Coffeezilla patreon

Rabbit Gaslit Me, So I Dug Deeper (EARLY ACCESS)

🕑 Added 2024-05-22 23:33:21 +0000 UTC

Comments

Angie aka Mrs_Cuddelsworth

I would have laughed so hard if rabbit had said “get your lazy ass up and make breakfast yourself”

SeeBiscuit

So happy you remembered your password! 🧡

Osman M

To be super honest though, a *lot* of startups start off like this. Get the basic thing working, prove the concept, then refine it. OpenAI itself was the same way — they trained their models on copyrighted data, aiming to produce “value,” with legal concerns being punted down the line. Rabbit is bad but not as blatantly fraudulent as those startups with fake users (and there are a lot). I think they need about 1 more year of development time before being close to doing what was advertised. And that also includes contractual negotiations with the API service providers (DoorDash, Uber, etc). The guy’s mistake was launching too early and calling it “production ready” when it should’ve been more like a very early access private beta.

Lil_Ellie

I was not even halfway through this and started wondering about security and data privacy. Is this company selling user information? I'd assume you'd need to login to all of your accounts on those apps which gives them a lot of information about the user. I could be incorrect about users being required to login to the apps.

Zero

The guy made such a smart move, just lie sell a plastic square with screen, profit.

Chazzaroo

Your eyes at "The DoorDash app is under maintenance" sent me, I am howling.

Mithurshun Jeyatharan

Awesome investigation. I wanted to ask you Coffee. In the world of entrepreneurship there is a term called Minimum Viable Product (MVP) which serves to understand the market, whether or not anybody will buy your product. Is it possible that the Rabbit is just a MVP and now that they have a proof-of-concept the real product can now be build?

Ross Vincent

"I'm not gonna answer your questions because the answers would be so technical you just wouldn't get it, bro."

Brandon

Itching for the next video….

Ant Wakefield

How freaking stupid do u have to be to try and hoodwink coffezilla?!? History shows it never ends well…

hazard_fox

the apron 💅

Stephen Ford

2 years from now Coffee’s career will have completely pivoted from crypto related scams to AI ones.

JBones RVA

Scary how many psychopaths there are out there who say and do anything with no hesitation or shame just to make money. Thanks for making us all more aware Mr. Coffeezilla!

DaMu

I'm glad you reenacted the entire commercial of how Rabbit was demonstrated only to show how broken it is.

Elizabeth Woodley

I love the commercial remake! Great way to show the extent of the over-promises, and also your interactions with the rabbit are hysterical 😂

V L

Real talk even the term LAM is just a bastardization of the actual research term LLM for large language model. There’s no such thing as a large action model. I’m so done with these fools.

V L

Is the LAM in the room with us right now? 👀👀👀

Kilrah

Did you look into the ties between Rabbit and Teenage engineering? Apparently it's also not a "coincidence", but shared board member(s)

/// _

I really have to wonder if what they are doing can fall into bypassing API limitations of these app companies for their product- I feel like they might not have consulted with these companies before pushing their product to store

Brian Gibson

Another angle of theirs. Victim status. If we refuse to give you our side we can say you never gave our side. Scamological.

TigerHix

Awesome investigation, as always! A few things that I would like to mention, as someone reasonably familiar with this field: 1) Their claim that Rabbit is "faster than chat assistant apps like ChatGPT" might not be completely false, but it's a shady claim. OpenAI's API access to its GPT models, in many cases, *are* faster than the ChatGPT web app (accessed by non-developers). Of course, this doesn't change the fact that Rabbit is just a poor ChatGPT wrapper. 2) The concept of Rabbit's "Large Action Model" and Jesse Lyu's explanation of it is most likely inspired by WebArena (https://webarena.dev/), a benchmark to evaluate the web-browsing capabilities of LLMs. I attended a LLM agent workshop earlier this month, and the most recent results using GPT-4V (the state-of-the-art vision model) only achieved less than 30% accuracy on all the tasks. Keep in mind that those tasks (click through a few pages on a fake Amazon) are much simpler than the ones Rabbit claims to solve (booking Uber rides, ordering DoorDash, solving captchas, etc.). Knowing this alone, people should already be highly skeptical of Rabbit's "Large Action Model". 3) Theoretically, it makes sense to combine a tool like Playwright with a LLM to automate tasks. This is just like ChatGPT's function calling feature, e.g., it "browses the web" by invoking an external tool (that is not ChatGPT itself), which returns the webpage as a string, so that ChatGPT can process it. So, if Rabbit's "LAM", which is at best a finetuned/prompted vision LM that knows how to invoke external tools, can generate correct, *contextual* interaction instructions for Playwright, then Rabbit's claims may have some credibility (at least this is what academia is currently trying to achieve!). Note this is indeed different from Playwright's codegen, which generates static code that doesn't adopt to both the conversation context and the webpage's UI changes. But again, likely Jesse doesn't know what he's talking about, and Rabbit doesn't seem to have their own vision LM at all. 4) In the conclusion you mentioned Rabbit is "ChatGPT with an auto-clicker" - you are actually praising Rabbit too much! LLM researchers are trying very hard to make "ChatGPT with an auto-clicker" happen by improving LLM's web parsing capabilities :) Rabbit is more like "ChatGPT for answering questions, and Playwright scripts for using apps".

