The AI Daily Brief: Artificial Intelligence News and Analysis - Amazon Launches Q Chatbot: Is It A Disappointment or AI On Trend?

Episode Date: November 29, 2023

We didn't get Amazon's rumored "Olympus" model, but we did get an enterprise focused chatbot called Q. NLW explores whether it's a disappointment that continues to show how behind OpenAI the big tech co's are, or whether it actually exemplifies a broader AI trend of workflow integration. Also: is Stability AI facing financial pressure? ABOUT THE AI BREAKDOWN The AI Breakdown helps you understand the most important news and discussions in AI.  Subscribe to The AI Breakdown newsletter: https://theaibreakdown.beehiiv.com/subscribe Subscribe to The AI Breakdown on YouTube: https://www.youtube.com/@TheAIBreakdown Join the community: bit.ly/aibreakdown Learn more: http://breakdown.network/

Today on the AI breakdown, we're looking at Amazon's just-announced Q chatbot. Before then on the brief, is Stability AI in Money Trouble? The AI breakdown is a daily podcast and video about the most important news and discussions in AI. Go to Breakdown.network for more information about our newsletter, our Discord, and our YouTube. Welcome back to the AI breakdown brief, all the AI headline news you need in around five minutes. Stability AI is clearly one of the most prolific companies in the AI space. Every week, it seems like, they release some new model that has people excited and creating new things and generally thinking in new ways about the frontiers of AI's capacity to help people create.

And yet, even with all that, it appears that the company is under some fairly significant financial pressure. Bloomberg today published a piece called Stability AI has explored sale as investor urges CEO to resign. Now, a couple interesting things from this piece. First of all, apparently in a letter to management last month, one of Stability's largest investors, Kowatu Management, called for CEO Amad Mastok to step down. Notably, this demand came just a year after Kowatu helped lead a funding round that made Stability a unicorn. Kowatu's argument in that letter was that, as Bloomberg put it,

Mostok's leadership had prompted several senior managers to leave and place the startup in a tenuous financial position. Now, of course, one thing that we did talk about either last week or the week before that was how someone who had been working on audio projects inside Stability left and did a little mini press tour about how much he disagreed with the company's approach and thinking around copyright. However, beyond just this letter and questions of leadership, there's also some serious dollar and cents type concerns. According to Bloomberg's sources, as of October, stability was burning around $8 million a month compared to revenue that was in the low single digit millions.

In a post on X, Mastok wrote that Stability had made $1.2 million in revenue in August and was on track to make $3 million this month. Obviously, that's a far cry from the $8 million burn. It's also a significant enough gap that even deep coffers could go away fairly quickly. Stability raised $101 million last year and another $50 million in October, that latest time from Intel. But if you're losing between $5 and $7 million a month, even those big eye-popping numbers can go away pretty fast. Apparently surrounding that, Bloomberg has reported that Stability has, quote, presented itself as an acquisition target in recent weeks and held early-stage conversations with multiple companies. Bloomberg's sources said that a deal is not imminent and that the company

could decide not to sell. Bloomberg identifies some of the interested parties as cohere, and Jasper, who themselves have been subject to some amount of fud as they've had to do multiple rounds of layoffs this year. Now, there are two possible, very different dimensions of this story. One is, of course, questions around specific leadership decisions, and that I think is a little bit less interesting for our purposes, one, because we don't have that information, and two, because if that is really what's going on, well, that it's just about personalities and leadership and inner company struggles, all of which are fairly normal, but not necessarily extrapolatable to some larger trend or insight. However, to the extent that there really is a sale process, emanating from

the gap between the cost of running this company and the amount of money it can make, well, then in some ways it might be reflective of one of the real serious difficulties of competing in the AI space right now, which is the incredible cost to do so. Compute is scarce and expensive, talent is scarce and expensive, and all of that adds up to a recipe in which very few players can actively compete. It's not an accident or a coincidence that we've seen a very new type of relationship form between the big tech companies that have the biggest coffers, i.e. the Googles and Microsofts and Amazon's of the world, and the startup AI labs working on frontier models, who even with basically blank checks from VCs, still need even more capital than is

realistically available on Sand Hill Road. This is why some are worried about how AI could increase centralization around big tech. The director of the Consumer Financial Protection Bureau, Rohit Chopra, said in an event on Tuesday that he's, quote, concerned that a handful of firms and individuals could wield enormous control over decisions made throughout the world because of advances in AI. Now, that was part of a larger conversation around the difficulty in regulating AI, but it points, of course, to a reality. Now, of course, to some extent, we're just going to have to wait and see how this all plays out, but it certainly does seem like stability AI is making some moves to shore up their financial situation. The day before this story broke, Amad tweeted out, stability AI memberships.

