LAMs need a lot of work to find place in future marketing strategies

The humanity had just started to get used to the post holiday work routine, chilly winters, and large language models, when AI hardware company, Rabbit announced its first large action model or LAM based product R1 at CES 2024.

That single product has caught attention of the tech and tech influencer community almost instantly. It managed to find dedicated space in MKBHD's WVFRM podcast episode, and appreciation by Microsoft CEO, Satya Nadella!

I was covering CES 2024 from the comfort of my room, scrolling through the Twitter feed on my iPhone. The way one would do in this chilling Delhi winter. I discovered R1, its noticeable orange square shape, and almost scrolled up till I read, large action model. With LAM in brackets right below.

And my brain went, "LAM? What is that now?". I quickly turned to my laptop and Googled about it. And since then, I have been consuming materials about LAMs, media coverage of R1, and polarising reviews of R1.

Don't worry, I'm not going to review Rabbit R1. That's not what this episode and channel is about. My curiousity about LAMs pulled me into the rabbit hole of a technology that has been in the works for a while now, waiting to be consumerised the way ChatGPT did for generative AI, and AI in general.

So, without a further ado, let's talk a little bit about R1, see what we know about LAMs so far, and what the technology needs to do to get the attention of the marketing community.

Let's talk a little bit about Rabbit R1

R1 is squarish orange device with a 2.88-inch touchscreen with a 2.3 GHz MediaTek processor, 4 GB RAM, and 128 GB storage. It looks like a bigger version of Nintendo Gameboy with a rotating camera, and a scroll wheel button.

But, it's not the looks that R1 has got so much attention for. It's what it is able to do with its LAM-based operating system called RabbitOS. It is able to go beyond the tranformer capabilities of the LLMs like ChatGPT or Gemini and basically ACT on behalf of the user. In layman terms, R1 is able to take multi-modal input from the user via voice, text, video or photo and come up with actionable suggestions or simply take the action.

For example, in the keynote at CES, Jessy Lyu, the CEO and Founder of Rabbit, asked it to build out an itenary for a family trip to London. R1 developed a day-wise plan, identified flights, and showed options of hotels available for confirmation.

This feat is huge from the point of view of using AI for taking complex inputs, processing them, and presenting relevant actionable options to the user. And what is most interesting here is that it does not require any integration with the apps or devices that you might be using.

RabbitOS requires you to login to their web portal called Rabbit Hole to login to your apps or services once, and then it acts as a universal controller for those apps and services. There is limited understanding of the UI, which is only through the media reports and the keynote. It appears to have a category-based cards for music, transportation, or video chats. The screen is also nothing like what we are used to on our mobile and computing devices.

Still, the capabilities exhibited by the LAM-based device are truly magical. R1's LAM has been trained to navigate through the UI of major apps and operating systems that are in use today. For those apps R1 is not trained, it offers a training mode wherein a user can train it to learn interface of the apps of their choice, allow it to process it, and then it will be able to action out the user inputs for those apps! This is truly something out of a sci-fi movie!

Understanding R1's LAM and LAMs in general

Large language models or LLMs, are something that most of us have experienced at some point or the other in the form of ChatGPT or Bard or SGE. These LLMs are based on a technology known as the transformer. A transformer has enabled LLMs to rapidly generate content, text and image based content. But what if LLMs could generate actions? This is where large action models or LAMs step in.

LAMs are based on the technology known as reinforced learning which allows to build decision-making agents and learn optimal behaviour through trial and error. LAM used by R1 learns through imitation or demonstration. It observes how a human is interacting with a UI and tries to reliably replicate it. Since its approach is based on observation, R1's LAM continues to gather deep understanding of the UI over time and output actions even if there is a change in the interface.

Because R1's LAM is based on what is known neuro-symbolic systems, it can act on apps the same way humans do. What this also means is that LAM can learn any interface of any software regardless of the platform they are working on. Not just that, LAMs can be trained to complete a process that runs across multiple applications to complete a task. Something a professional does in a day-to-day work environment. And this does not require any kind of programming! In that respect, LAMs are able to go beyond language processing and perform actions based on the instructions given by the user.

