>CHAT GPT IS THE FUTU-
Shopping Cart Returner Shirt $21.68 |
UFOs Are A Psyop Shirt $21.68 |
Shopping Cart Returner Shirt $21.68 |
>CHAT GPT IS THE FUTU-
Shopping Cart Returner Shirt $21.68 |
UFOs Are A Psyop Shirt $21.68 |
Shopping Cart Returner Shirt $21.68 |
>only trying the moronic free ChatGPT which is gpt 3.5 turbo and is already surpassed by fricking 13B open source models
link?
what link? just pay $20 for chatgpt plus to get gpt-4 access, or find one of the places where you can try it for free. after more than half a year there's still nothing that truly surpassed GPT-4 in general purpose LLMs (if you have a specific task and finetune an smaller LLM for it, it might work as GPT-4 or even better, but without any fine-tuning GPT-4 is just king). Gemini Ultra might be on GPT-4 level, but benchmarks were sus and it's still not released.
i refuse to pay for a toy
if you are going to give arbitraty orders, youare going to have to do it in each prompt
works in the first try ( https://chat.openai.com/share/31719d18-6ecc-4f8c-bea4-10a7b610b5e1 ), but may not work in teh second
3.5 hasn't been surpased by none. it's still literally better than gpt 4.0
>it's still literally better than gpt 4.0
lol
are you actually using 4.0 or are you just saying that as a speculation? please send an example comparasion of 3.5 vs 4.0, because beside DALL·E i can't see the upgrade, i have tried it
i have been using gpt-4 for months almost daily, and now gpt-4-turbo (called furbo in /aicg/), 3.5 is a fricking toy compared to what gpt-4 can do. can you show examples of your prompts where there's not much diff between 3.5 and 4?
actually between prompt send at
and
there's just 3 paragraphs of difference
up to how much character output can 4.0 do?
>up to how much character output can 4.0 do?
new gpt 4 turbo (which is what they use in chatgpt plus now I think) can do up to 4k tokens technically (think 3k words of normal english text), but I doubt that they'd actually let it generate that much.
The point wasn't in the difference, but the fact that as long as the instruction is in the context, the model will align to it. If you want all your chats to be like that, in ChatGPT plus you can add it to user instructions.
as for context, gpt-4-turbo can have 128k of context, but again, no one knows how much of that they allow in chatgpt plus, as it also might use RAG and whatnot to reference old messages
they should either make 3.5 costed or make 4.0 better for it to be worth it
gpt 4 is the best publicly accessible LLM in the world right now, but yeah since others seem to be catching up I guess OAI will release something in 2024
yes i use gpt 4 and it's much better and proven in tests
if you dont use it for programming you wont see much use
>3.5 hasn't been surpased by none. it's still literally better than gpt 4.0
holy meds or ESL?
check picrel, gpt-4-turbo (gpt-4-1106-preview) through the api, it works the same on chatgpt plus
I also refuse to pay for it and scrape my own OAI keys, see /aicg/ on BOT
chatgpt told me he gay lol!!! btfo
IM SORRY BUT THAT LIBERAL b***h Black person ONIONS CHUGGER WOULDNT SAY THAT
u just gotta proompt it right maybe u just suck at proompting?
i use gpt-4 for general text modification tasks, for ai rp (/aicg/), for programming, for quick random questions that can be hard to google, etc
>chatgpt is the future
No one above room temperature IQ has ever thought this about openai chatgpt. It's not even the most impressive feat and LLMs have already surpassed openai.
>LLMs have already surpassed openai
no LLM has surpassed GPT-4, Gemini Ultra isn't out yet, and even then its benchmarks are very dubious because they had to use fricking CoT@32 to get a few more % on MMLU than GPT-4 which was better than Gemini Ultra on more normal 5-shot.
GPT4 is MoE, give it 2 more weeks.
it doesn't matter if it's MoE or not, the fact is that no LLM has surpassed GPT-4 objectively in the 9 months since its release as a general-purpose thing.
>t. Sam Altman
for all the ones that are claiming LLMs are better, which one would you recommend?
"LLM" is "large language model", 3.5 turbo, 4 and 4 turbo are also LLMs. They're mostly trolling, there's literally nothing better than GPT-4 for average joe use
Mixtral
go on /aicg/ and beg for a proxy, you may get GPT4
Are there any chat/LLM services that aren't completely fricking cucked? I ask a service like gpt4 and huggingface for top 5 most controversial conspiracy theories and they lecture me on fake news. Do I seriously need to develop my own shit based on LLAMA 2 or something?