Amazon Gets In On AI Foundation Model Game with Nova
16:14 Share

Today, in the AI daily brief, amazon gets in the foundation model game. Before that, in the headlines, google veo rose out to more customers. The AI daily brief is a daily podcast in video about the most important news and discussions in A I to join the conversation for all the discard link in our shower notes.

Walk back to the a daily brief headline addition, all the daily AI news you need in around five minutes. One of the big themes we thought that the twenty twenty four was going to be all about was video generation, and to some extent has been the air kicked off with a preview of sora, which really reset people's expectations of what could be.

And while sora never came out, at least not in a way that was broadly available, we got great updates from runway with their genre, luma labs, dream machine, pek a and now google is rolling out their latest video and image generation models to a vertex customers in a private preview called vio and image three, respectively. The models round out google's genii offering into a swe focusing on vio, the video model. Google says that cora has already integrated its features into their poo chapt platform, while orio owner model, ys international, is using IT to create marketing content and collaboration with agency partners.

Unveiled back in April, the model is capable of outputs six second videos at ten N, D, P. Users can add a generated video OS, including changing camera movements, speaking to a long wait for A P. I access, warn barklay, the senior director of product management and google cloud, said IT was all about ensuring quote enterprise readiness, he added.

Since veo was announced, our teams have augmented harden to improve the model for enterprise customers in for a tex ai. As a today, you can create high definition videos in seven twenty p and sixty nine landscape or nine sixteen portrait aspect ratios, similar to how we have improved capabilities of other models such as gemini on vertex A I. We will continue to do this for vio tech on rights, vio understands the effects reasonably well from prompts that has some one of a grasp on physics, including include dynamics.

The model also supports mass editing for changes to specific reasons of a video and is capable of stringing together footage into longer projects. However, they also say reflecting the limitations of today's AI objects in viols videos disappear and reappear without much explanation or consistency. Env o often gets physics wrong.

I vig in many ways, the state video generation is all about what you can use IT for. You have tones of creators who are pushing the boundaries of short film making. But from a business perspective, this is a technology who's probably ready for prime time when IT comes to advertising, social media, but not necessarily yet longer film content.

Still, with this update, IT seems pretty clear that twenty twenty five is gna have a lot more video generation than twenty twenty four did. Next up, we've been following the recent ftc proof microsoft and IT appears that their deal with open a eye is explicitly part of the concern the information rights ftc officials have been asking microsoft rivals about the impact of its ideals and range of products. Specifically, the ftc is asking about microsoft deal with OpenAI.

The ftc is questioned rivals about how microsoft cells is OpenAI infused copilot, where and how IT resells open the eyes models to developers on its azure club computing platform. They continue the questions implied that the ftc is pRobing whether microsoft dominance in the cloud computing market has given the company an unfair advantage. Sales of A I software.

Of course, this is a very fluid situation. We have a new administration coming in, and many feel that is basically inevitable that although the next ftc chair nominee has not been announced, IT is very likely that there will be less and tagish towards big tech than linea con husband. Still, according to the information several of microsoft largest rivals quote, believe they can convince the trumpet administration to keep scrutinising the company.

Adding more created to that is, of course, the close role of adviser elon must get that administration, who has been a fierce critic of microsoft deal with OpenAI, going so far to actually sue microsoft as part of his lawsuit against OpenAI as well. Speaking of OpenAI, the company has just hired its first cmo, and that person is he had another refugee. From the craft to space, kate roh is the outgoing cm of coin base and represents the latest major higher as OpenAI built out at sea sweet sar A E S in A C F O Kevin wal as chief business officer, although as of yet they have not hired A C T O to replace me.

Mali k tweet could not be more excited to help show the world what A I that benefits all of humanity looks like. Congrats to kate and good scoop. Open an eye for now that we've got a slightly longer than north main episode is so we are going to cut the headlines here.

Appreciate you listening as always. Now it's time for the main episode. Today's epo de is brought you by rocket money.

We are coming up on the beginning of the new year, and that is a perfect time to get organized. That goals prioritize what matters, which for many of us is going to be financial wellness, thanks to rocket money. Those goals, especially around money, feel achievable.

Rocket money shows you all of your subscriptions right in one place, helping you easily cancel those that you may be for. Got that you're actually paying for rocky money also post together all of your spending across your different account so that you can clearly track spending habits and see where you can cut back. Rocket money is a personal finance tap that helps find and cancel unwanted subscriptions, monitors your spending and helps lower your bills.

You can grow your savings. Their dashmore gives you a clear view of your expenses across all of your accounts. You can easily create a personalized budget with custom categories.

