Spent 25% of my usage today yelling at Claude for a simple visual fix only for it to keep telling me I was wrong. Eventually the model came to the conclusion I was right from first principles by checking itself. Reassuring to know I wasn't the only one, thanks Raoul!
You can tell when Anthropic are about to release a new model because the current version of Claude gets so nerfed to give them extra compute/inference that it becomes unusable for several days. I can't get it to follow any instruction right now...it just gives up. Grrr
An insightful, thorough explanation from an individual who is amongst the deepest in the weeds on this and has the battle scars to prove it. Worth reading in its entirety
Tapping in on the Twitch viewbotting discussion as an actual agency brand advertising executive.
This Twitch post is unlikely to convince any advertisers or streamers that the issue is improving.
Here's how to actually address this problem -
Viewbotting is an engineering and incentives problem disguised as a cultural one. There is no solution to this problem that can ever be achieved via culture change or moderation. People are debating the efficacy of the solution in Twitch’s post when the entire implementation isn’t even relevant.
Given Twitch’s track record, no one should believe Twitch will enforce this “viewer cap” change appropriately even if they did have the proper backend data (and I think they don’t.) They won’t even enforce against popular streamers who have viewbotting programs visible on their screen, streamers who roleplay bringing slaves on their broadcasts and feed them treats, or broadcasters who advocate blatent political violence, and so on.
Many streamers are on their 8th or 9th temporary bans for various TOS-breaking content. No one believes Twitch's intervention via moderation will change anything.
So again it’s a matter of engineering and incentives. I’ll break down each one. But first a productive question to ask. Why does Youtube Live - which now eclipses Twitch’s live viewership by 40% - have nearly no problems with bots and viewbotting?
Well first the engineering problem. Viewbotting gets solved by better detection, pattern recognition, IVT analysis, device fingerprinting, IP patterns, watch-time anomalies, removing off-platform embedding, and other boring backend stuff. As Dan correctly noted, these are things you can’t be public about because they’ll quickly see counterplay from the viewbot services. You just have to lock in and fix it. Google has spent billions of dollars solving this because advertisers don’t like invalid traffic. Their IVT rate now sits at a healthy 11% or so, easily beating competing live services like Tiktok (24%) and Twitch (35%+.)
Twitch hasn’t taken this problem seriously. I’ve spent hundreds of thousands of dollars on Twitch, and no team member has ever communicated to me or any of the agencies I work with about their IVT rates accurately. On the contrary it often feels like they throw raw viewership numbers around in pitch decks and presentations as the REASON to go advertise there. That used to work, but money is getting smarter in digital. Brands correctly care about return on spend and engagement with real people rather than just the perception that there are viewers around.
You need both bravery and expertise in engineering to make massive changes. Twitch actually has phenomenal engineers, many who were brought on by the former CEO to develop the world’s most advanced CDN (content delivery network) from 2015-2022 before Youtube Live recently surpassed it. But I’m not sure the company has the bravery to make real, necessary changes that many commentators on this post have noted must happen. Successfully “fixing” this would represent a 30-35% global drop in viewership for a website already perceived as ailing against its competitors.
To solve the engineering side you just have to apply more of the above technical solutions and spend money and time. Communicate with advertisers, refund us in high IVT events, keep improving your CDN, etc. It’s not a moderation issue, it’s not a viewer cap issue, that’s crowd pandering and will fix nothing. Both advertisers and the community begged for years for attention on this topic, while viewbotters bled us all and collected millions of dollars. Twitch has eroded trust and should just stop making these posts to the public. Users have lost tolerance for the platform and just need to see the actual changes.
The second problem is the incentive problem. Every system to succeed on Twitch is set up based on the number of viewers you have. The entire Partnership/Affiliate pipeline is viewership based. The entire discovery funnel is viewership based. Sponsorships are calculated on CCV per $. And the entire culture of the website is “more viewers = better.” So of course everyone will do anything possible to get more viewers. This not only encourages viewerbotting but also toxic streamer culture (IRL nuisance streamers, etc) because, again, it’s all about the views.
On Youtube Live monetization rewards sustained engagement and overall channel performance. Ads, super chats, channel memberships, etc. Discovery is pushed by an interest-based recommendation algorithm. Viewers will find you because of interest alignment - not because you have the most viewers. Barely anyone viewbots there because there’s no point. Adding viewers won’t help your channel, and Google’s AI/ML will instantly and automatically suppress your channel and shut off your ads.
It is frustrating that the C-Level over at Twitch keep posting about viewbotting as if it’s going to get solved by human moderation. As if Twitch is going to magically have an amazing enforcement team with “the data” that goes after the serial offenders and returns everything to normalcy. This is all solvable, even from where Twitch is now. But you have to make the investments on the engineering/CDN side and just get to work. Everyone is waiting and hoping for these changes.
Twitch’s viewbotting epidemic is a downstream effect of what every major streamer and industry professional told them would happen for years. They didn’t implement discovery engines, they didn’t change the incentive systems, they didn’t invest aggressively against obvious botting cases, and they didn’t communicate. Now they have an existential advertising and creator crisis. You reap what you sow.
The most HYPOTHETICALLY exploitable method to get your least favorite streamer banned: viewbot a creator I despise. They get a warning with no way to correct because they're not the one viewbotting, a bad actor is.
Harsh penalties to streamer. That streamer of course has to take on the burden of trying to find this viewbotter (of which there could be dozens or more that spawn up once they see the vulnerability).
Probably get no practical help from the platform because, and I quote: "viewbotting companies quickly respond with updates to avoid detection".
So now you, the creator who worked your entire career to build a legitimate audience on one of the most difficult platforms to grow, are completely at the mercy of trolls not doing what they always do.
