[neom]

I dunno what Cohere is, but it's answers are super weird. I asked it "Where does the phrase "the rain in spain falls mainly on the plane" come from?" and it said:

Does anybody know a green song? What happens if a girl doesn't have a twin brother? What musical instrument did larry black sing 'jealous heart' on? Is country music more country than country music. I'm confused.? What is the difference between a copyright and a trademark? What is the best piano chords software u can get for beginners? What is the best music software for beginners? What is the song from the new dexter episode? Did he kill his brother? What is the name of the song dueling banjos? Was the raiders of the lost ark the first jungle hero movie? Who really invented reverse tape effect? Why can't me and my neighbor get along? What prophet spoke to the teachers of the law at jerusalem about their treatment of jesus? Are there any songs that describe the emotions of a cult member? What is the most ugly song ever made?

[jamilton]

IIRC, GPT-2 and 3 are somewhat like this, asking a question often results in it giving you more questions instead of an answer, unless you give. ChatGPT and GPT-4's tendency to answer questions instead of continuing them comes from the RLHF, I believe.

[spacemanspiff01]

Clearly the most advanced model we've seen to date.

It's already trolling you.

[quickthrower2]

And the comment means the next bot gets trained on this answer

[jayalammar]

For Cohere, make sure you're using Command-Xlarge-Nightly.

Otherwise, you may be prompting a Base LLM expecting the behavior of a different kind of LLM (an instruction-tuned chat model).

Cohere's Command model builds on top of the base model, giving it the capability to follow instructions and user commands.

[d1sxeyes]

The rain in Spain falls mainly on the plain. Not sure if it makes a difference to the response.

[IanCal]

Coheres model is Claude and it's a completion style one. So it's trying to carry on what you wrote, and a list of questions is reasonable. It worked just fine for me copying and pasting but you might try this format:

Q: your question

A:

Then it is more likely to fill in an answer for you.

[jayalammar]

Cohere actually trains its own models and they are not based on models from other providers [I work at Cohere].

Your prompt suggestion is a good one for LLMs as a whole. Any information added to the context informs the model and nudges it towards the expected answer format.

[IanCal]

Oh sorry I've got mixed up with who runs what model, I thought coheres one was Claude, but that's anthropic. Wasn't trying to say it was based on another.

[IanCal]

I got mixed up with the models, Claude is from anthropic.

[sangnoir]

Just like your own question - it seems to me it's riffing on "Bar Quiz" type questions. Have tried the 'Question:'...'Answer:' prompt format?

[jozvolskyef]

You didn't tell it to act as an assistant so it just assumes that your input is some list of questions that it's supposed to complete.

[psygn89]

When I asked if elephant farts were loud it answered with a lot more detail than ChatGPT.

[rambojohnson]

it seems that Cohere might be suggesting that the phrase is as devoid of meaning as its unrelated questions, considering that the phrase was originally just a vocal exercise for a character in a film.