A couple of days in the past, Microsoft introduced an “evolution” of their partnership with OpenAI, which suggests a notable shift of their collaboration dynamics. OpenAI has been utilizing Microsoft’s cloud providers solely, however now it has the choice to select from different cloud platforms. Microsoft now has the best of first refusal (ROFR) when OpenAI requests extra computing sources from the corporate.
“This new settlement…contains modifications to the exclusivity on new capability, transferring to a mannequin the place Microsoft has a proper of first refusal (ROFR),” the announcement acknowledged.
Notably, there was hypothesis for fairly a while that the connection between these two firms has deteriorated.
What was stunning is that Salesforce CEO Marc Benioff, on a uncommon event, made a optimistic prediction about Microsoft. “It’s extraordinarily necessary that OpenAI will get to different platforms rapidly as a result of Microsoft is constructing their very own AI,” he stated.
Benioff additionally speculated that Microsoft would have its personal frontier fashions and stated, “I don’t suppose they are going to use OpenAI sooner or later.”
Whereas Microsoft has not formally introduced any plans to construct a big frontier mannequin, sure indicators might counsel such a chance.
Microsoft Will Must Take a Phi-4 Strategy
The most recent in Microsoft’s Phi sequence, the Phi-4, packs in 14B parameters. As per the benchmarks, the Phi-4 outperformed the Llama 3.3 70B and OpenAI’s GPT-4o in a number of benchmarks.
Not solely did this point out spectacular efficiency on a small mannequin, however Microsoft additionally solved an important drawback that many AI researchers are discussing: knowledge shortage. The mannequin depends totally on high-quality artificial knowledge to realize success. In contrast to OpenAI, the corporate has additionally not used any inference optimisation strategies.
That is vital progress in breaking the barrier of scaling legal guidelines. “Blindly scaling, like how individuals have been doing with trillion parameter fashions, isn’t simply wanted, proper?” Harkirat Behl, member of technical employees at Microsoft AI, earlier informed AIM.
Benchmarks are one other solution to gauge a mannequin’s efficiency. For a very long time, leakage of benchmark knowledge within the coaching corpus and overfitting the mannequin’s efficiency to the check units have led to unfair outcomes.
Nevertheless, with Phi-4, this concern might have been eradicated in some methods. Microsoft examined the mannequin on a benchmark developed after gathering all of the coaching knowledge. On these benchmarks, specifically the November 2024 AMC 10/12 exams, it outperformed a number of rivals.
On condition that Phi-4 is open supply, a number of builders had been in a position to check its capabilities. The mannequin hit 50,000 downloads on Ollama, an open-source AI mannequin platform, in simply three days and 40,000 on Hugging Face.
A number of builders had been in awe of the mannequin’s efficiency, particularly in restricted {hardware} circumstances.
I’m amazed on the phi-4 efficiency and outcomes even on modest {hardware}. That is on a 12GB GPU through @ollama—it opens up all types of edgy & native use instances. pic.twitter.com/3AruIt1rbB
— Ben Younger (@benyoungnz) January 9, 2025
Furthermore, given the legacy of the corporate, and the capital, there’s little question about investing in GPU clusters to take action.
Amidst the hypothesis, Microsoft CEO Satya Nadella stated, “We’re not going to do issues twice.” His assertion signifies that the corporate will take advantage of OpenAI’s present fashions.
“The extra necessary factor for me is to construct worth on high of OpenAI. So we have now a unbelievable post-training stack,” he added.
‘Mustafa Suleyman and Sam Altman Aren’t Greatest Pals’
In line with Benioff, Microsoft employed its AI CEO Mustafa Suleyman from Google DeepMind to construct frontier fashions.
He added that “Suleyman and [OpenAI CEO] Sam Altman will not be greatest pals”.
Nadella, nevertheless, dismissed any indicators of a deteriorating relationship. “Our partnership continues,” he stated, including that he believes that the brand new particulars are solely going to be useful for each, particularly within the context of The Stargate Undertaking.
For this mission, large tech firms, together with Oracle, Softbank, and OpenAI, are pooling $500 billion {dollars} to construct knowledge centres the place OpenAI, amongst many different AI firms, can have entry to unprecedented ranges of sources.
“Sam needs to proceed with the scaling legal guidelines to construct out extra compute to ensure that him to coach extra fashions and we have now a roofer. So he involves us first,” Nadella stated.
“If we meet these wants, then we clear it. If not, he can go to those different suppliers. And so, I feel it really works out properly for Sam and for us,” he added.
Notably, Microsoft’s newest announcement in regards to the partnership carried the ROFR element on the very finish.
A majority of it principally targeted on the exclusivity of OpenAI IP to Azure, their revenue-sharing agreements. The announcement additionally acknowledged that “OpenAI just lately made a brand new, giant Azure dedication that may proceed to help all OpenAI merchandise in addition to coaching”.
Nevertheless, the connection carries a number of layers. Not too long ago, Microsoft and OpenAI have reportedly agreed on a brand new, particular definition of synthetic common intelligence (AGI).
OpenAI can solely obtain AGI when it has constructed a system that may generate $100 billion in income. That is necessary as a result of a clause within the settlement reportedly means that Microsoft will now not have entry to OpenAI’s fashions when the latter achieves AGI.
If the brand new definition of AGI is something to go by, OpenAI is way from it. This yr, OpenAI noticed a $5 billion loss, with $3.7 billion in income. Experiences additionally counsel that OpenAI is not going to flip right into a worthwhile entity till 2029.
Furthermore, OpenAI’s losses raised scepticism about how they may contribute to The Stargate Undertaking. In a publish on X, xAI founder Elon Musk claimed that OpenAI doesn’t even have any cash.
The publish Will Microsoft Construct Frontier Fashions to Scale back Reliance on OpenAI? appeared first on AIM Media Home.
