There are rumors about a new AI Chat coming soon based on Llama 3. Is that true? Don’t get me wrong, the current AI Chat is great, but its goldfish memory ruins any chance to roleplay with an elaborate lore or world-building, not to mention the constant repetition of expressions like “maybe, just maybe” and so on.
I think it is already Llama 3. Also, lots of new models were released recently. I guess it is better to aim for large context length models now.
No way this is llama 3: https://perchance.org/ai-chat Unless there are other “official” chatbots. This chat has a very short memory. I’m sure it is Llama 2; it would make sense, since I suspect it has around 4k token context, while Llama 3 has 128k. Maybe I am missing something. Are there other chats like this that use more advanced LLMs?
From what people write on reddit, its indeed llama 2, and gets updated to 3. I used all official chats - characters and rpg, and it seems they use the same LLM.
Oh, that’s great. So I was right. I’m not crazy, then. Is a new version coming out? I heard they were gonna update it in July, but it never happened.
It looks like the dev is in the middle of implementing it :
https://perchance.org/image-gen-update-aug-8th
People saw this announcement in the AI image generator
Thanks! I can’t wait. Just imagine going from 4k tokens to 128k tokens. This is gonna revolutionize free chatbots.
I mean if it’s just Llama 3, it will feature context window of 8k tokens, not a big uplift from Llama 2 and its minuscule context window, I heard it’s very impressive but it’s being held down by its small context window, if it’s Llama 3.3, then yes, it’s 128k context window size would be a game changer.
Well, I hope it’s 3.3, after two years of waiting, and with the image generators using Chroma and Flux. I can’t complain about a free service anyway. It will be a bittersweet victory; having 2x context will surely improve the roleplay experience, but it will still be impossible to create lore and world-building unless we have at least 50-60k context tokens. And yes, the actual model is amazing, but it’s being held down by the context window; you cannot create a deep lore or word building, and worse of all, the chat forgets the key events around 3-4 pages of convo, having to rely on summaries that miss A LOT of important elements.

