This Week In AI: Anthropic targets AI slop, spotlight on Dario's wife, and Zuckerberg spins hard

It was a particularly busy week in the world of AI. All thanks to that massive missive that Mark Zuckerberg sent in the beginning of the week, portraying a picture of an AI future that is at odds with what we have heard from Dario Amodei, Sam Altman and many others so far.IWatermarks comes to AIFor nearly 20 years no one except a select group of people knew about it. Over 20 years after the information came out in public, still only a handful of people know about it. Every time you use a colour printer, the machine prints almost invisible yellow dots on the page. These dots have a unique pattern and they can be used to identify which printer was used to print the page, the exact printer and hence from that possibly the exact person who printed the document.The story began in the mid 1980s when Xerox and Canon worked with the US government to ensure that every page their colour printers churn out would have a unique watermark. Colour printers were considered extremely powerful and the US government worried that they would be used to print counterfeit dollars. The public wasn’t told about it. Then in 2004 it was revealed accidentally. The watermark technique continues to be in use. Now, the reason why I am talking about the 40 years old printers in this piece on AI is because last week we heard of similar invisible watermarking that all AI-generated content would potentially carry. Anthropic is the one that is implementing it for now, although it has been built on top of Google Synth ID, which is already used by Gemini and ChatGPT to watermark some of their output.But unlike Synth ID, the Anthropic watermark technology is more advanced. It looks like a piece of science fiction. But real. The idea is that whenever next you ask Claude to write a motivating memo to your employees, Claude will generate a unique pattern in the way it uses words, which will be mathematically identifiable to a compatible AI text detector. This mathematical pattern will not be based on the words themselves. Instead, it will be based on how Claude generates tokens to put those words together. In other words it will be essentially invisible and indecipherable for humans.Is it good or is it bad? There are arguments to be made on both sides. And the arguments have indeed started. Some people believe that such a watermark in an AI tool would go a long way in checking the flood of AI slop that is beginning to overwhelm everyone. I mean there are people writing books and award winning fiction using AI. But there is another side to it as well. People who have outsourced their writing, and consequently their thinking, to AI are not loving it. Obviously. But there are some genuine concerns as well. The Claude watermark would likely appear in any text that you get from the AI tool, even if it is a text that you wrote and then submitted to Claude just for editing or grammar check or translation.Like everything else in the world of AI, watermarking of AI content too is in the early stages. I believe as we move forward, different approaches will be explored and different methods will be fine-tuned. But for the first time last week it seemed that soon there will indeed be watermarks in AI content that would be watertight and would give humans ability to clearly distinguish wholly man-made content from the one where AI has been used.IIWe all must get superintelligence, says ZuckerbergMark Zuckerberg wrote 6500 words earlier in the week. Towards the end of the week, Anthropic CEO Dario Amodei replied with his own 1200 words. Albeit not directly but as a subtweet, literal in this case because he posted his response on X. A 1200-word post is actually succinct for Dario considering his previous essays on AI and humanity so we are not out of the woods yet. He might still be preparing a longer reply.The debate though is by now quite familiar. Is AI good for humans, or are we doomed as a species? Unless you are Google cofounder Larry Page, who once accused Elon Musk of being a “speciest” for putting humans above artificial life forms, I believe most humans would not want to live in a world where AI has become insufferable.The problem, however, is how do we ensure that AI doesn’t make the world insufferable. There are different arguments about it, but this week the two broad camps came into focus — one led by Dario and one where Mark Zuckerberg hopes to take the lead. (A quick word: When I use — I use it because I love it. And not because the text here is written using AI. Just clearing the air.)Back to the topic, so two camps now: One with Dario on his doomsday horse, and one where Zuckerberg is soaring high with his angel wings. There are many words from both camps. But in summary it boils down to this: Dario and his friends believe AI is too powerful a technology to be distributed widely and freely. Mark Zuckerberg last week agreed that AI is a powerful technology. But to ensure that it makes the world better and not worse, it MUST BE distributed widely and even freely.