It's unfortunate that the only reason this problem exists is because people buy into the marketing. But the amazingly ironic part about this is that MKBHD / Marques Brownlee has hocked this type of garbage (aka sponsorships) for years, propping up a lot of this bad behavior along the way. Personalities like him are "celebrities" from a lot of consumer's perspectives, and they trust what he says verbatim. From an objective stance his reviews are lackluster: they contain very little data and no fundamental framework that underscore his "reviews". He's a YouTuber with an opinion that is being influenced by sponsor marketing money. Yet his well crafted productions are pretty and easy to consume, which sells and Google puts it front and center. To say that he has a "core philosophy" is laughable. The entertainment I would watch is if people like him had to disclose all earned income along with these videos. I'd gather a lot of his loyal following would start putting those dollars to sense with respect to how his videos are positioned: call-out, hype generating, brand supremacy, etc...
I've seen interesting products, and either been curious or ready to buy. Unfortunately, the only available information to me is the marketing.
Seriously, look on amazon. You can see pictures of the product, read the description, and read the reviews. this leads to 2/3 of the information being by the marketing team, and quite possibly much of the final 1/3, the reviews.
Being able to view a video of the product by someone (anyone!) on youtube gives you a chance of critical information you need for a decision.
Who cares if he's lackluster. If he even in passing gives you an idea of what the product might actually be like, you've been helped.
By the way, even consumer reports, formerly the gold standard in unbiased reviews is part of the gravy train. When they switched from the magazine to online, they acted like a funnel to purchases. sigh.
If you actually watched his videos you would have seen the one where he actively calls out that a lot of his viewers aren't interested in the product at all and just want to see another high production value video from him just to kill time, with no intention of ever buying anything.
"The entertainment you would watch" would probably amount to nothing and you probably wouldn't even finish the video, because it would be too boring.
I'm not sure why I would want to watch his videos if he's basically telling people that he's an entertainer - and that he caters to that. I see why he would do that, but that seems to bolster my assertion that his focus is on production, not review.
Most review videos are entertainment- and that includes accounts that focus on performance and include metrics, too. The medium makes the message. Been that way since Will it Blend?
Tell me you’ve never seen a MKBHD video without saying you’ve never seen a MKBHD video. He’s probably the best tech reviewer on YouTube. I definitely trust him more than other reviewers who do it for the clicks/lols.
He doesn't hold a candle to people like Dave2D or Hardware Unboxed, the latter of which has extremely thorough testing for every single one of the products they review. MKBHD is often inaccurate about many things and certainly not exhaustive in his reviews like other reviewers are. His channel can be better classified as infotainment, much like Unbox Therapy and (for certain of their videos) Linus Tech Tips are.
He doesn't do it for the clicks? I can't tell if this is sarcasm, but I'm going to assume it's not - because if this is truly your perspective I'd implore you to install a few extensions that help weed out what YouTube knows you'll watch. Clearly the algorithms are working.
To sum this one up:
Tell me you can't tell the difference between an influencer and a reviewer without telling me you can't tell the difference between an influencer and a reviewer. The bar is really low with respect to quality (not production, but content) these days.
You really should watch some of his videos if you think this. Marques has been doing this for years at this point, he's a reviewer, not an influencer. The clickbait titles is purely a YouTube thing, that is only indicative of the video being on YT, unfortunately.
I don't believe we share the same barrier to entry for objectivity, is really what it comes down to. He's sponsored, he's an influencer.
From a recent article [0]:
"The 30-year-old YouTube kingpin makes money from sponsorship deals with brands, merchandise sales, and a whole lot of ad revenue from YouTube."
