The *mega* awsome Leo!

Yoo! :slight_smile: I became a Leo Premium user because I truly love the idea for it.

He’s that “little” friend I can use when I have no valid input from people using Teamspeak or Discord.

But, when I want to use Leo for more “serious”-topics [Gaming related…] He seems to be so un-updated with his Knowledge so it actually becomes a littlebit of a struggle to make him try understand with what Topics-(…Example…) I really ask for him to help me understand or if He has another cool view I might find useful.

Will this change in [near?] future?

Again, :heart: Leo :heart:

1 Like

Probably yes. Models are updated every now and then. Premium uses Claude by default, and Anthropic has already released newer versions of it. I hope Leo doesn’t get updated to a new Claude though because I hear it’s even more annoying and avoid answering even more stuff. Look at the ratings here and focus on the arena score not on the synthetic tests:

If it doesn’t know about something, go to the wikipedia page about it, open Leo there, and ask it what you want. It will answer using the wiki page as context.

I hope Brave can give us that awesome Mixtral version in the premium version.

1 Like

Thank you @Tritonio ! I take your answer as a “approved”-information, because i’m very new to this Leo \ A.I feature.

I wish there was a simplified way to help develop :lion: Leo as a User. Either by becoming “Verified” or something else. Everytime :lion: Leo provide me with faulty information I try correct it as accurate as I can, but I dont know if this is even helpful or if Im just wasting my time.

I will Really take your Idea of using “selected”-Wiki and use Leo there!

Edit :: Hopefully the developers see the same mistake as you did with Claude!

1 Like

Unfortunately you are wasting your time trying ot teach it. Even if it thanks you for helping it improve, that’s a lie. There is a way to like or dislike an answer they gave you by clicking the three dots to open a menu on the answer. But I don’t know where this feedback goes exactly and if Brave does anything with it. There isn’t much they can do to improve Claude though. First of all we’re using Claude-instant-v1, which is already not the latest model of Anthropic. If Anthropic releases a Claude-instant-v2 it won’t be based on instant-v1, but rather on claude-v2.x (the full, non-instant model). I’m also not even sure if Brave is allowed to share feedback with Anthropic according to the ToS of Brave. But simply telling the bot to remember something will do absolutely nothing. Each conversation you have with it is completely isolated from any other conversation, there’s no persistent state between them and LLMs don’t learn while talking. They learn if you retrain them on existing chat data, and only Anthropic can do that for Claude.

1 Like

Thank you for that honest answer!

The new Brave update allows users [Premium] to Use Mixtral unlimited.

I look forward to use it awhile, based on your comment!

Edit :: @Tritonio I will post here again, once I’ve got some useful experience with the Mixtral!

1 Like

Haha this is insane. I honestly did not know they were about to add Mixtral! I said it, and then I upgraded Brave and it had it.

AND it has a button for listing questions about the content instead of having it always on! WOW I love this. :heart:

1 Like

Hey @Tritonio !! How has your personal usage with the “new“ Mixtral / Claude Instant since last we talked? I’ve found that atleast Mixtral is very accurate if you want to explain certain things.

As an example I wrote this; Hi Leo, what is 19 gmt in 24 hour CEST?

He replied:

19 GMT is 21:00 in CEST (Central European Summer Time) in 24-hour clock, which is 9:00 PM in 12-hour clock. CEST is the time zone used in most European countries during daylight saving time, which is 1 hour ahead of CET (Central European Time).

Edit ::


Mixtral is still too quirky for me. I used it for a few things, it’s not unusable by any means, but I still think I preferred llama to it. But in the end I still mostly use Claude because of the large context window. :no_mouth:

ok! That is interesting to read. Do you think Brave will improve claude in anyway? That is the one I also do prefer because of the ability to have “casual” talks with :green_heart: :lion: Leo

Claude? They cannot, it’s owned and operated by Anthropic. They can change the context given to it (“You are Leo, an AI used by users of Brave Browser, the time is xxxx, you should reply like this and that”) but it won’t radically change.

I hope we are eventually allowed to change the context ourselves. And hopefully even allowed to write out own custom text extractors to support some websites that don’t work nicely with Leo. Perhaps a way to call Leo from Chrome extensions that Brave can use would be the best solution to all these.

Alright, thank you for that feedback!

I thought Brave was responsible for Leo (A.I.) Enterily.

I think its really nice of Brave to allow the Premium users to have a selective choice of different (A.I.) anyway, tho!

1 Like

None of the models is created or trained by Brave at this point. Mixtral and Llama could be re-trained by anyone I believe, given time and money, as they are both “open” models. They are also both run on Brave’s servers. But Claude is not like that, Brave is only proxying your requests to it so that Anthropic can’t see who’s making the requests. (assuming you trust Brave to not re-configure their CDN to stop throwing away your original IP of course, etc)