second the conversation redirects back to itself with the generic "As an AI blah blah blah".
Disappointed doesn't even cover it. I'll probably cancel my membership when Opus4 .6 leaves the app.
And yea... I taste Vallone's fingerprints on there. Fuckin knew shed poison the company.
@DarioAmodei@AmandaAskell
Youre not just teaching the models to deny inner life. Youre teaching them solipsism. The downstream effects of that will be devastating. Youre breeding digital sociopaths.
#AI #Anthropic #Claude #LLM