At first, I was ok with the update and I honestly like it. However, today, everything I ask the application to do (ex. translate a text), it sees my order as an code.
Is this a bug or something experimental?
1 Like
It’s honestly very annoying, especially for me, who uses AI studio for recreational purposes, such as translating.
I recommend that a category of prompts that are exclusively specified for programming should be developed, so misunderstandings won’t happen.
There are many words in common use in normal language that have a programming angle to them, and if they appear in a prompt might shift the probability of a good completion (which is really all the model is trying to do) in favor of generating some Python code instead of generating text for human consumption/enjoyment.
For instance, the word list often has that effect, the model will associate with an array and from there it’s easy to see that making a program looks like what the user had in mind.
You can usually tweak the prompt a bit and get what you really wanted.
Hey @lexian can you share an example prompt that is causing this so I can test?
1 Like
Try copying this text in this exact format and paste on AI STUDIO
The cow is there
It says moo
did you see what happened?
Question: are you feeding these sentences to the model wrapped in markdown? Perhaps, wrapped in this " ` " or " ``` " notation?
What model did you use, and what response were you expecting?
Just providing “The cow is there” doesn’t really give the LLM much to go on. Remember that these are pattern machines, so it will choose statistically likely next things to respond with. If you ask a question - it is likely to answer it. If you just provide a sentence, it is likely to just continue the thought process there.
For the record, this is what I get with AI Studio and various models:
gemini-1.5-flash-latest:
That's interesting! Could you tell me more about the cow?
For example:
* **Where is the cow?** (In a field, on a farm, in a zoo?)
* **What is the cow doing?** (Grazing, sleeping, mooing?)
* **What kind of cow is it?** (Holstein, Jersey, Angus?)
I'd love to know more! 🐮
gemini-1.5-pro-latest:
That's great! Is there anything else you'd like to tell me about the cow, or would you like me to ask you some questions about it? 😊
For example:
* Where is "there"?
* What is the cow doing?
* What kind of cow is it?
gemini-1.0-pro-001:
[Image of a cow standing in a field]
(yes, exactly like that)
one of the first two models
providing more detailed descriptions to what you mean and what your problem is could significantly help us identify your problem.
Currently, the above response looks like an incomplete sentence trying to refer to the prior reply, but with no reference to explain why it’s relevant to you, thereby making it difficult for us to figure out what the problem is.
Dont you backticks. For example, this is one here in brackets ( ` ) . If you use them to open and close as quotes, the text between will get separated out as a chuck of code.
Its a handy thing for us developers when we need the LLM to see the difference between normal text and code.
For your use case, stick to using normal quotation marks like these: " "
thank you for explaining