Google steps up AI arms race with Bard

Following the cultural success and early commercial promise of ChatGPT, Google has announced its competitor with access to up to date information and a far larger language model to draw on, just one of the major search companies now coming to market – here are a few ideas about what makes Bard’s release different.

Why it matters

OpenAI’s ChatGPT was exciting and novel, but it was based on Transformer technology (the T in the GPT) invented by Google, which also has a much bigger and more developed language model in the shape of LaMDA.

However, powerful applications at the scale of Google are difficult to do, so it is interesting to see the company treading carefully in the space while also leaning on its existing leadership in search, and adding a little detail to the AI promises of its most recent earnings announcement.

This said, the announcement so hot on the heels of ChatGPT’s smash hit demo does a lot to confirm reports that Google had been in a ‘code red’ situation – it’s worrying that such world-changing technologies could come down to a rushed arms race.

Introducing Bard

In a blog post, Google CEO Sundar Pichai introduced the new product as a conversational AI service intended to outflank ChatGPT in one key dimension: “It draws on information from the web to provide fresh, high-quality responses.” This is an area that ChatGPT’s public demo has been unable to do and will be far more difficult for the smaller company to emulate.

Conversational AI will also start to bleed into the search experience “soon”, Pichai writes, noting that lots of searches are moving beyond the factual and into the space of insights – he gives the example of asking which instruments are easier to learn, which requires an engagement with different perspectives.

But it’s worth thinking about some of the other issues that Pichai addresses:

• Bard will initially run on a lightweight version of LaMDA. This is interesting because it lets Google attack the space at scale, potentially without much of the throttling that has hampered users of ChatGPT at peak times.
• Ultimately, these systems are very costly to provide at the level of computing with some estimates suggesting natural language searching is seven times more expensive than a traditional search.
• Like ChatGPT’s public release, this iteration of Bard is a test and user feedback is critical. Arguably, Google’s reputation raises the bar that Bard’s accuracy and truthfulness must meet far higher than that of the startup.
• There remains a significant question, however, about how these services will affect how Google monetises its services should it pivot further into AI.

Baidu and Ernie

IT publication The Register adds more colour to the rumours of Baidu’s new chatbot, which will be known as Wenxin Yiyan in China and ERNIE outside China. Based on a language model larger than GPT3, and born bilingual, the chatbot could turn out to be extremely important in the development of the field (and to the operations and fortunes of global brands operating in China) once it completes testing in March.

Sourced from Google, WARC, The Register, New York Times