Amazon – 6 years late

Alexa might finally deliver on a 6-year old promise.

• Amazon’s annual Alexa event was dominated by generative AI, but the reality is that Amazon promised all this 6 years ago and remains demonstrably behind the curve when it comes to generative AI.
• It is ironic that Amazon has by far the largest installed base of hardware that could be upgraded to generative AI with great effect but remains very far from a leader in this field.
• A series of new hardware products including a new Echo Show, speakers and tablets for kids, Fire TV sticks and a smart home controller were launched but the real focus was on the software that would run them.
• Amazon has put a lot of thought into the creation of the large language models (LLMs) that will run all of these devices and I am pretty certain that the quality of the Alexa voice experience is about to take a big step forward.
• Alexa today remains pretty similar to the one that debuted in 2014 and is not much use other than as a convenient way to play songs, turn lights on and off or set timers when one’s hands are busy doing something else.
• This is because the AI powers Alexa has been substandard for years but just like JVC and Betamax, the inferior product has fared much better thanks to the Amazon retail juggernaut.
• This has not hindered its usage as, according to Amazon, the 1bn deployed devices are accessed by users on average 3 times per day.
• This puts Amazon in an excellent position if it can improve the user experience and turn the device into the personal assistant that it promised over 6 years ago but failed to deliver on.
• RFM research has concluded that one of the biggest steps forward offered by LLMs is the ability for machines to converse fluently with humans making voice a realistic user interface for the first time.
• Amazon is trying to capitalise on this, but the demonstrations clearly indicated that a lot more work is needed as, while Alexa is much better at conversation, its replies remain stilted, delayed and often go wrong.
• There is a lot more to this than just sticking a text-to-voice engine in front of ChatGPT, and despite its massive resources, Amazon is still having difficulty with it.
• Amazon continues to offer Alexa as a series of skills and the LLM function will be deployed as an app that the user enters called “lets chat” and then exits at the end of the conversation.
• This has been fine-tuned for the activities that Amazon expects the user to undertake which is a theme that ran through all of the product launches.
• For example, the kids’ products have been optimised to have conversations about animals in a way that will interest kids and all possibility of craziness has been scrubbed from the LLM.
• This is not the way other LLMs operate where they are trained on a massive trove of scraped internet data and then let loose to see what they can do and what they can’t.
• In practice, the scrubbing means that the LLM has been lobotomised removing all possibility of it doing anything that will upset users but at the same time preventing it from doing anything really groundbreaking.
• In effect, Amazon has created a series of LLMs that have been fine-tuned for specific activities which is not that different to the training of the Alexa bot to do specific functions that has been going on since its launch in 2014.
• What is different is the ability to understand what the human is saying as well as respond using context and circumstance.
• However, I still think that Amazon does not really understand the realities of LLMs which was underlined by its announcement that it was going to combine its speech-to-text, text-to-speech and other algorithms into one large LLM.
• This is the opposite direction that the industry seems to be going in as the focus now is to make models smaller with more specific functions that have better performance by training them with more data.
• This makes them cheaper and easier to deploy and research indicates that they can be just as good if not better than their larger counterparts.
• Amazon seems to have started in the right place but now seems to be going backwards.
• I still think that Amazon has a shot at becoming a major player in this space solely because it has by far the largest installed base of products that can deliver LLM services directly to consumers.
• This is like the VHS / Betamax war where Betamax was a better product but superior marketing by JVC allowed the VHS standard to win the day.
• Hence when the new batch of LLM companies starts running out of money, Amazon will be able to purchase the assets it needs to make the hardware it has installed in its customers’ homes really work properly.
• Until then, I think users are still going to use Alexa to play music, set timers and turn the lights on and off but the new voice algorithms should allow Alexa to execute these better than it has in the past.