Discussion You think GPT-5 will be released later this year?

[MelaninQueen]

Some OpenAI (famous for its closed-source LLM models) employees are teasing the last quarter of this year regarding the continuation of GPT-4... we'll see.

 

[Magonia]

Doesnt matter.

They can decide to call it Chat GPT Gigachad, Gpt 4o.9, or completely passing it calling it Chat GPT 9000. Its all arbitrary when you get down to it.

What Iwant to know is if i can talk to chat gpt like in the video and show it my dick with a video camera

 

[MelaninQueen]

What Iwant to know is if i can talk to chat gpt like in the video and show it my dick with a video camera

Yeah, but OAI engineers have already 'safeguarded' the multimodal model with voice conversation before they launch it soon (in case you end up showing your dick). I hate their ridicule policies regarding NSFW.

 

[MelaninQueen]

@Amadeus Mozart

 

[WestEuropoor]

@Amadeus Mozart

Amadeus amadeus, amadeus , amadeus amadeus, amadeus, amadeus.. aMaDeUs

 

[MelaninQueen]

Amadeus amadeus, amadeus , amadeus amadeus, amadeus, amadeus.. aMaDeUs

 

[Deleted member 2206]

Ya probably it's supposed to be a yearly thing. They did make all of their models free so the next logical step is introducing a new paid tier to justify the price of the premium subscription.

 

[MelaninQueen]

They did make all of their models free

GPT-4 is still rate limited even on Plus tier. GPT-4o is limited to 10 messages (as well as 3 requests for multimodal feature) every 6-7 hours approximately.

I hope GPT-4 API price rates go down as long as they release another iteration (I play AI Dungeon)... we will al benefit in the long run, I suppose. GPT-4 Turbo is meant to be "cheap", as well as GPT-4omni. According to some Microsoft employees, they have been training "GPT-5" for a while now in their own datacenters with thousands of (Blackwell chip GPUs or they use their own NPUs?) GPU clusters.

 

[Magonia]

GPT-4 is still rate limited even on Plus tier. GPT-4o is limited to 10 messages (as well as 3 requests for multimodal feature) every 6-7 hours approximately.

I hope GPT-4 API price rates go down as long as they release another iteration (I play AI Dungeon)... we will al benefit in the long run, I suppose. GPT-4 Turbo is meant to be "cheap", as well as GPT-4omni. According to some Microsoft employees, they have been training "GPT-5" for a while now in their own datacenters with thousands of (Blackwell chip GPUs or they use their own NPUs?) GPU clusters.

Limited to 10 messages on +? I haven't even noticed limitations

 

[MelaninQueen]

Limited to 10 messages on +? I haven't even noticed limitations

I think they upped the rate limitatiots up to 80 requests or something? I've only used their API for my webui.

 

[404Errorcel]

Yes

 

[gooner23]

you can use perplexity for search engine with gpt 4o and without limit same as poe.com but poe doesn't have search engine but it gives better responses imo and better memory choices. can get both for like $10 a year too i think poe got patched tho

 

[MelaninQueen]

you can use perplexity for search engine with gpt 4o and without limit same as poe.com but poe doesn't have search engine but it gives better responses imo and better memory choices. can get both for like $10 a year too i think poe got patched tho

I wish I could afford 2x 4090s so I could run WizardLM 70B off SillyTavern.

 

[Magonia]

(I play AI Dungeon).

This is nice im trying it now. I didnt know you can run these directly on chat gpt. this is how mine looks.

Is yours better or worse than this?

Reporting the Findings​

You relay the details of your scouting mission:

• Approximately a dozen gnolls patrolling the ruins.
• A larger gnoll likely leading them, equipped with a distinctive weapon and crude armor.
• Makeshift fortifications around the ruins.
• Two sentries posted on elevated positions for a better view of the area.

Almah listens intently, nodding as you provide the information. Garavel takes notes, his expression serious.

Almah: "This confirms our suspicions. We must strike swiftly and decisively. Marcus, your task will be to lead a strike team to take out their leader. Without him, their ranks will likely fall into chaos."

Garavel: "We will provide you with a small team of capable fighters. Timing will be crucial. We need to hit them hard and fast before they have a chance to regroup."

Almah: "Prepare yourself and your team. We move at dawn. Any questions or additional preparations you need to make before we proceed?"

 

[MelaninQueen]

This is nice im trying it now. I didnt know you can run these directly on chat gpt. this is how mine looks.

Is yours better or worse than this?