Recent weeks have shown how alignment of business models is important in AI. We are bringing in stability memberships to help solve this alignment. We're doing okay as a business and ramping nicely. However, we want to move to a more aligned model focusing on our core talent of building great models as the releases over the last year and next period shows. Basically, they're trying to figure out a business model where more people who are interacting with their AI models actually pay them to do so. Now, I will be clear that I hope that they are able to figure it out. Like I said, I have no insights into the inner workings of the company, but what I do know is that they are continually pumping out interesting new tools and features. Last week, we got stable video

diffusion. This week we got SDXL Turbo, which is a nearly real-time text-to-image generation model, that people are lauding as just incredibly fast in practice. Indeed, it is so fast that it starts generating imagery while you're typing the prompt. And while it's totally possible that inside a different home they could still produce just as excellent of work. I'm always biased towards seeing startups stay independent for as long as possible. Now, speaking of startups, a few more interesting little bits of AI startup news. Hey Jen, which is a video avatar creator that I have used in the past. You might have caught a video avatar episode after Elon announced Grock when I was on an anniversary trip down in Mexico. Well, that company has just added 5.6 million in new funding and reported that

their revenue has gone from $1 million in annual recurring revenue in March to $10 million in ARR in August, and now all the way up to $18 million. Now, there's another really interesting part of this story where the company is working to break from its Chinese origins, including replacing a board member from Sequoia, China, with a new board member from Conviction Partners who led this round. But for me, as a user of the product, I'm just glad to see they're getting the resources to keep building. Speaking of startups we use here at the AI breakdown, 11 Labs, the voice synthesis company,

has announced a new grants program. They write, we're helping early stage startups and solopreneurs use voice AI to get innovative projects off the ground. Recipients get 11 million characters a month for three months to develop, test, and scale their products. For those who don't have the instant translation in their heads, 11,000 characters is around 200 hours of generated audio. To qualify, they have to have a monetized product use case.

It can't be a short-term or one-off project. It has to be a startup or small company only, i.e. have no more than 25 full-time employees, and each company can only file one application. Overall, there's definitely a sense of momentum and startup momentum specifically in the AI space with a lot of excitement around image generation and, of course, video generation. We talked about how PICA Labs had raised $35 million and launched their new PICO 1.0, which some have even gone so far as to call AI videos chat GPT moment. Will things actually play out that way?

That, I think, will be a question for 2024. And speaking of battles in 2024, the regulatory battles around AI do nothing but increase in significance. The Senate is getting ready for its next AI forum, which this time will focus on IP and copyright issues. And over in the EU, there are fierce debates right now going on around how foundation and frontier models should be treated in the EU AI Act. Now, this is deserving of an entire show in and of itself. And if you've been on AI Twitter at all, you've probably seen some caustic words around it. But suffice it to say that this will be a central battleground around AI policy in the short term. And it feels like even though we are rounding the corner on the end of the year,

This industry is just not slowing down. Indeed, a mod from Stability AI tweeted earlier this week. So many orgs have new AI-related announcements and releases lined up for the next few weeks, some super amazing stuff coming out. Have we ever seen a pace of cumulative innovation for actually useful stuff this fast? Can barely keep up. That is very true, but I'm glad you're trying to do so here with the AI breakdown. Thanks for hanging out. And next up, the main AI breakdown. Welcome back to the AI breakdown. This week, was the Amazon Reinvent Conference, or I guess is the Reinvent Conference as it's still happening, which is notable for a few different reasons.

First of all, it was at this event last year that Amazon very nearly announced their own chatchip-T-style chatbot, only to scrap it at the last minute, and lucky they did so, because, of course, as it was happening, ChatGPT launched and changed the world. Now, that instigated a whole new set of changes to Amazon's AI strategy, including stealing away the bedrock name that they had given to the chat bot initially and turning bedrock into instead a sandbox environment in which they could help their enterprise customers customize available models out there, taking a position that there won't be one winner take all. And yet, heading into this event, there was some amount of anticipation that

perhaps we would get an actual bona fide chat GPT competitor. Around three weeks ago, the information reported that Amazon was developing something called Olympus. Now, even if this got people excited, there were, to some extent, tempered expectations. As the information wrote, Amazon's Olympus model faces steep odds. That piece writes, I know there are tons of AI models to think about, but Olympus is worth paying attention to. For starters, it's a clear sign that Amazon wants to control its own destiny in AI. While Amazon also sells LLMs from other model providers such as Anthropic, investing money and developing another giant AI model shows that the company ultimately doesn't want to rely on others for key technologies it offers to cloud customers. That's probably smart because who knows what

