New Delhi: There was more amuse than surprise when Meta CEO Mark Zuckerberg agreed to a siege with Tesla CEO Elon Musk was there. In fact, the expectation of witnessing two billionaires fight in a cage, regardless of the outcome or speculation about who was the stronger fighter, was enough meme food for netizens. .
New Delhi: There was more amuse than surprise when Meta CEO Mark Zuckerberg agreed to a siege with Tesla CEO Elon Musk was there. In fact, the expectation of witnessing two billionaires fight in a cage, regardless of the outcome or speculation about who was the stronger fighter, was enough meme food for netizens. .
Musk and Zuckerberg were never friends, but the brain-battle was sparked in a podcast interview with Massachusetts Institute of Technology researcher Rex Fridman on June 9, in which Zuckerberg said: . A billion people use it…”
Hi!reading premium articles
Musk and Zuckerberg were never friends, but the brain-battle was sparked in a podcast interview with Massachusetts Institute of Technology researcher Rex Fridman on June 9, in which Zuckerberg said: . A billion people use it…”
Musk’s Twitter deal closed Oct. 27, 2022, with dramatic events including the dismissal of Chief Executive Officer Parag Agrawal, Chief Financial Officer Ned Segal and Chief Legal and Policy Officer Vijaya Gadde Some may remember Musk justified the layoffs, citing a loss of $4 million a day. But while the Musk administration prompted many to abandon Twitter, the microblogging site appears to have weathered the storm.
That said, Twitter has only a handful of users compared to Meta, which owns four of the most popular global social media platforms: WhatsApp, Facebook Messenger, Facebook and Instagram, and has over 3.5 billion monthly users. The fact remains that there are only 354 million people. And the irony of Meta Chief’s podcast remarks wasn’t lost on Musk, resulting in an online verbal exchange that led Zuckerberg to accept the challenge of a physical fight.
There may or may not be a duel in the end, but Musk’s mother, Maye Musk, insists on a verbal duel rather than a physical one. But a closer look reveals that the war of words isn’t just a Twitter issue, so the two companies are still battling for AI mindshare, and it’s not over yet, but it’s not as obvious as the battle for online centerpieces. .
Note that Musk and Sam Altman founded OpenAI in 2015 before the Tesla president stepped down from the company’s board of directors in 2018. Currently, he is reportedly launching a team to develop a chatbot BasedAI that utilizes AI. Musk is wary of Microsoft’s investment in OpenAI. OpenAI’s generative AI bot ChatGPT has swept the internet by gaining over 100 million users within two months. Musk also accused OpenAI of “seeking profit” rather than developing technology for humanity.
In April, Musk was among 1,000 participants who asked for a six-month moratorium on a training system “more powerful than GPT-4,” prompting the world to believe such a system could contain risk. I argued that it should be developed when Ironically, Musk recently founded a new AI company, X.AI Corp, which was incorporated in Nevada and listed Musk as its sole director, with Jared Virtual, director of Musk’s family office. as a secretary. It’s not yet time to make things public.
Some might argue that synergies between companies owned by Mr. Musk are unlikely. For example, SpaceX builds spacecraft and rockets, Tesla designs and manufactures electric vehicles, home to grid-scale battery energy storage systems, solar panels and roof tiles, and Twitter is a microblogging platform.
However, a closer look reveals that Musk has strategically aligned with his company, accessing vast amounts of personal data from tweets and using the help of Neuralink, a company he co-founded. It is suggested that they are subdividing and building an information powerhouse. , develops high-bandwidth brain-machine interfaces, enabling seamless interaction between computers and humans. By harnessing the power of intelligent data, enormous economic opportunities await Musk, which will provide the potential rationale behind his incorporation of X.AI.
Mr. Zuckerberg is by no means passive. With Google focusing on his AI and Microsoft making progress with his investment in OpenAI, he launched his Meta, taking a metaverse-first approach and distanced himself from Facebook’s roots to create its own built the way
Unfortunately, history hasn’t been kind to Zuckerberg. By October, Meta had invested a staggering $10 billion to $15 billion in the Metaverse, and investors and shareholders weren’t too happy because so much cash was wasted in the effort.
Meta now seems to strike a balance between the Metaverse (not as abandoned as many people think it is) and AI. The company has created an AI sandbox that “serves as a testing ground for early versions of new tools and features, including AI-powered generative advertising tools.” Meta Advantage is a portfolio of automated products that use AI and machine learning to optimize campaign results. We also announced Meta Lattice, a new AI-driven model for improving the performance of ads on the network.
Additionally, Meta had to shut down two AI chatbots, BlenderBot and Galactica, due to glitches, but released Large Language Model Meta AI (LLaMA) and made the code available for researchers to test. regained trust. Meta’s LLaMA claims that “much less computing power and resources are required to test new approaches, validate the work of others, and explore new use cases.”
In addition, Meta’s chief AI scientist Yann LeCun, one of the so-called godfathers of modern AI and who did not sign a call to suspend models like GPT4, said that introducing new models would help AI It took the battle for mindshare one step further. It’s called the Combined Image Embedding Prediction Architecture (I-JEPA), which Meta touts as the first AI model that will “get one step closer to human-level intelligence in AI.”
In a June 13 blog, LeCun explained that I-JEPA learns by creating an internal model of the outside world that compares abstract representations of images, rather than comparing images at the level of individual pixels. I’m here. The representations learned by I-JEPA can also be used for various applications without requiring “extensive fine-tuning”.
LeCun said, “We were able to run a 632 million parameter Visual Transformer model in less than two days using 16 A100 graphics processing units (GPUs), compared to other methods that typically take between 2 and 10 times the GPU time. Here’s an example of training with only 12 labeled samples per class.” Training with the same amount of data has even worse error rates (https://arxiv.org/abs/2301.08243). Meta has open sourced their training code.
Generative architectures typically learn by removing or transforming some of the inputs to the model, says LeCun. For example, corrupted or missing pixels or words after erasing part of a photo or hiding part of a word in a passage of text. But along the way, generative techniques can make mistakes that humans never make. This is because generative techniques focus on irrelevant details instead of capturing “high-level predictable concepts”. In this context, LeCun gives an example of how difficult generative models can be. To accurately generate human hands (models often add extra digits or make other obvious mistakes).
The idea behind LeCun’s vision, as he puts it in his blog, is that humans learn a lot about the world simply by observing it, or “common sense background knowledge.” is. Meta says he wants to extend JEPA’s approach to other areas such as images. The availability of text pair data and video data improves existing generative AI models.
Even with Google and Microsoft monitoring Meta’s AI developments, it will be interesting to see what Musk pulls out now.