You can see your monthly spending trends in each category to know exactly where your money is going. Rock money will even try to negotiate lower bills for you. They automatically scan in your bills to find opportunities to save, and then you can ask them to negotiate.

They'll deal with customer service so that you don't have to rock money. Has over five million users in a safety total of five hundred million in council subscriptions, saving members up to seven hundred and forty dollars a year when using all of the apps, premium features cancell your unwanted subscriptions and reach your financial goals faster. With rocket money, go a rocket money to calm.

Flash A I break down today. That's rocket money. Duck com slash I break down. Today's episode des brought you by vantage, whether you're starting in your company. The security program demonstrating top notch security practices and establishing trust is more important than ever.

Venta automates compliance for I S O twenty seven, O O one, soc two gdpr and leading AI framework like I S O forty two thousand one and N I S T A I risk management framework, saving you time and money while helping you build customer trust. Plus you can streaming mind security reviews by automating questionaire and demonstrating your security posture with a customer facing trust center. All power by vent to A I over eight thousand global companies like LangChain lea AI in factory A I use vantage to demonstrate A I trust, improve security in real time, learn more adventure.

Doc com slash N L W that's ventadour com slash N L W today's eps ode is brought to you, as always, by super intelligent. Have you ever wanted an A I daily brief but totally focused on how A I relates to your company? Is your company struggling with A I adoption either because you're getting figuring out what use cases will drive value or because the A I transformation that is happening is individual teams, departments and employees and not able to change the company as a whole? Super intelligence has developed a new custom internal power test product that inspires our teams by sharing the best AI use cases from inside and outside your company.

Think of IT is an A I daily brief, but just for your company's A I use cases, if you'd like to learn more, go to be super di 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 be super da I slash partner.

Welcome back to the daily brief. Amazon has had a long an interesting journey when IT comes to their relationship with foundation models in the A I space. We've had context to discuss them a couple times in the last week and noted that much of their strategy was formed in reaction to ChatGPT coming out and being, Frankly, much Better than the version of that, that they had planned on releasing themselves.

In fact, they took the name that they had been planning to use on their ChatGPT equivalent, which was bedrock, and turn that into an A W S service to help enterprise customers figure out which models to use. Since then, they've double down on the relationship with anthropic as well as really focused on their infrastructure play with their training um chips. But IT appears that they are not content to not be in the foundation model game as at A W S reinvent this year.

The biggest announcement of that event was the unveiling of a new family of proprietary models called nova. The range includes four sizes of aleem microlight pro and premier nova micro as a text only model describe as being optimize for speed and cost, nova light is a low cost multi model model capable of quickly analyzing image, video and text inputs. Nova pro is described as a highly capable multi model model with the best combination of accuracy, speeding costs for a wide range of tasks, and nova premier is amazon's most capable multi model model designed to excell IT quote complex reasoning tasks, and for use as the best teacher for distilled stop models.

The line also includes an image generation model over canvas and a video generation model of a real. Each is claim to be state of the art in their respective fields. All of the models of the novel mier are now available within the bedrock model library on A W S.

With premier expected to arrive sometime early next year, amazon C E O andy ji said, we've continued to work on our own frontier models, and those frontier models have made a tremendous amount progress over the last four to five months. And we figured if we're finding value out of them, you would probably find value out of them. We don't know how many parameters these models are, but based on their descriptions, they seem to line up with similar latest generation model families from leading labs.

Context windows are also comparable to rival models, but amazon promised to deliver an ultra long two million tok in context window for some models next year. The aleem all support find tuning using text, images and videos s as well as model installation canvas. The image model looks on par with leading models from rival labs.

While real, the video model feels a bit like a teaser version, only supporting six second videos, which take about three minutes. Generate, amazon says, that diversion that can generate two minute videos is coming soon. Quality appears up to standard in the demo video provided by on.

However, the user generated videos s that have come out so far ranged dramatically from good enough to extremely junky. Coming next in the novel lining up is a speech to speech model expected in q one of next year. And in any to any model expected in the twenty twenty five, the benchMarks appeared first plants to be competitive.

Amazon is claiming that real outperformed runways genre alpha and A B testing, achieving a sixty one percent when rate for video quality and a seventy one percent when rate for video consistency for the language models. No APP seems to be at least competitive with CD three point five thousand and GPT four o claiming our performance in some areas. One of the things that I think is important when we talk about benchMarks is that it's probably more valuable to understand these in terms of competitive ranges rather than in terms of exact specifics.

