With New Deal, Anthropic Becomes Even More Key to Amazon's AI Strategy

0:00Today on the AI Daily Brief, Amazon makes another$4 billion investment in Anthropoc And the CEO of JP Morgan thinks that we'll have a three and a half day work week in just a generation to come. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes.

0:24Welcome back to the AI Daily Brief Headlines Edition. All the daily AI news you need in around five minutes. In the US, this is a short week for Thanksgiving, which is on Thursday. and the CEO of JP Morgan, Jamie Dimon, thinks that this might not be that different than a normal work week in the future. In an interview on Bloomberg TV, Dimon said, your children are going to live to 100 and not have cancer because of technology. And literally they'll probably be working three and a half days a week. Dimon's predictions are backed up, at least in practice, by just how much JP Morgan has gone in on AI.

0:55He has previously called it, quote, critical to our company's future success, dedicated an entire section to AI in his shareholder letter this year, and said that J.P. Morgan already has more than 300 use cases in production. One of the things I talk about a lot here is the need for a discussion around a new social contract. And this is exactly what I mean by that. One of the major questions, if AI does replace a huge amount of human labor, is whether we're going to stay on the same paradigm of how much we work and just fill it with the same amount of time on our new, more advanced tasks, or if we actually get comfortable saying working a smaller amount is enough to have contributed to society.

1:31It's hard to have that conversation in the abstract, and it's hard to imagine a future that doesn't look exactly like today. But these are the types of conversations we are simply going to have to have in the years ahead. My prediction for the AI feature, which most feels insane to those of us now who have grown up in a very specific era of computing, but will feel completely and absolutely normal for people in the future, is the type of constant surveillance represented by Microsoft's recall. The controversial feature takes screenshots of everything you do on your computer to create a searchable database of memories.

2:02AI is used to index images and text from the screenshots and power the search function. The company announced this some time ago, but has finally released their first preview version. A couple things to note. One, they say that Recall is entirely optional. You have to opt into it. Second, they say that they've provided strong controls over privacy and security. Users have full access to all of the screenshots and can manually delete them as necessary. The feature can also be configured to exclude certain apps and websites from being recorded. Can't imagine what they're thinking with that. And Microsoft also says that personal details like credit card info, passwords, and ID documents can be automatically detected so those snapshots are not saved.

2:38Finally, Microsoft won't have any access to any recall snapshots. They're not sent to the cloud or used to train Microsoft's AI models. Now, the reason that I think that this is going to become completely normalized is one, that the general pattern on the internet has been that we get more comfortable with surveillance. But two, if they can solve for these security and privacy concerns, the number of applications that this opens up is immense. This is so potentially useful. And even beyond just the search application that Microsoft is thinking about, there are many, many opportunities to use this sort of total information about what someone is doing to create customized products and services around them that could be really, really powerful.

3:18The question just, of course, is what people get comfortable with. And for that, we will just have to wait and see. Next up, an update on an AI-powered device that had a ton of excitement and then fell off quite quickly. AI Wearable Rabbit says that they're now rolling out a new agentic upgrade. Users of Rabbit R1 can now teach it to perform certain tasks with a feature they're calling teach mode. The R1 can learn through demonstration. For example, a user could show the device how to fetch social media updates or save a song to your Spotify account. Using the web interface, users create a lesson by describing the task and then record themselves performing it.

3:51The R1 can then recall the lesson on command. This is something that they had promised when they first announced the R1, but which wasn't immediately available. The feature has come online after the R1 received a big update last month with the addition of automated website browsing. Some reports criticized Teach Mode for being a little laborious, but also recognized that no-code agent training is a powerful idea. Fortune, meanwhile, highlighted how common experimental and unpredictable AI features are becoming. Rabbit CEO Jesse Liu defended the practice, stating, You have to kind of encounter all the edge cases and tweak on the fly and continue.

4:22That's just the whole nature of developing with AI models. He pointed out that Rabbit doesn't have a 10-year runway or the ability to fully test edge cases, saying, we have to make sure that we take our shot and move fast. The question, of course, is whether move fast and break things is an appropriate mantra for the generative AI world. Lastly today, another Chinese lab has claimed a major breakthrough in the use of reasoning models and inference time scaling. Last week, DeepSeek claimed they have produced a text-based reasoning model that exceeded the capabilities of OpenAI's O1 preview model.

