Despite being a major player in the tech world and a pioneer in AI research, Google has left me scratching my head with their latest AI chat assistant, Google Bard. With ChatGPT from OpenAI setting a high bar in the AI space, other tech behemoths have been trying to replicate its success, especially in the wake of the post-pandemic tech slump and worsening global economic conditions. However, Google’s recent offering seems a step in the wrong direction.
Google Bard, intended as a competitor to ChatGPT, falls noticeably short in performance. When put to the test, Bard often delivers incorrect or low-quality outputs, failing to compete with the older GPT-3.5 Turbo, let alone the advanced GPT-4.
What’s intriguing is Google’s choice to use LaMDA as the underlying technology for Bard, bypassing their more advanced language models like PaLM or Minerva. Even more perplexing is their decision to ignore DeepMind’s Sparrow (DeepMind is a subsidiary of Alphabet Inc), a model that uses the internet to enhance accuracy, much like Microsoft’s integration of Bing with GPT-4. This decision seems to echo a sense of isolated development rather than collaborative innovation.
However, it’s not that Google is lacking in research prowess. Their groundbreaking work, the Transformer design from their 2017 paper “Attention is All You Need,” played a fundamental role in developing GPT models. More recently, Google published papers emphasizing the remarkable finding that large language models (LLMs) can reason. This revelation culminated in their paper on Minerva, which they claimed outperformed GPT-3 regarding reasoning capabilities.
But here’s where things get a bit foggy. After the Minerva paper, there’s been no further news about the model. Despite releasing two more potentially revolutionary papers early this year, there’s been no subsequent announcement of a new, more powerful LLM. One might expect that Google is on the cusp of announcing a model surpassing GPT-4, perhaps even reaching the much-anticipated AGI milestone. But as of now, we’ve heard nothing.
While it’s clear that Google is investing in AI research, their recent moves suggest a disconnection between their research and product development. Despite a wealth of promising research, their practical output—in the form of Google Bard—doesn’t reflect the innovative spirit they portray in their papers. Bard feels like a rushed and quickly slapped together answer to ChatGPT to keep the shareholders happy, not a competitive or impressive response to OpenAI (who is currently dominating the AI discourse).
The question is, will Google translate their pioneering research into competitive products? Or will they continue to lag behind, allowing other companies to lead the charge in AI innovation?
It’s too soon to write off Google in the AI race, but their recent actions—or lack thereof—signal a need for a more strategic approach. Their actions need to align with their words, and their impressive research should pave the way for equally impressive products. After all, in AI’s high-stakes world, actions speak louder than words.
The race is on towards being the first company to achieve AGI. Some say we’re only years away from achieving it. Still, given the amount of resources dedicated to AI research and development, it’s possible it might come sooner rather than later.