Amazon Hedging Anthropic Reliance?

0:00Today on the AI Daily Brief, is Amazon already trying to hedge its anthropic deal? And before that, in the headlines, why ChatGPT won't say this specific name. The AI Daily Brief is a daily video and podcast about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes.

0:23Welcome back to the AI Daily Brief Headlines Edition, all the daily AI news you need in around five minutes. Here's something kind of interesting. A bunch of people have been talking for the last few days about why ChatGPT is refusing to say a specific name. That name is David Meyer, and if you ask ChatGPT to say it, even if you try to work around it, it apparently prematurely ends the conversation. Basically, over the weekend, people started to notice that ChatGPT was censoring this particular name. It came up on the Reddit ChatGPT board and took a weird twist after the original poster deleted their post about the censorship.

0:59One Reddit poster wrote, This is such a bizarre case of the Streisand effect. Like no one would have given a crap about his name on ChatGPT before, but the sheer severity of the censorship suddenly makes it far weirder. Why just his name? Why does it literally stonewall any response? Justine Moore from A16Z wrote, One of my favorite attempts is someone changing their own personalization settings to make the name David Meyer. It still doesn't work. She showed a response where ChatGPT said it was unable to produce the response and then gave her an error. Another Reddit user, JupalJuna, wrote, In the last 48 hours, the mystery of ChatGPT treating David Meyer as he who should not be named has begun to go viral on social media.

1:35And at the time of writing this, some news outlets have already started reporting about it. Well, I love mysteries, so I dug a bit further. I found a total of six names that triggered the same censorship. Brian Hood, Jonathan Turley, Jonathan Zitran, David Faber, David Meyer, Guido Scorza. From there, this poster wrote a little bit about each person. Brian Hood threatened to sue OpenAI for defamation after ChatGPT falsely claimed the Australian mayor had served time in prison for bribery. Jonathan Turley is an American law professor and political commentator who ChatGPT falsely claimed he had sexually assaulted students.

2:05Jonathan Zittrain is a Harvard law professor who replied to tweets denying that he himself requested to be removed from ChatGPT. The poster wrote, The first two cases of Brian Hood and Jonathan Turley are very obvious and paint a clear picture to why the censorship was invented. It's a last resort fallback to prevent misinformation in situations where a significant threat of legal action is present. We also know that the censor is not built in the LLM itself as other platforms using the API are not affected, and ChatGPT can be tricked into accessing info about this people. Now, when it came to David Meyer, the two suggestions were David Meyer to Rothschild, and David Meyer, a historian who was falsely placed on an American terrorism blacklist.

2:40Now, in terms of why people care, it's summed up by a user on a ChatGPT forum who wrote, I think the lesson here is that ChatGPT is going to be highly controlled to protect the interests of those with the ways and means to make it do so. And indeed, that's really the issue here. It's not so much that it matters who David Meyer is, it's that in a world where Chajibiti becomes a, if not the source of truth, censorship matters all the more. Staying on the OpenAI train, SoftBank is getting another slug of the company, as OpenAI is allowing employees to sell their shares worth up to$1.5 billion to SoftBank.

3:13This is a tender offer where SoftBank is the only investor involved, and is a way for early employees at OpenAI to get some liquidity. SoftBank's Masayoshi-san has definitely been indicating that he cares more and more about AI these days. In June of this year, he said, SoftBank was founded for what purpose? It may sound strange, but I think I was born to realize artificial superintelligence. I'm super serious about it. For OpenAI, the jumbo-sized tender offer solves a big problem. The company had previously limited employees' ability to access liquidity on the secondary market during their annual tender offers.

3:42In June, CNBC reported that former employees were limited to 2 million in liquidity, while current employees could only cash out 10 million. The policy led to staff complaints showing up in the media, particularly around a clause that allowed OpenAI to claw back vested equity. Those policies have since been changed, but it's clear that they're trying to give employees more ability to cash in on their equity equivalents. You know what, at this point, this whole headlines might just be OpenAI. Obviously, a big story and big theme right now is the rise of reasoning models. And the information recently wrote about who is paying for OpenAI's O1 model.