4:50They said they merely took OpenAI's chain of thought logic and added more time, demonstrating that this approach to scaling could be viable. Now, a consortium of Chinese universities have produced an image-capable model called Lava-01 based on the same principles. Chain of thought prompting has been used for visual language models or VLMs in the past, but it generally produced only marginal gains. The issue has been that VLMs can struggle when the chain of thought is not sufficiently systematic or structured, often getting lost and losing track of the specific problem they're trying to solve. The researchers behind this new model wrote, We observe that VLMs often initiate responses without adequately organizing the problem and the available information.

5:23Moreover, they frequently deviate from a logical reasoning towards conclusions, instead of presenting a conclusion prematurely and subsequently attempting to justify it. The researchers of Lava01 took a similar approach to OpenAI's 01, breaking the reasoning process down into four steps. The model first provides a high-level summary of the problem it's being asked to solve. Next, the model captions the image input, describing the relevant parts and focusing on elements related to the question. The model then performs structured logical reasoning to produce a preliminary answer. Finally, the model presents a summary of the answer based on the prior reasoning step.

5:52Only the final step is visible to the user, with the rest taking place behind the scenes. The researchers wrote, It is the structured output design of Lava-01 that makes this approach feasible, enabling efficient and accurate verification at each stage. This validates the effectiveness of structured output in improving inference time scaling. And while there is excitement about the possibilities, Rabia Shakur also pointed out, love how the Lava01 paper releases zero information about their training, no code, no annotated dataset. Super helpful. Still, there is clearly a lot going on with the inference and test time compute approach to scaling, so anticipate some more developments there.

6:26That, however, is going to do it for today's AI Daily Brief Headlines edition. Next up, the 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. Plus, you can streamline security reviews by automating questionnaires and demonstrating your security posture with a customer-facing trust center all powered by Vanta AI.

7:02Over 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 by Superintelligent. Every single business workflow and function is being remade and reimagined with artificial intelligence. There is a huge challenge, however, of going from the potential of AI to actually capturing that value. And that gap is what Superintelligent is dedicated to filling. Superintelligent accelerates AI adoption and engagement to help teams actually use AI to increase productivity and drive business value.

7:41An interactive AI use case registry gives your company full visibility into how people are using artificial intelligence right now. Pair that with capabilities building content in the form of tutorials, learning paths, and a use case library. And Superintelligent helps people inside your company show how they're getting value out of AI while providing resources for people to put that inspiration into action. The next three teams that sign up with 100 or more seats are going to get free embedded consulting. That's a process by which our super intelligent team sits with your organization, figures out the specific use cases that matter most to you, and helps actually ensure support for adoption of those use cases to drive real value.

8:20Go to bsuper.ai to learn more about this AI enablement network. And now back to the show. Welcome back to the AI Daily Brief. This headline from The Information just about says it all. Anthropic keeps Amazon in AI contest. The news specifically is that Amazon announced that it's investing$4 billion more into Anthropic. This is the second$4 billion investment Amazon has made into the company. Anthropic noted that Amazon will remain a minority investor. And separately, Anthropic is still raising money from other investors at a valuation of up to$40 billion, dollars, which notably is less than the money that Elon was able to raise at for XAI.

8:58What we're interested in discussing today, though, is not just the terms of the deal, but how it reflects where Amazon sits in the entire AI conversation. Let's go back to the beginning of this generative AI era with the launch of ChatGPT. Before OpenAI announced ChatGPT, Amazon Web Services had been developing something pretty similar. They were planning to announce their own version of a chatbot, which they had been calling bedrock inside the company at their annual conference in November 2022. Due to some technical issues, however, at the last minute, they decided to postpone their plans, which ended up being a pretty good thing because right in the middle of that conference, OpenAI released ChatGPT, which as we know, exploded, became the fastest software to get to 100 million users, and so totally has defined the era of generative AI that many normal people know it as ChatGPT, not AI.