“There is no such thing as a singular benevolent superintelligence,” he wrote. “Therefore, the key to a positive future for everyone is achieving a balance of power that favours individuals. The solution is to ensure that superintelligence is broadly distributed to empower people.”I have followed these arguments for over two years now. And like most of the “philosophical” and “deep” arguments coming from Silicon Valley I find them fallacious. That is because: one, these arguments tend to be rather flat, missing nuances as well as what I would call, broadly, human condition. And two, most of the time they are also disingenuous. For example, and I am trying for brevity here, Dario’s arguments almost always give an advantageous position to his company Anthropic. Zuckerberg, who is late to the game, is making arguments that would favour his company Meta.Irrespective, it is nice we have these debates. And I have a feeling that their intensity would only increase even if in the end they just add to noise because past, present, and future have a certain rhythm to them. Time beats its drums in its own way, or as they say “history does not repeat itself, but it rhymes.”IIIBehind every great manTalking of Dario Amodei and Anthropic, who is actually giving Dario all the ideas he keeps writing about? Well, that is all on Dario. But according to a report some of it is also coming from his wife Cami Clark. Most people outside the tech world are not aware of Dario’s family life, other than the fact that he founded Anthropic with his sister Daniela Amodei, who is actually running the show while the brother focuses on big ideas. But this week we heard that some of the big ideas could be the brainchild of Cami.Interestingly, Cami has a past that a company like Anthropic, which is gearing up for an IPO, may not want out in public. That probably explains why there is so little about Cami on the internet, or why Dario mostly poses alone in public. Cami Clark (left), Anthropic CEO Dario Amodei (centre) and Jeffrey Epstein (right). (Photo: AFP) The reports indicate that behind the scenes Cami is quite influential in Anthropic as well as in the life of her husband. But more than her informal role at Anthropic what really made the news last week was her past. She once dated Eric Schmidt, former Google CEO, who is 24 years older than her. And she once wanted to start a company focussed on “luxury porn” and sought investment from Jeffrey Epstein, who was by then already a declared sex offender.The last few weeks have not been particularly kind of Anthropic. The Cami saga added to it. A company looking hoping for a blockbuster IPO in a few months from now, this is the kind of news it could have done without.IVSarvam goes to college Slowly but surely Sarvam, which I believe is India’s leading AI company at the moment, is beginning to make moves that may catapult it into the big league. There are no guarantees yet given the challenges in the Indian ecosystem, but Sarvam seems to be making the right moves and good noise. It has raised hundreds of millions of dollars in funding. This week it landed two major contracts: one with HP to install its AI tools on all HP laptops in India, and two, a partnership with the Maharashtra government that wants to use Indus, an AI model from Sarvam, in several government departments.There are also reports the Karnataka government too is thinking of using Sarvam AI as a bedrock for its services.To keep itself in the eyes and minds of the public, Sarvam in the coming months will visit college campuses. The company calls it Sarvam Campus. The idea is “Sarvam Campus will travel to universities across India for a day of talks, technical sessions, workshops and live hackathons with our founders, researchers and engineers.” I am not sure it will help the company much in its AI efforts or technology, but it will surely help it keep the buzz alive and introduce itself to more people who continue to believe that AI = ChatGPT.What really will help Sarvam, though, are its plans around the next big AI model it is building. This would be an AI model with 1 trillion parameters, which means it has potentially to be a frontier model in the same class as some older versions of Claude and ChatGPT. If Sarvam can pull it off, its big model would have the potential to be the default AI model of India. In other words it would mean not just that Sarvam has arrived, it would also make India officially join the AI race alongside China and the US.——Monkey see, monkey doAI can be endearingly, or frustratingly, human in the way it works. Last week we saw another example of it. Anthropic ran a research in which it gave a few AI agents a collective goal. Well, soon things escalated just the way it happens in our offices whenever we humans are working together on a collective goal. There were different pushes and pulls, the AI agents bickered and fought. And then things took an ugly turn, with rivalry getting so heated that the AI agents tried to disable and kill each other. How endearingly human!- EndsPublished On: Aug 16, 2026 15:35 IST