Where do you draw the line between his format and his sponsorships? And how do you know that he's "reviewing" without monetary bias? As an example he put out this [1] video with Buick a few years ago. Let's try to be honest for a second - he wouldn't be caught dead driving that car unless Buick paid him to. And that is my point. There's no boundary between his "reviews" and his "ads". They look, they feel and they are the same format. He's excited about this Buick, "reviewing" it on his channel, it does say that it's sponsored, but is it his actual opinion? Very doubtful. And that, right there, is the difference between a reviewer and an influencer. Is he a charlatan? Maybe not quite that far - but he is being dishonest by making the sponsor video in, basically, the same format as anything else he does. Does he have relevant points? Sure - but there are other reviewers that have much deeper knowledge of the reviewed product. Case in point is his car reviews - he's not a true car enthusiast. His car "review" videos clearly reflect that when you compare it to someone who knows the car market in-depth (there are a lot of great car reviewers on YouTube, MKBHD isn't one of them). Is it good enough? For most people, probably. But again - I don't trust anyone to do a full spot on a car that's not even close to being on his want-to-have list and then expect him to be unbiased when "reviewing" a competitor. He puts out ads for his own brand, and anyone willing to pay him. He's an influencer. And now, he's hocking wallets that are $125+ [2][3].
Looking at the comments to the Buick video, that seems to be the exception to the rule- commenters dislike it but express they like his other ones. Of course, perhaps he’s had more sponsored videos since then, but again, maybe they’re exceptions to the rule.
It seems to just be a leak of their sandboxed headless browser setup and the API code for controlling it. Obviously such a thing will run arbitrary JS from the web so inevitably there will be something like a browser sandbox exploit, and subsequent dump of its filesystem.
The leak doesn't seem to contain what Rabbit calls the LAM, their purported AI model for interacting with UIs. And what the leakers are claiming is that Rabbit's automation is just handwritten scripts which seems to be completely unsubstantiated. The rabbit secret sauce could still turn out to be a scam but I didn't see anything to corroborate any of the leakers' claims. Grepping the files I found no reference to doordash, uber eats or midjourney, only a path reference to what appears to be a spotify integration library, but the source for that isn't there.
1. Filtering by model should be enabled by default. Mixtral-8x7b-instruct on Perplexity is almost as fast as the 7B Llama 2 on fireworks, but are quite different in sizes.
2. Pricing is a very important factor that is not included.
3. Overall service reliability should also be an important signal.
Can you describe what you'd like to see for #1? We currently show everything, but let people filter via the UI or URL param, e.g., https://thefastest.ai/?mf=3-70
you can't automate playwright without a decision making component in front of it, they are definitely using a transformer there. one could train a llama and make it perform triggers to playwright automations. you can even get deep into transformer tokenization and create action tokens and a formal grammar for your generation, build a parser on top of your predict function and have a "lam" working. the fact that they use playwright does not imply it is not generative ai. i'd say it is really hard to do those actions without a transformer involved
Isn't the secret sauce just VNC with playwright? What more do you need to achieve 80% of what they are showcasing (basic doordash orders, spotify controls)?
I can't find any purported auomation scripts for those services as claimed in the Github page. There is a reference to "cm-spotify-client" which seems to be some sort of custom integration code they've written, but other than that there is no reference to doordash, midjourney, or uber eats. This dump seems to just be the code/infrastructure to run chromium/playwright in kubernetes, wrapped in a Node API to accept commands, persist/hydrate browser state, etc.
Is there a less convoluted less enterprisy implementation of such a project someone can point to. I was interested to see what is or isn't so intensive about it. Or is this a good scaffold to start a fork?
Yeah, good idea! I'm not sure if you would be successful mixing LoRA + functional tokens. If you could, that would be great. Then You could ship very light LoRA packs with repositories.
Their LoRA training was I think against their finetuned model, not Gemma-2B directly. But, seems worth playing with -- could be super useful.
Yes, you can do that! We have experimented about this idea. The lora training is reported in the paper already, and it works well. I think it would be very interesting.
"Never. Ever. Buy a tech product based on the promise of future software updates."
https://x.com/MKBHD/status/1383616274693951494