Anthropics' long-term thinking is, and the startup is also staying close to Google. And at the same time, this article pointed out how difficult it was going to be to differentiate. Part of that was that Amazon's existing proprietary model Titan had just had a very so-so reception. And still, it felt like something was necessary. The piece concludes, I wouldn't be surprised if AWS executives talk about Olympus at ReInvent the company's annual conference in Las Vegas, because if they don't, it might make customers in the press think it's even further behind OpenAI. So what did we get?

Well, we did get a chatbot, but it wasn't this rumored Olympus. Indeed, it was something very different. What we got was called Amazon Q, and here's how Amazon CEO Andy Jassy described it on X. Jassy writes, really excited to share with customers Amazon Q, a new type of generative AI powered assistant that is specifically for work and can be tailored to your business. Amazon Q can help you get fast, relevant answers to pressing questions, solve problems, generate content, and take actions using the data and expertise found in your company's information repositories, code, and enterprise systems. When you chat with Amazon Q, it provides immediate relevant information

and advice to help streamline tasks, speed up decision-making, and help spark creativity and innovation at work. Amazon Q is both your expert for building on AWS and for analyzing your business. We've built it to be secure and private, and it can understand and respect your existing identities, roles, and permissions, and use this information to personalize its interactions. If a user doesn't have permission to access certain data without Amazon Q, they can't access it using Amazon Q either.

And from day one, we've designed Amazon Q to meet stringent enterprise customers' requirements. None of their content is used to improve the underlying models. Now, in the preview video, they show a theoretical person asking questions like, what product features are causing the most problems for customers, to which Q is able to dig into the repository of information that comes from this company specifically and identify the advanced reporting feature as a problem. From there, the user is able to upload a training schedule and ask, based on this document, how fast can we get to education around this topic? Now, of course, the point of all of this is that Q is offering something very different, a deeply integrated chatbot that is powered by

the documents that your company runs on, that you've already presumably given Amazon or AWS access to. In that way, it's much more of a competitor for the Microsoft co-pilot type tools than it is to something like ChatGBTGBT, although ChatGPT's enterprise business is, of course, a huge driver of revenue. as well. Now, this clearly is part of Amazon's strategy to say to the world, we're not behind, we're just taking a different approach. Bedrock was the first part of that, and the vehicle by which Amazon made it very clear that they believed that the future of AI integrations into companies was not going to be winner-take-all, but was going to be lots and lots of different customized solutions, including both closed-source and open-source solutions that were tailored to companies

based on what they specifically needed. This further walks down that enterprise AI pathway, and of course builds on the trust that Amazon already has with lots of its users to try to, as Amazon Web Services CEO Adam Silipsky said, become a work companion for millions and millions of people in their work life. In their press around Q, they reinforced how many companies had had to resort to banning chatbots like ChatGPT because of concerns around security and privacy. One interesting technical note from the New York Times, unlike ChatGPT and Bard, Amazon Q is not built on a specific AI model. Instead, it uses an Amazon platform known as Bedrock, which connects several AI systems together. including Amazon's own Titan, as well as ones developed by Anthropic and Meta. Now, another

differentiation point is price. Whereas Microsoft and Google are both charging $30 a month for each user of their enterprise chatbot offerings, Amazon Q is starting at $20 per user per month. Indeed, the message they were trying to send was clearly picked up by Reuters, whose piece about the announcement was Amazon's AWS appeals to corporate customers with new chatbot. Amazon is trying to lure big corporate customers to its AWS cloud computing service with a new chatbot for business, and by offering to guard them against legal and reputational damage that can come from the output of artificial intelligence. So what were people's responses to this? First of all, there was quite a bit of laughter at the fact that Amazon named this service Q just after news had

broken that inside OpenAI, something called Q-star, had represented a technical advance that was sufficiently scary that it might have been part of the cause of Sam Altman's firing, although that has sort of been denied. Another response is basically a carryover of existing Amazon antipathy. L.A. Times tech columnist Brian Merchant writes, knowing how Amazon has pursued worker surveillance and productivity regimes over the last 10 plus years, this release about its plans to use and sell a genitive AI product Q should send at least a faint shutter down your spine. Now, the reality is that whatever they had announced, that probably would have been the type of response you were going to get from some circles. There hasn't been very much in terms of practical feedback from actually using

