So I heard we're talking about Anthropic's watermarks...

OpenAI is doing the same thing but they’re being much quieter about it (and being less transparent about detail) Funny you don’t talk about it

No hate but this ain’t YT bro just post the link

I can't bring myself to watch 30minutes of theo yapping but does he just call out ant? All the labs are watermarking now. Including OAI

The problem with Anthropic isn’t that they’re watermarking anything It’s that they are insanely untrustworthy

@Ph4seOn3 i love the intro LOL

They scraped and now watermarking

Im so glad not being an anthropic customer you guys are incredibly enduring

every theo video starts with a question mark and ends with a sponsor read, works though

Honestly I just hope it doesn't break the vibe of my AI writing side projects lol. Curious how strict they'll be with smaller apps like mine.

Got some weird looks from my girlfriend when she walked in & saw the Theo face foot on the screen

Claude eu drops after august 2 2026 carry the mark

If I turn an AI generated text into an image or use AI generated images to create a video the watermarks may not survive. Similarly if the watermark relies on metadata, simple screenshot can break the watermark. I think there's still a lot of work needs to be done.

These images are real guys , I took them :/

Lmao which rockstar from Queens took over theo's account?

Watermarks get stripped in seconds by anyone motivated to remove them. This is for optics, not safety.

Shipping quiet infrastructure upgrades that quietly fix agent reliability builds more long-term developer trust than hype cycles

how much money do we raise to achieve buzzcutheo

Guys please read up on Article 50 of the EU AI Act. Every company has to do this Anthropic is just being open about it.

Both Anthropic and OpenAI are so silent lately. Do you have any insider access to some new models? 👀 I mean you cannot tell but still. 👀

how do they even pick which tokens get watermarked? semantic-level or just statistical patterns?

fun fact: openai has a website for detecting images containing their own watermarks

我很不理解这个行为，为什么要加水印？难道网上搜索的信息经过他们的模型也当作是他们家的？很不理解

Yeah, it's a fascinating shift, especially how it works without changing the text quality. We actually broke down the full technical and compliance angle here:

Yessir do u think it can be edited in the future tho

Anthropic "hater" follows @DarioAmodei btw

By the way, did you know that all modern printers print out a unique ID on every printed sheet of paper? 🤔

i will never get used to seeing full-length videos on Twitter/X. Great video nonetheless.

One of the problems is that a large number of the people who advocate for this are just unconditionally against AI. Which makes any debate on solutions disingenuous. Because the person doesn't care if a "solution" destroys the product. They in fact welcome it as a perk.

I love how the only thing you asserted as unquestionably real is the picture of the foot with your face on it

Flagging human written text makes so much more sense. Is it realistically possible though?

I don't mean to backseat drive, but does this not seem like a huge waste of effort? How hard would it be to have a 4B thesaurus agent rephrase the watermark out?

🤣🤣🤣 upscale and downscale

u think it actually makes a difference for the trust issue or is it just more bloat for the models

This is one of the dumbest things I’ve seen done… at this point we need to ban the EU off everything and leave them on the Stone Age where they want to live. They literally hold back technology all the time. Their government try to control everything it’s nasty 🤮