Industry experts agreed the announcement of AI large language model (LLM) Gemini by Google last week marked a turning point in the search giant’s approach to the technology, though noted there remain questions around the company’s broader strategy.
Michael Orme, senior analyst at GlobalData, and Richard Windsor, founder of research blog Radio Free Mobile, separately noted Gemini is a much-needed rebound for Google after problems surrounding the launch of its Bard chatbot.
The GlobalData expert told Mobile World Live “Google has clearly got its act together” following a Bard launch he branded a “debacle”, which prompted parent company Alphabet to merge the research teams behind Google Brain and DeepMind.
“These first results from this shotgun wedding suggest that with the just-launched Gemini LLM, Google has got its act together.”
Windsor also believes Gemini addresses what he branded a panicky response by Google to the rise of OpenAI’s ChatGPT.
Related Articles
In a blog, Windsor noted Gemini is “Google’s first foray into the AI ecosystem” and predicted the company would quickly look to launch a kit to build relevant “services on top of it, as well as a store to distribute and sell these”.
He suggested the split of Gemini into Ultra, Pro and Nano editions means Google already has an eye on different customer bases and use cases.
Orme said Gemini “is designed to bring world-competitive AI benefits to Google’s 4 billion-strong user base” spanning search, email, browser and Android devices “by 2025”, a point also separately made by Windsor.
Vague
Both highlighted some questions over timing: Orme noted it will take until sometime in 2024 “to get a handle on just how Gemini stacks up against OpenAI’s GPT-5”, based on the assumption this product launches within the next 12 months or so.
“The promise is that Ultra will offer a breakthrough in multimodality and agent AI. But then, we’ve heard all kinds of whispers from OpenAI along the same lines.”
Windsor stated Google had been “deliberately vague” on some of the key characteristics of Gemini, citing the size of each edition of the LLM as an example.
He predicted the Ultra version would “almost certainly be cloud-based” and capable of handling “the most complex tasks” spanning billions of parameters. The Pro edition suggests Google has realised “users will have to train and run multiple copies” of the LLM to handle numerous tasks simultaneously, while the Nano option “has been optimised for efficiency to run on-device”.
Windsor noted Nano would most likely begin life on Google’s Pixel smartphones, but backed the company to already be in talks “with Qualcomm and MediaTek to ensure their processors are optimised” to run it.
He said the line-up would likely spearhead Google’s ambition to lead the AI market, citing a perceived improvement in the accuracy of the company’s Bard chatbot, which is already using Gemini Pro.
Orme diverged on this point, saying the AI in Bard is currently “not proving a sensation by any means” when measured against GPT-4.
Windsor cautioned there is an element of smoke and mirrors to Google’s announcement, explaining that while all the features cited for the model appear to be “a big step forward”, Gemini “has no causal understanding of what it is doing”, a common trait in all current LLMs.
Progress
While Windsor and Orme each offered a dose of reality around Google’s move, they were broadly confident the search giant would ultimately get Gemini right.
The GlobalData analyst noted Google already occupies a leading position in the field, citing AlphaFold, an AI used to predict protein structure.
“If and when Gemini can amalgamate the best of Alpha self-teaching, deep reinforcement learning with transformer LLMs, then step change” will be possible in the sector, along with potential “glimmers” of artificial general intelligence (AGI).
Windsor believes Google “has finally got its act together” and “laid a viable claim to the AI ecosystem”, though he noted the company’s reticence to bare all about Gemini at this stage is “yet another sign of how the once open and collaborative world of AI has become a hotbed of fierce competition”.
Comments