Last update: Sunday 3/24/24
Welcome to our 24Mar24 podcast + TL;DR summary of the past week's top 5 AI stories on our "Useful AI News" page ... (1) More powerful Nvidia chips, (2) Musk releases his chatbot's code, 3) Microsoft hires DeepMind co-founder, (4) OpenAI's GPT-5 "soon", and (5) Apple might use Google's Gemini for iPhones
A. TL;DR ... Top 5 stories in past week ...
1) More powerful Nvidia chips
This week Nvidia announced the production of "Blackwell", its newest graphics processing unit (GPU). Blackwell chips are 30 times as powerful as Nvidia's H100 chips and use far less energy.
In 2023, Big Tech's insatiable demand for Nvidia's H100 GPUs to support their generative AI services propelled Nvidia's market capitalization high enough to make it the third most valuable company in the world, behind Microsoft and Apple. Nvidia anticipates even greater demand for Blackwell.
Nevertheless, we should remember two important exceptions: Google has been designing its own GPUs for a few years now, and Microsoft recently signed a contract with Intel to manufacture GPUs that Microsoft will design. The importance of these exceptions derives from their current positions as the leading providers of generative AI services, their vast corporate wealth, and their determination to spend whatever is required to maintain their leadership positions.
2) Musk releases code for Grok-1
On March 17, 2024, the xAI corporation posted a notice on its blog that it was releasing the weights and architecture of its 314 billion parameter language model, the model underlying its chatbot Grok. In other words, Grok-1 was now an open source model.
This move was consistent with Musk's ongoing lawsuit against Sam Altman and OpenAI because of OpenAi's failure to create open source models. Musk claims that this failure violated the commitment they made to him made when Musk invested the $44 million that launched OpenAI back in 2015.
xAI's announcement did not provide the results of benchmark tests that could be used to compare its performance with the performance of other models. Both GPT-4 (Microsoft and OpenAI) and Gemini (Google) have over 1.5 trillion parameters, making each of them five times as large as Grok-1's 314 billion parameters.
CORRECTION 3/31/24 -- When we posted this TL;DR last week, we stated that Musk had not included any benchmarks. However, a subsequent look at the xAI blog note revealed a link that we had overlooked, a link to a page that displayed benchmark tests results. These tests showed that Grok-1 outperformed all other models having about the same number of parameters.
3) Microsoft hires DeepMind co-founder
Microsoft hired Mustafa Suleyman as CEO of its new Consumer AI division that will manage its Copilot, search, and browser applications. Suleyman is a co-founder of DeepMind and the current CEO of InflectionAI, a start-up whose biggest initial funding came from Microsoft. Karén Simonyan, Inflection’s co-founder, will also join Microsoft as chief scientist for the new consumer AI group, together with other members of the Inflection team.
It is highly unusual for a sponsor of a start-up to cannibalize the start-up's founding team in this
- "Inflection disclosed a licensing deal with Microsoft to make its models available for sale on Microsoft’s Azure, and said other cloud platforms would follow. It has also told investors they will fully recoup their investment, and more, as a result of the licensing agreement, according to a person briefed on the arrangement."
The Information conjectures that Microsoft is making this substantial payment to Inflection AI in order to (1) compensate Inflection AI for losses that might result from Microsoft's hiring so many of its key staff, and (2) to ward off review of the hiring by government antitrust regulators had the hiring resulted in losses for Inflection AI.
4) OpenAI's mid-year release of GPT-5
OpenAI will release GPT-5 "in the coming months" or "soon" or "mid-year" or "likely during summer" according to Business Insider in an article cited by other tech publications. In other words, this story is a well founded rumor ... that has considerable credibility because GPT-4's many shortcomings are well documented and have caused major enterprises to limit their purchases of the pricy services provided by GPT-4.
A section of last week's TL;DR ("Does generative AI really boost profits?") noted that the biggest providers of generative AI services are tempering expectations with their salespeople, saying the hype about the technology has gotten ahead of what it can actually do for customers at a reasonable price.
Indeed, if we are at the initial peak of a Gartner hype cycle, then a large scale collapse of demand for GenAI is inevitable ... unless ... unless demand can be stimulated by the release of a "new and improved GPT" ... Enter GPT-5, stage right ... applause signs flashing ... :-)
5) Apple might use Google's Gemini for iPhones
According to Bloomberg:
"Apple Inc. is in talks to build Google’s Gemini artificial intelligence engine into the iPhone, according to people familiar with the situation, setting the stage for a blockbuster agreement that would shake up the AI industry.
The two companies are in active negotiations to let Apple license Gemini, Google’s set of generative AI models, to power some new features coming to the iPhone software this year, said the people, who asked not to be identified because the deliberations are private.
Apple also recently held discussions with OpenAI and has considered using its model, according to the people."
Why would Apple prefer Google over OpenAI? Apple's commitment to privacy is more consistent with a genAI model that's small enough to run on an iPhone, rather a cloud-based model. Google has Gemini Nano, whereas OpenAI only has large cloud models, like GPT-4 and GPT-5 (forthcoming). Furthermore, Apple and Google already have a flourishing partnership via Google's search engine ... although federal regulators are challenging that partnership.
Taking a step back, we have to ask why Apple needs an AI partner? Why can't Apple go it alone? Perhaps (1) because Apple is coming very late to the genAI table and (2) because Apple may be feeling financial pressure from government regulators.
In a recent article the NY Times discussed the U.S. Department of Justice's new antitrust suit against Apple. The Times then moved on to provide a comprehensive overview of actions taken by government agencies all over the world that have contributed to Apple's slippage to second place behind Microsoft as the most valuable company in the world. As consequence, government pressure may also be moving Apple to jumpstart its late entry into generative AI via a partnership with Google.
B. Top 5 stories in past week on "Useful AI News"
- Misc
"Nvidia's next-gen AI chips are way more powerful and use a lot less energy", Alex Perry, Mashable, 3/19/24 ***
-- This story also covered by Bloomberg, CNBC, VentureBeat, FastCompany, Reuters, - Other Models
"In Latest A.I. War Escalation, Elon Musk Releases Chatbot Code", Kate Conger and Cade Metz, NY Times, 3/17/24 ***
-- This story also covered by TechCrunch, The Verge, Wired, VentureBeat, Mashable, ... and xAI - Microsoft
"Microsoft Hires DeepMind Co-Founder Suleyman to Run Consumer AI", Dina Bass, Bloomberg, 3/19/24 ***
-- This story also covered by The Verge, TechCrunch, NY Times, Forbes ... and The Information focuses on Microsoft's substantial payment to Inflection AI in order to (1) compensate Inflection AI for losses that might result from Microsoft's hiring some of its key staff, and (2) to ward off review of the hiring by government antitrust regulators if it resulted in losses for Inflection AI - OpenAI
"OpenAI is expected to release a 'materially better' GPT-5 for its chatbot mid-year, sources say", Kali Hays and Darius Rafieyan, Business Insider, 3/19/24 ***
-- This story also covered by Mashable, ZDNet, - Google
"Apple Is in Talks to Let Google Gemini Power iPhone AI Features", Mark Gurman. Bloomberg, 3/19/24 ***
-- This story also covered by The Information, Reuters
This page contains links to responses by Google's Bard chatbot running Gemini Pro to 12 questions that should be asked more frequently, but aren't. As consequence, too many readily understood AI terms have become meaningless buzzwords in the media.
No comments:
Post a Comment
Your comments will be greatly appreciated ... Or just click the "Like" button above the comments section if you enjoyed this blog note.