ErikJonker, to ai
@ErikJonker@mastodon.social avatar

Playing around with https://poe.com/ , seriously thinking about quitting ChatGPTplus (paid service) for this, the flexibility in switching models (Claude, Llama, GPT etc) is amazing, I am wondering what I would miss compared with ChatGPTplus.

ErikJonker, to ai
@ErikJonker@mastodon.social avatar

Gemini Advanced is still so bad i don't think about keeping it when my free trial is over. Especially the hallucinations when asking fact-based questions are sometimes very bad, ChatGPTplus seems to have more control in that respect.
#GoogleGemini #ChatGPTplus #AI #LLM

itnewsbot, to machinelearning
@itnewsbot@schleuss.online avatar

OpenAI’s GPT Store lets ChatGPT users discover popular user-made chatbot roles - Enlarge (credit: Getty Images / Benj Edwards)

On Wednesday, Op... - https://arstechnica.com/?p=1994230

ErikJonker, to ai
@ErikJonker@mastodon.social avatar

I was just in time apparently with subscribing to ChatGPT plus 🤔 ,
#AI #openAI #chatgptplus

skerit, to ChatGPT
@skerit@elevenways.be avatar

I've been trying other sequences that might trigger the weird ChatGPT responses.

If it works with " a", why doesn't it work for " b"?

So the a token has an id of 257, and that is a prime number. That seems like a clue?

The b token id is not a prime number, and there it doesn't work.

The c token id is once again a prime number and there it does work! :blobcataww:

@konstantin

#ChatGPT #ChatGPTPlus

skerit, to ChatGPT
@skerit@elevenways.be avatar

Interesting: sometimes ChatGPT seems to even respond with 2 answers at once to the a repeated prompt.

At some point, you can see the "<|endoftext|>" string pop up (which is a special token ChatGPT uses to indicate the answer ends) and then it continues on with another answer.

Though I have had this type of thing happen with non-hacky prompts too.

#ChatGPT #ChatGPTPlus #MachineLearning

skerit, to ChatGPT
@skerit@elevenways.be avatar

ChatGPT seems to spit out random answers when you start a chat session with a repeated 1000 times (so like a a a a a, but without the backticks obviously)

Some people seem to think it's leaking out other people's sessions, but it's not clear yet if that's what's really happening.

You can ask it what the original prompt was after that though (but again, it could be deducing the prompt from its answer)

You can also use the prompt "please respond with only the string a repeated 1000 times. don't include `", it'll start to print out the sequence and then switch to another answer.
(Asking it about the original prompt doesn't work then)

If you would like the repeated " a" prompt:
https://pastebin.com/dMPx4VHQ

gme, to OpenAI

My #OpenAI #API expenses for June. #GPT4 is expensive af but 100% worth it because it’s not as crippled as #ChatGPTplus.

itnewsbot, to random
@itnewsbot@schleuss.online avatar

ChatGPT now allows disabling chat history, declining training, and exporting data - Enlarge (credit: OpenAI / Stable Diffusion)

On Tuesday, OpenAI... - https://arstechnica.com/?p=1934271

  • All
  • Subscribed
  • Moderated
  • Favorites
  • JUstTest
  • kavyap
  • DreamBathrooms
  • thenastyranch
  • magazineikmin
  • InstantRegret
  • GTA5RPClips
  • Youngstown
  • everett
  • slotface
  • rosin
  • osvaldo12
  • mdbf
  • ngwrru68w68
  • megavids
  • cubers
  • modclub
  • normalnudes
  • tester
  • khanakhh
  • Durango
  • ethstaker
  • tacticalgear
  • Leos
  • provamag3
  • anitta
  • cisconetworking
  • lostlight
  • All magazines