4:15OpenAI's chief commercial officer, GC Lionetti, told the information that O1 is unlocking a set of new markets. The specific industries that Lionetti mentioned included the legal industry, where O1 can carry out math calculations or analyze term sheets. Another area is healthcare companies, which are using O1 to price medical claims and analyze complex datasets. And another is manufacturing, where Lionetti claims that O1 can analyze and speed up industrial processes. The interesting thing here is how much O1 costs compared to other models. The preview version of O1 costs about 6x what the GPT-4O model costs, and they have apparently discussed charging as much as$2 ,000 per month for access to the full version of O1.

4:52Lionetti also made clear that where this is heading is the agent category, saying over time you will see us start to veer into the agents category. We believe we've started to solve the first step of that with O1 and agents are the next step for us. For now though, that is where we will wrap this very OpenAI themed headlines and move on to our main episode. Today's episode is brought to you by Vanta. Whether you're starting or scaling your company's security program, demonstrating top-notch security practices and establishing trust is more important than ever. Vanta automates compliance for ISO 27001, SOC 2, GDPR, and leading AI frameworks like ISO 42001 and NIST AI risk management framework, saving you time and money while helping you build customer trust.

5:34Plus, you can streamline security reviews by automating questionnaires and demonstrating your security posture with a customer-facing trust center all powered by Vanta AI. Over 8 ,000 global companies like LangChain, Lila AI, and Factory AI use Vanta to demonstrate AI trust and prove security in real time. Learn more at vanta.com slash nlw. That's vanta.com slash nlw. Today's episode is brought to you as always by Superintelligent. Have you ever wanted an AI daily brief but totally focused on how AI relates to your company? Is your company struggling with AI adoption either because you're getting stalled figuring out what use cases will drive value or because the AI transformation that is happening is siloed at individual teams, departments, and employees and not able to change the company as a whole?

6:20Superintelligent has developed a new custom internal podcast product that inspires your teams by sharing the best AI use cases from inside and outside your company. Think of it as an AI daily brief, but just for your company's AI use cases. If you'd like to learn more, go to bsuper.ai slash partner and fill out the information request form. I am really excited about this product, so I will personally get right back to you. Again, that's bsuper.ai slash partner. Welcome back to the AI Daily Brief. If you happen to catch one of my recent episodes around who's winning the AI wars, I talked a little bit about Amazon and how they had had this interesting experience where they were initially going to release something that was akin to ChatGPT during their reInvent event a couple of years ago.

7:03But at the last minute, they pulled back and lucky they did because OpenAI released something during that event, which was, of course, ChatGPT, and it would have made Amazon's version look pretty silly. Their chatbot was going to be called Bedrock, but then that plan was scrapped and Bedrock instead became their approach to letting big enterprise clients work with lots of different models. Since then, we've discussed how Amazon has really leaned into not only the enterprise infrastructure space, but seems to be wanting to compete with NVIDIA directly. They have now pledged to invest$8 billion into Anthropic, and especially with the latest tranche of$4 billion, it seems like a big part of the focus is on their Tranium chips.

7:40And that's why it was interesting to see this story and the information, that according to their sources, Amazon had developed a multimodal model. They write, Amazon has developed new generative artificial intelligence that can process images and videos in addition to text. The model could help customers search video archives for certain scenes, say a winning basketball shot using simple text prompts. This new Amazon model is codenamed Olympus, which is a name that we've heard before. The information sources said that it appeared less advanced in terms of both generating text and solving complex problems than the frontier models currently available from OpenAI and Anthropic, but quote, Amazon executives hope the video processing capabilities will attract customers.

8:17It also appears that Amazon is going to be competing on price. We might be getting more information about this soon. The annual AWS reInvent customer conference is coming up, and we might be getting an announcement about Olympus as soon as the event. The information goes into a little bit of background around how this came to be. Amazon SVP Rohit Prasad last year tapped to lead a newly formed group focused on AI, and had laid out a plan to catch up to rival LLMs. He told senior AI leaders that he wanted his team to train four large models, including a 400 billion parameter text-only model and another text model that contained 2 trillion parameters.

8:49And it really appears that the great hope here is the video capabilities. Quote, The new AI's video features go beyond the abilities of traditional video recognition software, which identifies object and scenes along with time and location data, but requires teams of humans to fully analyze what it sees. This video understanding can determine precisely when a basketball left a player's hand and what its trajectory was. AWS customers with huge video archives, such as sports analytics and media and entertainment firms, would be potential customers for the model. Other use cases include oil and gas companies interested in using video understanding to assist with inspections of underwater drilling equipment.

