The days leading up to the holiday season are typically quiet for new technology releases, but Alphabet Inc’s Google is making a desperate push to catch up to its competitors. In response to being caught off guard by OpenAI’s ChatGPT last year, Google has been eager to showcase its advances. However, despite the hype surrounding its new AI model, Gemini, Google still lags behind OpenAI from a technical perspective.
Gemini, which Google claimed could detect magic tricks and excel in accounting exams, was unveiled in a surprise announcement after reports of a delay. While a demo video released by Google has impressed many on social media, it’s important to view the situation with a critical eye. From a technical standpoint, Google is still playing catch-up to OpenAI.
Google’s own table comparing Gemini to OpenAI’s top model, GPT-4, shows that Gemini Ultra (highlighted in blue) outperforms GPT-4 on most standard benchmarks. These benchmarks test the AI models on various tasks, such as high school physics, professional law, and moral scenarios. These capabilities largely define the ongoing competition in the AI field.
However, the improvements made by Gemini Ultra on these benchmarks are marginal. The model only surpasses OpenAI’s GPT-4 by a few percentage points. Essentially, Google’s best AI model has only made narrow advancements on something that OpenAI had accomplished at least a year ago. Additionally, Gemini Ultra, the most promising variant, is still under wraps and yet to be released.
Even the demo video, described by technologists as jaw-dropping, has its caveats. While the model’s ability to track a ball of paper underneath a plastic cup or predict a dot-to-dot picture represents significant reasoning abilities nurtured by Google’s DeepMind AI lab, many of the other showcased capabilities can be replicated by ChatGPT Plus, as demonstrated by Wharton professor Ethan Mollick. The reliance on edits and shortened outputs in the video, combined with the admission that the demo wasn’t performed in real time or voice, raises validity concerns.
Google seems to be emphasizing its vast research team and data access while capitalizing on the recent turbulence at OpenAI, which saw CEO Sam Altman temporarily ousted. As Google actively seeks to convince OpenAI’s corporate customers to switch, it is attempting to ride the wave of uncertainty by launching Gemini.
However, impressive demonstrations and marketing strategies can only go so far. Google has previously showcased remarkable but ultimately ineffective technologies, such as Duplex. Additionally, Google’s bureaucratic structure and layers of product managers have hindered its ability to swiftly ship products, in contrast to OpenAI’s agility.
As AI continues to reshape society, it is crucial to approach Google’s supposed sprint ahead with caution. While the company’s efforts are commendable, it is important to recognize that Google is still playing catch-up. Impressive marketing and extensive deployments are not always decisive factors in technology success. In the realm of software, market dominance is achieved through optimal performance.
In conclusion, Google’s latest AI model, Gemini, appears remarkable at first glance. However, when considering the technical aspects and overall performance, Google still falls short of OpenAI. While Gemini has demonstrated some unique reasoning abilities, many of its capabilities can already be replicated by existing models. Google’s marketing efforts should be met with scrutiny, as it aims to leverage recent uncertainty at OpenAI. Ultimately, the company’s track record and bureaucratic structure suggest that it is still trailing behind its competitors in the AI race.
Note: This article was generated by OpenAI’s GPT-3 language model.
