Meta is reportedly building an advanced AI model equating to the efficiency of OpenAI’s language model, GPT-4. The AI tool will reportedly produce text and analysis that will target businesses, in the the company’s latest bid to compete with the flood of new AI models, according to The Wall Street Journal, citing unnamed sources.
The AI model is aimed to assist businesses with services including building sophisticated text, and analyses, as well as assisting with additional output, sources told The Journal, with the company working to acquire H100s. The advanced Nvidia chips are used to train AI software and will allow Meta to build its new AI model.
Meta CEO Mark Zuckerberg had previously partnered with Microsoft, which owns OpenAI, to create his Llama 2 software using its Azure platform. Now, the endeavour to create a newly reformed product comes shortly after Meta introduced Llama 2, a free AI language model intended for research and commercial use.
Llama 2 was introduced to “benefit everyone,” Meta said in a July press release. “Giving businesses, startups, entrepreneurs, and researchers access to tools developed at a scale that would be challenging to build themselves, backed by computing power they might not otherwise access, will open up a world of opportunities for them to experiment, innovate in exciting ways, and ultimately benefit from economically and socially,” the release added.
Zuckerberg’s new AI model is reported to roll out next year, people familiar with the matter told The Journal but will still face competition from other major products, like ChatGPT and Google’s Bard, that aren’t slowing down on their own development. However, The Journal reported that Zuckerberg allegedly plans for Meta’s AI tool to be open source, meaning it offers free access for users.
Zuckerberg said he is optimistic about Meta’s “groundbreaking AI products in the pipeline” during an earnings call in July. “I’ve said on a number of these calls that the two technological waves that we’re riding are AI in the near term and the metaverse over the longer term,” he said. “Investments that we’ve made over the years in AI, including the billions of dollars we’ve spent on AI infrastructure, are clearly paying off across our ranking and recommendation systems and improving engagement and monetization.”
While Meta and other companies compete to produce the most advanced AI systems, concerns about these tools’ reliability continue to grow, with experts noting that AI tools are becoming increasingly inaccurate.
Researchers at Stanford University and UC Berkeley released a report last month that GPT-4’s accuracy has dropped in recent months. The report revealed that in March, GPT-4 was 97.6% accurate when identifying prime numbers, but when conducting the same study in June, researchers found GPT-4’s accuracy dropped to just 2.4%.
Meta did not immediately respond to Gizmodo’s request for comment, but Zuckerberg addressed future AI products, saying the company is “building leading foundation models to support a new generation of AI products.”