For those that were in the chat today where Jesse showed off Teach Mode, what were your thoughts?
5 Likes
Personally, I’m very impressed, I wasn’t expecting to see everything that Jesse showed off, but I am now even more excited to start training rabbit agents on tasks!
1 Like
I missed it, wish I knew more when Jesse would be around to catch cool things like this. I think I am 1/5 in catching them!
3 Likes
Same, I missed it too. I laboriously scrolled thru the discord conversation that they had. From what I could make out from the comments after Jesse’s demo folk were really fired up
1 Like
Anybody who was there willing to type out what happened?
I’ve got a few projects in mind for when we get access but I’d love a bit more insight into how it works and what limitations it has!
2 Likes
In short, Jesse basically confirmed the early versions of Teachmode are real and working, showing a few use cases where he made a rabbit agent that could search google images, join a discord voice chat, and I think one other that I can’t remember right now.
The highlight for me was after training the google image search to go to Google and search an image, because he was in Japan and their default search engine is Yahoo, it went to Yahoo and searched an image, that alone was amazing. It was able to figure out how to navigate the Japanese Yahoo site and still return a correct result.
Jesse mentioned this is still early and they are still aiming for the end of the year for Teachmode rollout and the potential of having a Agent Store so that people have to chose what they want to install on their devices and not have anyone creating anything malicious and pushing it out to all R1 devices, which I thought was super smart. He also mentioned the possibility of having the developer of the agents get paid for complicated tasks or charging for their rabbit agent they created, which was also excited to hear.
Anyone else feel free to add something I missed, but I think those were the highlights for me.
12 Likes
Ooooh awesome! I’ve always been a fan of the community curated content, and marketplaces for the projects people make. The art program Clip Studio Paint does something very similar, and imho its the best system for something with so much personalized content. I’ve uploaded a few assets, and it works how you would expect. If you price things reasonably people will buy them, if you make quality products, people will buy them. Only thing I worry about is that agent discovery wont be good enough to find what you are looking for…
4 Likes
Yeah discovery could be challenging, but I’m sure it would be included inside the Rabbithole somehow based on topics. Maybe even a future feature would be to have the R1 search the store for a helpful agent!
1 Like
For sure. With how this forum is, I feel pretty confident they could make a good storefront for searching up agents and finding the utility you need! It will probably fall more on the individual users to write good descriptions on what their taught rabbits can do.
3 Likes
So basically he demo’d a Selenium like session replay tool. That’s been real and working for 15+ years. That’s not a LAM, and does nothing to address any of the hurdles that make a solution that is based on old technology viable in a capacity that comes anywhere close to meeting the vision.
Show me an agent that handles any kind of authentication challenge that comes from two-factor auth, jumping http requests from multiple AWS IP’s, captcha solving, authorization access from another device/txt/email, or anything native (non https) at all - and then you got me. Until then - it’s all vapor.
I’m not dismissing your take on the recent demo from earlier, I’m intrigued to know if you’re basing your comment on the basis that you were present at the time that Jesse gave the demo?
Fair point. I based it on the visual overlay I saw on several illicitly taken screenshots and a running screen capture that hypothetically speaking may have been sent to me by someone in the industry who knows her stuff. Oh, and also still not an answer to any question I’ve raised.
I want to be wrong so hard it isn’t even funny - this stuff would change the world. Again, I will say it - even doing this in a different way than the typical http request, DOM read, js/field/post manipulation, http request, rinse, lather, repeat - still doesn’t solve the fundamental issues I have with the feasibility. You can do all of these things now. So what was different that I’m missing?
Thanks for responding, I’m all for constructive feedback clearly there’s a long way ahead for teach mode to become reality.
Like you I want to believe the r1 will deliver what it promised, I’m afraid my expertise is fairly limited when it comes to the prevailing technology 
Appreciate that!
You don’t need to be an expert to see some pretty glaring initial flaws. How many times when you login to your own email, or your favorite airline’s website, or to a retail store you frequent, or to a service you use often - do you get the message to the effect of “we don’t recognize this device, click on the email we sent you or enter the code we txt’ed you.” Happens most likely every time you change locations, and definitely every time your browser gets updated for security updates/features.
And that’s just for sites that you don’t explicitly have locked down with two-factor authentication (which we all should). If you ask for it or not, it’s still two-factor authentication being forced on you, to protect you, which is a good thing.
And passwords are in many ways on the way out, how many sites do you use that instead of a login, you just enter your phone number, get a txt code, and then have to enter that to login?
Now imagine that there is a browser, running in the cloud that Rabbit is paying for (which is just someone else’s computer) on your behalf that the Rabbit could interact with and do whatever it is that you taught it to do by replicating your behavior on a computer that you don’t have any access to and knows nothing about you. What happens to that automation when this inevitable scenario goes down?
And that’s just the tip of the proverbial iceberg.
1 Like
Interesting when you put the challenge in this context, certainly seems like a tall order.
Hopefully they’ll crack all the issues needed, I’m impressed with rabbits commitment and what they’ve brought to the table so far and what’s forthcoming in the immediate timeline.
I’ll be able to do some low level testing myself tomorrow, r1 due to drop 
Good luck with it! Enjoy it for what it is - many find it fun and interesting to use - which I totally get, especially if you are relatively new to conversational AI and haven’t tinkered much. It is an amazingly low barrier to entry to see some of the things that will blow your mind and hopefully pique your interest into using some of the more advanced tools out there - or just having fun testing the boundaries.
AI is great - and you can talk it into doing pretty much anything you want to if you are patient and persistent enough. Also, kind of scary if it has the capacity to login to all of your accounts with, or without you
1 Like
yeahhh, missed this. Wish I didn’t. Seems interesting.
I want to be able to teach it series of tasks. With the ability to group these tasks into “series” of tasks in a “group”.
Group > Task Series 1 > Task Series 2 > Task Series 3
Example of a Task nested within a Task Series;
Task - Search for Flights with specific criteria ( command word = “Find Flights” , variant = “price”, “stops”, “total time”)
- Number of days: [Enter number]
- Number of travelers: [Enter number]
- Departure date: [Enter date]
- Return date: [Enter date]
- Departure city: [Enter city]
- Destination city: [Enter city]
- Priority (1-Price, 2-Number of stops, 3-Travel time): [Enter priority number]
1 Like
I feel like this is a pretty big ask for something that can’t even run a single task as you describe, let alone grouped into a set of tasks, let alone rolled up to a group of tasks, let alone remember its own name.
I am willing to help. I can run tests, collect data, report findings. Point me in a useful way. I will take action.
1 Like