Direction LAMs must go in to find a place in marketing strategies

Whether you are zealot, skeptic or a naysayer of Rabbit R1 as a product, the capabilities exhibited by LAM that powers RabbitOS are very impressive. But, as with any bleeding-edge technology, businesses need to have clarity on its real-world applications. The problems that technology can solve for them, and deliver efficiencies in terms of time, cost or both.

Now that AI has become fairly consumerised, LAMs need to start giving signals that gets the attention of businesses, especially marketers. The signal could be around one of the 3 aspects of marketing - systems and processes, targeting, and data analysis.

Looking at the present state of maturity of AI, I think automation of marketing systems and processes seems to be lowest hanging fruit. Since LAMs have the capability to learn through observing, training it to use a particular set of tools for content marketing should not be very complicated. So, LAMs could be trained to access LLMs like ChatGPT to generate content for an e-book. Then, it could open up Canva, pull up an e-book template, copy-paste content generated using ChatGPT, search and insert nice looking images and graphics, format it as per user's training, and export the PDF to a specific folder. It could also ask ChatGPT to generate e-mail sequence to promote the e-book.

For targeting, LAM could be trained to pull out a segmented list of contacts from a CRM platform, upload the list to an e-mail marketing tool, use the e-mail sequence generated by ChatGPT, and schedule it using the tool's automation functionalities.

Then, it could export data from the e-mail marketing tool in an Excel file, slice and dice the data, create charts, and embed them into a PowerPoint presentation. It could again take help of ChatGPT, if needed, to generate insights and place them next to relevant charts and graphs in the presentation.

I know, this sounds almost like a sci-fi move scene, and it is, but only to an extent. Because marketers have been using automation for a while now. For many businesses, ChatGPT and Canva have become a part of their content generation process. And quite many are using AI-based tools to perform data analysis and generate insights. It seems that LAMs can play a very important role in integrating systems, processes, and data. This can be very helpful for many marketing organizations that are sitting on siloed data. It can also help in adopting a CDP platform or generative search. I will drop in a link in the description to previous episodes on these topics.

Hazy applicability of LAMs for businesses and marketers

As LAMs get trained for more apps and interfaces, their applicability in real-world business scenarios will be only limited by human imagination. Having said that, there are a few things about LAMs that are hazy.

For instance, how will they change online search and the current process of information discovery for the users. Its multi-modal capabilties seem to be more advanced than what other generative AI options are offering right now. Moreover, what about all the UI/UX capabilities that have been perfected by marketers, developers and designers over time?

The argument that LAMs are trained to learn the interface and use it the way humans do seems a bit too easy. Let's not forget that RabbitOS is the first version of a consumerised LAM that we are witnessing. What future versions and variants of LAMs will look like is anybody's guess right now.

Then there is the critical aspect of data privacy and security. Even in the CES keynote, privacy part got very limited screen time. It is not fully clear how the user data will be stored and used. Rabbit, and other companies that bring LAM-based products into the market will need to be more transparent about this aspect.

The Rabbit R1 seems to be more of a personal assistant than anything else right now. For all the right reasons, it is more focused on the individual user and not businesses. Assuming it remains a consumer device for a foreseeable period of time, what opportunities its minimalistic, monochromatic UI will offer businesses and marketers to serve ads or to send messages or to become part of the user's online information discovery is unknown.

Finally, Rabbit CEO's claim that R1 can learn any complex interface like that of Adobe Photoshop, seems to be of tall order. If, IF, it can deliver on these claims, then it could mean a very next level of efficiency. But its capabilities are yet to be fully understood.

R1 has got a lot of popularity in a very short span of time. No wonder it has managed to sell 10,000 units in the first 24 hours! Internally, they were hoping to sell just 500 units. Assuming that the product lives up to the hype and delivers really good user experience, the sales volume are likely to skyrocket in that case. Is R1 a smartphone killer? It's hard to say. Its the LAM that runs the product that definitely holds a lot of potential. However, from a business perspective, the technology still needs a lot of work to become part of future marketing strategies.