Open Source Llama 3 was Nipping at OpenAI’s Heels

When open AI will become more powerful: The Llama 3 beta release feels like a game-changer for Sizzle and Google

Meta’s previous model, called Llama 2, was already influential, but the company says that it made the newest version even more powerful by feeding it more high-quality training data, and using new techniques to choose the best mixture of datasets to use.

The release last Friday feels like a game-changer according to Pesenti. GPT-4 and other closed and open artificial intelligence models are currently being used by Sizzle to craft problem sets and curricula for students. His engineers are evaluating whether Llama 3 could replace OpenAI’s model in many cases.

But the near-term fortunes of Microsoft and Google, at least as far as their generative AI efforts are concerned, look different under the hood and in the comments of their executives. If investors, workers, and potential customers perceive the rivals in a certain way, it will determine which gets the better portion of the hundreds of billions of dollars in spending expected to flow to such software.

“It’s going to be an interesting horse race,” Pesenti says of competition between open models like Llama 3 and closed ones such as GPT-4 and Google’s Gemini.

The cost of running Llama 3 on a cloud platform is less than that of accessing GPT-4 through an interface. He says that Llama 3 can respond to queries quickly, a factor that is important for companies that rely on models from different providers. “It’s an equation between latency, cost, and accuracy,” he says.

The Llama 3 shows that if the models are smaller, they can be run on less powerful hardware. One version of the model has 70 billion parameters, a measure of the variables it learns from training data, and the other has 8 billion. The smaller model is compact enough to run on a laptop but is remarkably capable, at least in WIRED’s testing.

Two days before Meta’s release, Mistral, a French AI company founded by alumni of Pesenti’s team at Meta, open sourced Mixtral 8x22B. It employs only 39 billion of the 141 billion parameters it has at any one time, a design known as a mixture of experts. Thanks to this trick, the model is considerably more capable than some models that are much larger.

All of this openness is not altruistic. Meta CEO Mark Zuckerberg says opening up its AI models should ultimately benefit the company by lowering the cost of technologies it relies on, for example by spawning compatible tools and services that Meta can use for itself. He left unsaid that it may also be to Meta’s benefit to prevent OpenAI, Microsoft, or Google from dominating the field.

Alphabet CEO Sundar Pichai is confident that Google will find a way to make money selling access to generative AI tools. Microsoft CEO Satya Nadella says his company is already doing it.

Both companies reported sales and profit that were better than expected. And the stock prices of both soared on the results, with Alphabet further buoyed by its new plans to buy back more shares and issue its first-ever dividend.

Google Doesn’t Know It Can Cash In on Generative AI: How Microsoft Earns It Already Has? A Conversation with Financial Analysts

In a call with financial analysts on Thursday, Nadella touted that Microsoft now has 1.8 million customers for GitHub Copilot, a generative AI tool that helps engineers write software code. That’s up from 1.3 million customers a quarter ago.

Among Fortune 500 companies, 60 percent are using Copilot for Microsoft Office 365, a virtual assistant that uses generative AI to help workers write emails and documents, and 65 percent are using a Microsoft Azure Cloud service that enables them to access generative AI software from ChatGPT-maker OpenAI. “Azure has become a port of call for pretty much anybody who is doing an AI project,” Nadella said. The millions of dollars Microsoft has invested in OpenAI helped to win the clients.

The buzz of interest in artificial intelligence helped drive revenue for Microsoft’s biggest unit -cloud services – up by seven percentage points, and the company’s overall sales rose 17 percent. It also gained cloud market share, Nadella added. Microsoft landed more than one hundred million dollars in cloud deals during the quarter, an 80 percent increase compared to the same period a year ago.

But Pichai didn’t say how many signups Google had drawn to Gemini Advanced, a $20 per month subscription plan announced in February that provides access to the company’s most advanced AI chatbot.

While Pichai said the tests show that users exposed to generative AI-powered search are doing more searches, they are also potentially less profitable for Google because the underlying technology to power more advanced searchers is costlier than operating its longstanding systems.

Picahi was not concerned on either front. He was confident that they could handle the costs of how to serve the queries. I am certain we will be able to manage the monetization transition here as well. It will play out over time.”

Source: Google Thinks It Can Cash In on Generative AI. Microsoft Already Has

Microsoft Pays off $13$-Billion in the First Two Months of the Windows X Spectrum: Where Are We Going? How Well Does Microsoft Predict the Future?

It spent the same amount as Microsoft did last quarter, around $12 billion. But the results and comments on Thursday suggest that Microsoft is further along in delivering a payoff.