Hex

Your investigative work is great, but I'd love to see more analysis of the evidence in these cases and applicable laws to determine the (alleged) likelihood of legal liability. I know its not realistic to make every video a collab with someone like Legal Eagle, but I always feel like I'm left wondering what, if any, consequences the subjects of your investigations might face given the evidence you discover. Awesome content though! I've only ever subbed to a handful of Patreons, and yours is well worth it. =)

Tamir

I was on the fence for this product until I saw the huge picture of a lock. I had no idea they took privacy so seriously. Respect!

The Last Scientist

Good call: )

Coffeezilla

Yeah I don’t think they’re very happy with me…, which is funny bc I bet no one tried harder to get their side of the story

Coffeezilla

When I was authorizing DoorDash and Uber, it looked like they initialized a virtual machine and had me remotely control and login on a window browser within that VM. Uber thought it was a spam IP I guess bc they subjected me to the most horrific captcha I’ve ever done

Harry Moore

The people who bought this are victims*

Harry Moore

I think it's easy for millenials. Like if you're over 35 you probably remember internet jumping to broadband pretty quickly (tho it already existed), the sudden jump to smart phones, and then the sudden jump to everyone being online.... and a lot of these things made the leap the moment someone made it cheaper. In the case of the smart phone it happened breakneck. For a lot of millenials not versed in tech and understanding the work put in behind the scenes, that had to feel like how tech just works. The Rabbit probably just felt to them like "someone figured out how to do it"

Coffeezilla

Yep. I spoke w some of those ppl. Wild. Rabbit tries to claim you need “rabbitOS” to justify why you can’t download an app.

Coffeezilla

hopefully for their sake, no. I won’t be leaking it, it has really bad user privacy problems. I think someone leaked the android apk at some point but not sure if it’s public.

Harry Moore

Coffee, I think you misunderstood the point of Patreon. Was said to collect our donos... you're out here collecting W's

Brian Gibson

So basically, you aren't going along with their marketing grift so that's not good. 😂 Sign of the times with most corporations these days.

Jonathan Rivera

My hero ❤️

Coffeezilla

I’ll look into this

Michael DeRoy

Fantastic work coffee

Michael DeRoy

Same, love the aesthetic but if it was sorted in some way it’d be a lot easier to find yourself :)

Calvin

I don’t understand how anyone can watch that Rabbit R1 ad and think that they’re gonna an AI that is better than anything Google, Apple, and Samsung has developed at the low price of $200. A random Chinese company is somehow better than billion dollar tech companies trying to figure how to do this right now. Sure… The people who bought this are chumps…

The Last Scientist

I am enjoying the shift away from crypto. Is the codebase for rabbit available on Github?

Eric Aidinovich

Hype Coffee!! I’m really DIGGING this series so far :)

Alexander Demers

Hell yeah! The more content, the merrier. And as someone who has been watching people call rabbit out over the past couple of weeks, this two-part series has been another incredible deep dive into an incredibly shady business. Phenomenal job man! Hope you're doing well man, excited to see all the projects you have in store.

David Weintraub

A few people have gotten the Rabbit R1 to run on a standard Android phone. There could be some differences due to system permission issues. The Rabbit’s launcher app is installed as a root app. An Android install on a typical smart phone won’t get that permission. Still, the videos do show it doing many of the tasks as the official rabbit device — no LAM needed https://www.androidauthority.com/rabbit-r1-is-an-android-app-3438805/

Hamza H

They make users type login/password combinations and store them in their servers. I would be very curious to know what is the process they put in place to secure those passwords. Do they have one single encryption key for all users ?