And if that's how you take IT nov pro is in the classic models that include claw three point five senate GPT four o IT. Set up while I ocken veteran, a tech leader of U. S, pointed out an interesting area of out our performance, commenting one exciting aspect, the newly released him, is on nova genee models as their impressive performance on agent and multi agent benchMarks.

Offsetting this point, though, one of the areas that novel seems to be lacking in is in coding benchMarks. AI entrepreneur ban do ready ran her own testing on live bench, finding no approves further down the leader board, SHE writes the benefit of having a benchmark the changes every month is that IT can't be gamed. On our latest november chAllenge, amazon's nova cores below lam seventy b and is slightly Better than high co that net this model doesn't change.

The leader is significantly, though IT seems surprisingly fast, predictably saw at three point five, and the o online remain on top of the list. One of the more important features though, which is kind of gloss over, Frankly, IT reinvent, was how competitive nova is on Price. Nova micro and noble light, or both Price below gami one point five flash and GPT four mini, making them the cheapest models available from a major lab.

Novel light also has the distinction as the cheapest available model with multimodal inputs. Those two models are Price so close together that the differentiation seems to be speed with novel icr ro about ten percent faster than gami one point five ash nova light runs at a similar pace. E G P T four o, which is still faster than the rest of the pack.

Nova pro is available at around a third of the cost of either GPT o or called three five eight, and also position is slightly cheaper than cloud three point five high co, with much higher performance scores. The net of all of this is that amazon's in house EMS are now that cheap, is available for many use cases. Jerry Lewis CEO of lamon index rights amazon's nova should have advertised the cost production front and center instead of me having to dig up a hacker or new thread is a huge value proposition.

Nova is an exciting push towards way cheaper models that are comparable to the state of the art in terms of context, window performance and multi modality alums. S have come a long way, but a huge issue with using them for repeated loops is cost, especially for multimodal agented flows. Overall, the biggest take away from this announcement, if you're just trying to understand IT in a sentence two, is that amazon has gone from skipping this generation of ms to Frankly, fully and play with the complete lineup of competitive models, warn professor eth molex summed up.

And then there were six or so based on the stats that looks like amazon's novo is a competitive frontier model. This rounds out the GPT four one models, GPT four o gami one point five cloth, three point five, greg two on a three point two and maybe three on U. S. Models one year.

And mo, another question for me is that given that Jesse is talking about making progress over just the last four to five months, sort of implies that we've caught up to the other labs in summer, which that the case also lends credence to the growing sense that maybe there really aren't particularly strong motes when IT comes to models. The next week, group of announcements related to amazon's training on a eye chips training of two instances are now generally available on AWS for training and inference. The company also announced a new generation of chips training um three expected to become available late next year for the training mum two chips.

Amazon is cleaning of four x improvement in speed over the first generation, which Frankly saw very little adoption. They boast that training of two inference can deliver three x higher token generation through put on meta lama four five b model compared to offering available from other cloud providers in their announcement post amazon claims quote, training um two offers thirty to forty percent Better Price performance in the current generation of GPU based E C two instances, presumably referring in video h one hundreds A W C O at garmin claim that training three will be twice as fast as the second generation and deliver forty percent more energy efficiency, garden said today, there's really only one choice on the G P. U.

Side and it's just in video. We think that customers would appreciate having multiple choices. One big surprise regarding training um was a special appearance from apple, senior director of machine learning in A I benwell depend depend at the stage to promote training um stating that apple is currently using the chips to power services like search and will evaluate the use of the latest generation to in their apple intelligence models.

It's pretty unheard of for apple to endorse supplier, particularly in a segment as competitive as cloud. And so people are taking this as a pretty strong vote of confidence. Amazon also announced that the ign tic new AI training supercomputer dog rania design in collaboration with anthropic.

Amazon says the super clue will deliver five x the computer that was used for anthropic latest training runs. The design is somewhat unique, with the cluster's read out across multiple facilities network together. Usually training super clusters are housed under one roof, is the latency to transport data across the network is a major bottle neck.

Amazon claims to have solve this problem with networking technology that they call the elastic fabric adapter, amazon said in their press release. When completed, IT is expected to be the world's largest AI compute cluster reported to date. The current high water mark for completed supercomputers is elon mux colosse facility, which houses one hundred thousand and video age one hundreds.

Dd hutt, who works with customers at amazon's chip making unit and a panel labs, said the rainy will contain quotes significantly more than one hundred thousand training to chips. The cluster is expected to be ready sometime next year. So that is the story.

Lots going on on the training um side. But the big news is nova and amazon being in the foundation model game for reeled for now that's going to do for today's, I believe. Appreciate you listening or watching as always and until next time, peace.