Reporting the Findings​

You relay the details of your scouting mission:

• Approximately a dozen gnolls patrolling the ruins.
• A larger gnoll likely leading them, equipped with a distinctive weapon and crude armor.
• Makeshift fortifications around the ruins.
• Two sentries posted on elevated positions for a better view of the area.

Almah listens intently, nodding as you provide the information. Garavel takes notes, his expression serious.

Almah: "This confirms our suspicions. We must strike swiftly and decisively. Marcus, your task will be to lead a strike team to take out their leader. Without him, their ranks will likely fall into chaos."

Garavel: "We will provide you with a small team of capable fighters. Timing will be crucial. We need to hit them hard and fast before they have a chance to regroup."

Almah: "Prepare yourself and your team. We move at dawn. Any questions or additional preparations you need to make before we proceed?"

Pay for the latest tier (maybe Mythic if you can spare $50 every 2 months or so). The other ones like Mixtral and Llama 3 with more than 4k context are good enough and, in my opinion, WizardLM is one of the best models available (although I find it a bit too expensive; hopefully, it'll get cheaper eventually).

 

[Magonia]

Pay for the latest tier (maybe Mythic if you can spare $50 every 2 months or so).

What are you talking about? Mythic Tier?

WizardLM is one of the best models available (although I find it a bit too expensive

Arent the LLMS you talk about open source? you dont need to pay to use them

 

[MelaninQueen]

Mythic Tier?

It's the highest subscription tier for AI Dungeon.

 

[MelaninQueen]

Arent the LLMS you talk about open source? you dont need to pay to use them

No, AI Dungeon doesn't have in-house hardware to run 70B+ models (as you may know, they use third-party providers).

If I recall correctly, Wizard LM 8x22b requires 2x RTX 4090 + like around 10GB of RAM approximately where you can offload the remaining layers. Quantization is required (Q4 without severe hallucinations). You may think 128GB (DDR5) of RAM might do the job, but the inference speed will be lower compared to the first workflow.

 

[Magonia]

It's the highest subscription tier for AI Dungeon.

oh its on a separate thing from chat gpt. Who ever running AI dungeon making stacks i guess

 

[MelaninQueen]

oh its on a separate thing from chat gpt. Who ever running AI dungeon making stacks i guess

I mean aidungeon.com, buddy boyo.

 

[Magonia]

No, AI Dungeon doesn't have in-house hardware to run 70B+ models (as you may know, they use third-party providers).

If I recall correctly, Wizard LM 8x22b requires 2x RTX 4090 + like around 10GB of RAM approximately where you can offload the remaining layers. Quantization is required (Q4 without severe hallucinations). You may think 128GB (DDR5) of RAM might do the job, but the inference speed will be lower compared to the first workflow.

im looking at AI dungeon now looks fancy but already failing in character creation.

They dont seem to know the difference between ogre and Oread

 

[MelaninQueen]

im looking at AI dungeon now looks fancy but already failing in character creation. View attachment 137211
They dont seem to know the difference between ogre and Oread

You can still use auto-summarization when creating story cards if you intend to store real life data (or any other class for popular IPs). Story cards also take context from your story if I remember correctly, so they might seem to "hallucinate", but in general, it depends if "oreads" included in the model's dataset (what model are you using?) doesn't have vague context.

Not so long ago, you could train your own modules with NovelAI (I tried to make one with The Walking Dead lore), but I'm sure they have taken down this feature (I've read it on NAI's subreddit last week).

 

[Magonia]

You can still use auto-summarization when creating story cards if you intend to store real life data (or any other class for popular IPs). Story cards also take context from your story if I remember correctly, so they might seem to "hallucinate", but in general, it depends if "oreads" included in the model's dataset (what model are you using?) doesn't have vague context.

Not so long ago, you could train your own modules with NovelAI (I tried to make one with The Walking Dead lore), but I'm sure they have taken down this feature (I've read it on NAI's subreddit last week).

i just started free. I chose golarion. not the one in space but the fantasy one. It mentions eberron which is a different fantasy setting so even the starting description is wrong.

 

[Magonia]

can I make my own golarion that isnt retarded? i dont see an option

 

[MelaninQueen]

can I make my own golarion that isnt retarded? i dont see an option

Switch models or do the 7-day trial thing for the higher tiers.

 

[Magonia]

Switch models or do the 7-day trial thing for the higher tiers.

I made this, I think it sucks

AI Dungeon

AI Dungeon, an infinitely generated text adventure...

play.aidungeon.com

 

[MelaninQueen]

I made this, I think it sucks

AI Dungeon

AI Dungeon, an infinitely generated text adventure...

play.aidungeon.com

I like it, but my premium expired. I'll buy Mythic next month.