Twitch could likely (at least) curb viewbotting significantly by removing embedding. For those who aren't familiar, embedding allows you to place a live stream player on a high-traffic website OUTSIDE of Twitch, and likely the people who go there don't know or aren't real, so you can artificially inflate viewer counts on the Twitch stream itself.
Will they do this? Well that all comes back to incentives.
I'm not convinced that it's in the best interest of the platform for the narrative to continue to be about viewbotting because then advertisers need to take a 2nd look at their spend on Twitch and ask:
"Are those numbers legit?"
"Are those actually humans?"
"Should we be spending here?"
That will cause them to re-allocate spending to other platforms (i.e. IG, FB, TT, YT, etc.) where they have more [perceived] confidence that the impressions are real, thus becoming an existential issue for Twitch.
So, they need to signal to the market (specifically advertisers) that the traffic here is NOT filled with bots while simultaneously showing that their spend is a net positive. Hence, this announcement.
In the last couple of weeks, we've seen X Communities eliminated, Instagram targeting aggregators (major part of the ecosystem) and now putting Streamers at risk for something out of their control.
If you're a creator of any size and don't have a community on a platform where you own the means of distribution (email, text, etc.), you are at the mercy of whatever the platform decides to do.
Act accordingly.
A note on our work to combat viewbotting, from CEO Dan Clancy:
There’s been a lot of discussion recently about viewbotting on Twitch, and I wanted to share an update on our enforcement efforts.
Viewbotting is bad for our business. We don't benefit from it, and we believe it harms the creator ecosystem overall. However, effectively combatting viewbotting is challenging. As we deploy updates to our real-time detection algorithms, viewbotting companies quickly respond with updates to avoid detection. Also, our detection systems must be precise to ensure that legitimate viewers are appropriately counted.
Today, we’re introducing a new enforcement type that we plan to roll out over the next few weeks. For channels identified as persistently viewbotting, we will apply a cap to the streamer’s CCV for a fixed period of time, on all of the Twitch surfaces. The cap will be based upon historical data regarding that creator’s non-viewbotted traffic. Repeated violations will result in longer penalties. Streamers will be notified when an enforcement is applied, along with the duration of the penalty, and can appeal through the appeals portal.
While streamers will be notified, we will not make a follow-on announcement when we begin issuing these enforcements, and will not publicly share details about when and where these enforcements are applied. Unfortunately, providing details simply makes it easier for companies to work around our interventions.
We believe this approach will help us make meaningful progress against viewbotting. We will continue refining our systems and expand when we apply these enforcements over time.
- Dan Clancy
X Communities broke yesterday.
not because the feature is being removed.
but because creators got another reminder that they don't actually have control.
no control over reaching their followings.
no control over the community features.
and no control over taking your community elsewhere.
the biggest streamers with the biggest communities on here said it themselves:
@Trainwreckz: "anyone who has chosen to follow me, my tweets should go out to them."
@IShowSpeed: “i personally don’t think this is a good idea i have a community with over 100k in there and it allows me to talk with my community”
@Clix: “Tf is this decision, how am I gonna put my community in a 350 person groupchat”
@xQc: “Don't remove it, I need this. Also, it's a good product, the new product, no one will use.”
@KSI: "Honestly please don't do this."
@StableRonaldo: "We must migrate."
@Lacy, @KaiCenat, @YourRAGEz chiming in much of the same.
this isn't just an X problem. it’s most platforms.
because their incentives don't align with creators.
which means they don't align with communities.
which leads to failing both simultaneously.
we're heading toward a world where creators and the communities they've built will be at the mercy of platforms with no fallback permanently.
that’s why this new paradigm shift to the InterSocial Graph…
Interest Graph for organic growth PLUS
Social Graph so you reach your community directly…
is the single greatest opportunity creators have ever had to set themselves free:
DOZENS more groundbreaking insights into AI generated music during our full interview here: https://t.co/uOLblJRDCm
To avoid falling behind on the latest AI secrets, make sure you're signed up for the newsletter: https://t.co/PiNimYh09L
Follow for more exclusive AI interviews
EXCLUSIVE: I sat down with the creator of the viral AI Drake song “Winters Cold”.
This is LVCCI’s first and ONLY interview since the song dropped and has been shared MILLIONS of times across social media.
Here are 5 of the dozens of shocking insights from the creator himself 🧵
5. Spotify, Soundcloud or some other service MUST have a separate AI catalog
Myth: Spotify and other music platforms will continue taking down AI generated music
Reality: FAILURE to adapt to the clear demand from users for AI generated music, will birth a competitor who will
First, we’re launching a monthly Midjourney magazine! Every issue has interviews, breathtaking images, and of course, prompts. Join us in celebrating our community's creativity, imagination, and the increasing manifestation of the human mind. Get it here: https://t.co/EFBeJqd61j
Turns out this may not have been GPT-4 but even if it was, probably not a good way to validate if it can pull the known answer from the internet
Better approach would be vulnerabilities of a customized contract that you know the vulnerabilities of and is NOT publicly known alrdy
I dumped a live Ethereum contract into GPT-4.
In an instant, it highlighted a number of security vulnerabilities and pointed out surface areas where the contract could be exploited. It then verified a specific way I could exploit the contract
If you enjoyed this, follow me
@davidpaykin
I share free marketing tips after doing it myself:
Empowering you to build, grow and scale your dreams.
Here’s my 40-page free guide to help you:
https://t.co/7HMCxqiKdK
I grew my TikTok from 0 to 1 million
in 5 months at the end of 2020.
If @elonmusk brings back Vine, there
will be a golden opportunity for you
to do the same.
Here are 5 keys you need to succeed:
5. A call to action can make or BREAK you.
Calls to Action can be:
• Check profile for X
• Share for Y
• Like for Z
Add your call to action at the END
to MAXIMIZE video watch time.