9:48Apparently, when Amazon saw what ChatGPT was offering, they realized that their own version was so far behind that they had to totally shift their plans. And that's when Bedrock became the Bedrock that we know it as now, which is a service that allows developers to connect their cloud applications to a variety of LLMs. So the point of all of this is that Amazon has had to be pretty nimble and has been coming from a position of feeling behind for the entire last couple of years. And that's the context in which we got both the first Amazon Anthropic investment, which, as you might remember, was largely about a collaboration around Amazon's custom chips.

10:19And it's where this new deal came from. Back at the beginning of the month, the information started reporting that this investment appeared to be in the works. On November 7th, the information wrote, The new deal is similar to Amazon's initial$4 billion investment, which was struck last year, but this time Amazon wants Anthropic to make a concession. The cloud giant is asking Anthropic, which uses Amazon's cloud services to train its AI, to use a large number of servers powered by chips developed by Amazon. The problem is that Anthropic prefers to use Amazon servers powered by NVIDIA-designed AI chips.

10:49The information continued, the size of Amazon's total investment could depend on the outcome of this discussion, specifically on the number of Amazon chips Anthropic agrees to use. So why would this be a big issue? Part of it is actually the software. AI developers are used to NVIDIA's CUDA software, whereas the Amazon software the developers have to use with their Tranium chips isn't as advanced. Anthropic also might have had concerns around being locked into the Amazon ecosystem. Interestingly, the conversation between Amazon and Anthropic has apparently been ongoing. Once again from the information, quote, a senior Amazon official privately said Anthropic CEO Dario Amadei earlier this year discussed his interest in using a large-scale AI data center server cluster to develop technology, similar to the ambitious data center plans of rivals such as Elon Musk's XAI and OpenAI.

11:35And so that was the setup going into this deal. We didn't get perfect information about all the aspects of the deal, but reports suggest that Anthropic will now, quote, rely more heavily on AWS for training its models. They also did say that they will be using the Tranium chips, although exactly how many wasn't clear. A short announcement from Anthropic was called Powering the Next Generation of AI Development with AWS. Anthropic said that this quote, establishes AWS as our primary cloud and training partner. When it comes to AWS Tranium, they basically say that they are working closely with Annapurna Labs at AWS on the development and optimization of future generations of Tranium, although this does sort of skirt how much their commitment to use the chips is right now.

12:15While the details in the announcement itself might be a little sparse, it's very clear from how Amazon is talking about this where they want to place the emphasis. CEO Andy Jassy wrote on threads, excited about our deepening partnership with Anthropic. About a year ago, AWS became Anthropic's primary cloud partner. Today, we become Anthropic's primary foundation model training partner. Anthropic is excited about the price performance advantage that Tranium offers, and we're grateful for this confidence in partnership. In other words, this is about Tranium. It is about Amazon carving out a place in the future of AI infrastructure, not just ceding the entire chip space to NVIDIA.

12:52Some commentators think it's going to work. Andrew Carr writes, I know it seems like Amazon is eating Anthropic. However, I think the subtitle here of deep technical collaboration on directly interfacing with Tranium Silicon to improve the story of AWS chips could actually be immense. If this works as well as I think it will, AWS will maintain its market share and begin winning more AI training workloads from other cloud providers. But what about Anthropic themselves? Where are their motivations? The company has had a very good year. Menlo's recent report on the state of generative AI in the enterprise found that among enterprise customers, OpenAI's market share had dropped from 50 % to 34 % over the last year, while Anthropics had doubled from 12 % to 24%.

13:30That is a significant boost that firmly positions Anthropics in the number two slot and nipping at OpenAI's heels. And yet in spite of that, OpenAI has bigger brand awareness, is better capitalized, and of course has the heft of the Microsoft partnership, as well as now the Apple partnership, all of which gives it some serious advantages. For Anthropic then, digging in on the distribution and enterprise power of Amazon not only makes sense, it might be simply mission critical. Whatever the case, this battle just gets more and more interesting, and we will continue to follow it here at the AI Daily Brief.

14:04For now though, that's going to do it for the episode. Appreciate you listening or watching as always, and until next time, peace. Thank you.