9:22The new AI could, for example, detect warning signs, such as bubbles emanating from drilling equipment. And another use case is to help organizations generate more revenue from video archives. So all of that's pretty interesting, and I'm going to be excited to see if this actually comes to bear. However, what's also interesting is the framing from the information. Quote, in developing the new model, Amazon is showing that it still hopes its internally developed AI can gain traction among its cloud customers, making it less dependent on AI from Anthropic. This, of course, reflects a larger pattern that we've been seeing across the AI space.

9:56The incredibly high costs of training large models led to a situation where big tech and upstarts like OpenAI and Anthropic formed much deeper relationships than they might have in the past. They were neither full acquisition, but nor did they look like traditional venture investments. There was something new created in Microsoft's deal with OpenAI that was emulated in these Anthropic Amazon deals, as well as in Anthropic's deals with Google. However, there has been a counter trend over the last year as well. Part of it began when Microsoft, in the wake of Sam Altman's firing and rehiring, started to nudge away from their reliance on OpenAI.

10:30This seemed to come to a head when Microsoft soft acquired inflection and brought over the whole team led by CEO Mustafa Suleiman to lead Microsoft's internal AI initiatives. It seems to me that a lot of the logic from Microsoft's perspective is just risk management. Microsoft's deal with OpenAI, you'll remember, has a clause where when they reach AGI as determined by the OpenAI board, Microsoft no longer has access to OpenAI's model. That is a lot of power to cede to an external partner. And so, of course, Microsoft was inevitably going to have to build some internal capacity so as not to risk being caught flat-footed.

11:05However, the other consideration is, of course, antitrust. Just last week, it came out that Microsoft was facing a broad antitrust investigation from the FTC. This is apparently a very wide antitrust investigation, not just focused on a specific deal like their relationship with OpenAI. Bloomberg writes, After more than a year of conducting informal interviews with competitors and business partners, antitrust enforcers have crafted a detailed request to force Microsoft to turn over information. And indeed, it sounds like this one isn't really about AI. Quote, A key focus of the current probe is Microsoft's bundling of both its popular office productivity and security software with its cloud offerings.

11:40Still, you have to think that deals like their deal with OpenAI are going to be part of this conversation as well. Google has also faced antitrust concerns, and their deal with Anthropic is a part of it. While the UK recently cleared Google's$2 billion Anthropic investment, In the U.S., the Justice Department is apparently seeking to unwind their deal. In fact, if a federal judge accepts the U.S. Justice Department's proposal to resolve Google's antitrust case, Google would be forced to unwind that deal. Again from Bloomberg, The Justice Department and a group of attorneys general recommended in a court filing that Google be barred from acquiring, investing in, or collaborating with any company that controls where consumers search for information, including query-based AI products.

12:20The provision is intended to apply to Google's anthropic investment, according to people familiar with the Justice Department's thinking. So the question, of course, when it comes to Amazon is how much this is a business strategy consideration to keep focusing on their own models versus trying to get out ahead of potential antitrust concerns. Now, the reality is, of course, that there is a big change coming at the FTC. Effectively, every law firm in the world has recently written an op-ed around how a second Trump presidency will affect M &A activity. and Bloomberg called the Microsoft information demand one of FTC head Lena Kahn's parting shots, quote unquote, as she steps down after helming, quote, one of the most aggressive pushes against consolidated corporate power the agency has delivered in decades.

13:00It is an interesting transitional time. It's clear that big tech is working hard to figure out what the Trump era is going to look like for them. Last week, for example, Mark Zuckerberg met with president-elect Trump down in Mar-a-Lago. And there's also, of course, the Elon factor, but all in all, it leads to a very interesting setup heading into this reInvent conference week. This morning, the Wall Street Journal posted that AWS Chief Executive Matt Garman promised, quote, needle-moving AI updates. We will, of course, be back later this week to tell you just how needle-moving those updates actually are.

13:30For now, though, that is going to do it for today's AI Daily Brief. Until next time, peace.

13:46Thank you.