On Tuesday, Alphabet (NASDAQ: GOOGLE ), Google’s parent company, introduced its new search assistant, Bard, a rival to OpenAI’s ChatGPT.
The company said it will begin making Bard available in the coming weeks, opening it for testing with trusted users before releasing it to the general public.
The tool is based on LaMDA, a language model for dialogue applications developed by Google itself.
Reacting to Google’s new AI chatbot, analysts at Morgan Stanley (NYSE: MS ) told investors that the bank believes the tech giant has the technology and scale to maintain/increase its user base.
However, deeper-than-expected integration “increases the risk of cost overruns, as we estimate that for every 10% of searches that migrate to language models, operating expenses rise by about US$ 1.2 billion.”
“The artificial intelligence race has begun,” the analysts added. “Our work with natural language queries suggests they may be up to five times more expensive (on average).”
The bank’s analysts explained that the computational intensity of natural language models, which store, retrieve, analyze, and compile large amounts of text in the form of natural language responses, is very high.
The bank’s updated analysis of ChatGPT’s model size, its computation time, the average number of words generated per query, Nvidia A100 GPU pricing levels on Azure (NASDAQ: NVDA ) and an estimated 50% Azure gross margin (for this analysis) leads them to believe that GOOGLE’s average extra natural language cost per query should range between US$ 0.0022 and US$ 0.0220.
“Early demonstrations show how natural language computational costs tend to be higher. But, in our view, the biggest risk of this technology comes from the greater potential for additional costs due to deeper-than-expected integration of natural language into core search results,” the analysts wrote.
They concluded that if 50% of queries are integrated with natural language in 2024, the additional costs would total 6 billion dollars.
Reference source: investing.com
