Meta Platforms – Dial M pt. III

Here we go again.

• Even though the launch of a chatbot is a rinse and repeat of 2017’s Facebook M (see here), Meta Platforms’ complete dominance of the open-source community puts it in a powerful position with developers.
• Meta Platforms made three big announcements at its Connect 2023 conference, all of which involve AI (of course).
  • First, chatbot is not dissimilar to the attempt that it made in 2017 when everyone was jumping up and down about Alexa, Google Assistant and so on.
  • At that time Meta’s AI was so bad that M had to have humans available to answer questions and the chatbot did not last very long.
  • In the last 6 years, Meta’s AI has improved substantially, and so this time around I think the company should be able to produce a reasonable experience.
  • The new bot is called Meta AI and will be powered by LlaMa-2 (I presume the largest variant it has) and Emu which is an image generation model.
  • The AI will be linked to all of Meta’s properties and will function much like ChatGPT in that its knowledge will be frozen in time and unable to perform Internet searches.
  • To get around this problem, Meta AI will be able to call on Bing Search through the partnership with Microsoft, but this part of the service will be little more than Bing Search with a conversational front end just like Bard or Bing already provide.
  • Second, Personas and Studio. To make the chatbot more appealing and entertaining for consumers, Meta has implemented 28 variants that use the voices of well-known celebrities.
  • It has also added an AI studio that provides a series of simple tools by which users can create their own custom chatbots that can be used for customer service and so on.
  • The AI studio is a particularly good idea as the monetisation of WhatsApp is an ongoing challenge for Meta’s year of efficiency and this should help add value to business customers who may then be willing to pay more for using WhatsApp to communicate with customers (see here).
  • Third, Meta Ray Ban Glasses which is an enhancement of the product that is already in the market.
  • The new glasses can take pictures and videos with much better resolution as well as play music through external speakers that direct sound to the ears.
  • The idea here is that these will be upgradeable with Meta’s new assistant which in time will be able to provide near-real-time translation.
• Given Meta’s consumer focus and the recent effort it has made in AI, these products are a natural step.
• Part of the problem with using generative AI to power chatbots is that they still have to run in the cloud and, as Amazon demonstrated earlier this week (see here), this leads to a stilted and unrealistic conversation.
• Consequently, I think that we are still some way from LLMs providing realistic conversation as this will require an edge implementation but this is a crucial step forward.
• The last time everyone tried this, the AI was so bad that the bots could only reliably play music, set timers and turn the lights on and off.
• With LLMs, the digital ecosystems now have a reasonable chance to make good on all of the promises they made about their chatbots 6 years ago but spectacularly failed to deliver on.
• Furthermore, while Meta is still catching up with OpenAI and Google, it has a crucial advantage with developers.
• This is because LlaMa is the standard foundation model for open-source development.
• This means that many of the developers that Meta is seeking to attract will already be familiar with LlaMa giving it a head start over its peers.
• This will be crucial in the development of the generative AI ecosystem and here Meta is leading the pack.
• Meta Platforms has reversed all of the discount that was priced in when Mr Zuckerberg was resisting pressure to cut costs like everyone else in 2022.
• Hence, I am not inclined to chase it as I missed it and there are much better opportunities to look at.