Chatbots – Galloping obsolescence

Alexa, Google Assistant et al have no value.

• After many years of wrangling, there is finally a smart speaker in the market that supports both Amazon Alexa and Google Assistant in a clear sign that their spectacular failure to do anything very useful, combined with generative AI, has reduced their strategic value to zero.
• Harman has released through its JBL brand the JBL Authentics 300 which will support both Amazon Alexa and Google Assistant where users can choose one or both during setup and access them simultaneously.
• This is not a new idea and products have been planned that can do this for many years but sensitivities around user loyalty and ecosystems made the owners of the assistants wary to allow them to coinhabit the same device.
• The reasoning behind this is pretty simple and back in 2017 when these assistants were all the rage, it was thought that usage of one of these assistants would drive loyalty to the digital ecosystem in question.
• Furthermore, as Apple has exploited very successfully if a user owns one Apple device, he or she is more likely to purchase another Apple device given how well they work together.
• Both Google and Samsung have strived mightily to replicate this effect but so far have not been able to deliver very much.
• I have long argued that this is because they do not work closely enough together and insist on doing things themselves that would be much better sourced from the other party.
• The result is that Apple is the only digital ecosystem owner that has been able to really leverage its ecosystem across multiple devices with great effect.
• The problem with the digital assistant was that despite promises of a virtual personal assistant that takes care of everything, the result was a series of very stupid assistants that could only really play music, set timers and turn the lights on and off.
• Consequently, their ability to influence the user’s choice of where he or she lives his or her digital life has greatly underwhelmed expectations.
• When this is taken into context of what the latest large language models (LLMs) are capable of, there is finally a realistic prospect of having a capable digital assistant as long as the inevitable errors and hallucinations can be dealt with.
• RFM research has assessed their language capabilities on a test devised in 2017 to measure digital assistants and found that despite the errors, their ability to converse with humans and be an audio-based man-machine interface is substantially improved.
• This has driven the realisation that the older generation of chatbots are obsolete and of almost no value, explaining why it is only now that speakers with multiple assistants on them have properly made it to market.
• I think that there are substantial opportunities for LLMs to perform as proper digital assistants, but there remain substantial technical hurdles such as size, cost, privacy, mistakes and hallucinations that need to be overcome or at least minimised.
• Users of GPT-4 (paid) claim that it is far better than GPT-3.5 (free) but the benchmarks that have been run suggest that the improvement is not as great as it would seem.
• This is going to be key going forward as the barrier to entry of creating an LLM has been well and truly removed meaning that there will be thousands of LLMs available.
• Quality and performance will become the key differentiators that separate the men from the boys in this latest round of AI competition.
• I think that OpenAI has a slight lead over the others having been one of the first to move in this direction, but Google is certainly not far behind and could easily catch and surpass OpenAI and Microsoft.
• I suspect that Google Assistant, Alexa, Siri et al will be quietly retired over time as LLMs become more practical to deploy and use.