it, but we did get a little bit from Brian Romley, who writes, testing Amazon QAI with a large client right now that has been sideswiped by OpenAI Drama. I have been impressed thus far. So again, doesn't say much, but at least what it does say is positive. Still, I would argue that probably the biggest line of conversation is reflected in this tweet by Professor Ethan Malik who writes, one year since GPT3.5 was released and the released LLMs of all of the tech giants are barely catching up. Google Bard slash Palm 2 is worse. Amazon Q seems similar. X's Grock is similar. Meta Lama 2 is similar. Apple has nothing. No one is close to GPT4 yet. Maybe Google. Surprising. Two smaller labs, Anthropic and Infliction both have models that beat GBT 3.5, but not GPT4.

The question is whether GPT4 has some secret that prevents others from catching up. So, digesting this a little bit, the critique here is a general one that applies to all of these big tech players, which is why they can't catch up with the state of the art from a startup lab like OpenAI. That also gets back to the comment from the information piece we started with, that the longer than Amazon waits, the higher the expectations are going to go, and the more the perception of them being behind might be solidified. The counter argument to that, however, is that this is a very specific trajectory that just makes sense for this particular company.

There are negative and positive ways to look at that, and both were reflected in the conversation on X. Tiernan and Ray, an AI reporter for ZDNet writes, Amazon's Q-bot is the commodification of AI. Amazon has announced a $20 a month chatbot Amazon Q, which takes generally, and AI to where it was always going to go, which is a garden variety service that will be widely available and mostly sold on price rather than capabilities. Now, obviously, that framing is pretty negative, but it also contains within it a sort of inevitability, which is that these technologies

are so powerful and so useful in a workplace context that of course they were going to become commodified. Of course there was going to be at some point some amount of feature and capability parity across the big players. And from there, the decision point on which system to invest in was yes going to include things like price, as well of course as promises of security, privacy, and more intangible questions of trust that come from a legacy of who companies have worked with. Ex-user Amit writes, Amazon has massive distribution through AWS, so the question becomes if they can bundle this to sell the clients and take away from customers potentially looking to full-fledged solutions,

ultimately will be determined on the success of Amazon Q and how aggressive they are in trying to pair this with selling more compute. In other words, this is just part of what the cloud business looks like. Grimmie underside at Monk Chips writes, Q is going to be a big deal. It's Amazon's version of what Microsoft calls co-pilots. In the console, in the docs, in IDE, and Slack. This is the future of docs.

Also, we've all been complaining AWS has become too complicated and unwieldy. Q is an abstraction that could help. And this was something that I saw quite a bit as well, that AWS users specifically are very excited about a chat GBT-style bot whose main purpose is to navigate the AWS experience. Reinforcing that is a tweet from Deepak's, Singh who writes, Amazon Q is here. If you're building on AWS, our goal is to change how you build and interact with AWS. From helping you with getting information in the console to developing new features

in the IDE to Java migrations, Q is here for you. And all of a sudden, we have a different picture emerging, one that's not so much about how does Q compare to chat GPT, but instead, how does Q reflect a different phase of generative AI in which the priority is not only on the expansion of frontier capabilities, but on the integration of AI into existing workflows. If you listen to this podcast a lot, you will probably start to be getting sick of hearing that phrase, integration into existing workflows. But that's exactly what it seems like the next phase of generative AI is going to be about. And after our story this morning about the financial challenges of generative AI and stability AI, it kind of makes sense why this is the part of the story that we're getting into next.

Now, the other big announcement from Amazon's keynote was, as Adam Solipsky summed up, next generation AWS design chips, AWS Graviton 4, and AWS Trainium 2, deliver advances in price performance and energy efficiency for a broad range of consumer workloads, including ML training and generative AI applications, i.e. Amazon moving farther down the path of producing its own silicon, although interestingly at the same time that they announced these new chips, they also announced a deepening relationship with Nvidia, so it seems to be a both and strategy rather than an either or. So when all is said and done, I think in many ways this announcement was less sexy and exciting than people had hoped for, but much more reflective of where the industry is right now,

as it heads into 2024 and frankly increased expectations of actual value in the workplace, not just the first blush excitement of a brush with an advanced new technology. Thanks for listening or watching as always, and until next time, peace.