New York Times Report
Editor: Editorial Department
New York Times Digest: These days, the AI community's attention has been captured by Musk and LeCun. Whether it's about AI safety regulations or the mission of xAI, the two have engaged in a fierce online dispute.
Yesterday, two heavyweight figures in the AI field, Yann LeCun and Elon Musk, clashed on social media over the safety and regulation of artificial intelligence.
AI pioneer LeCun expressed his views on ongoing AI regulation on platform X.
LeCun believes that it is premature to worry about the "existential risks" posed by artificial intelligence, let alone regulating it.
"Artificial intelligence is not a natural phenomenon that suddenly appears and becomes dangerous. We design it, we build it."
He emphasizes that the safety of AI depends on human design, and there is currently no need to worry excessively.
In a humorous response, Musk quips, "Prepare to be regulated," hinting at possible future regulations.
Musk and LeCun Engage in Intense Debate on AI Regulation
When it comes to AI regulation, Musk and LeCun have never been on the same page. Their recent exchange sheds light on their differing perspectives.
In response to Musk, LeCun doesn't hold back, asking, "Are you happy with 'not training models exceeding 10^26 flops'?"
The mention of "do not train models with more than 10^26 flops" by LeCun comes from the first White House regulation on generative AI signed by Biden last October.
In fact, the introduction of this regulation owes much to Musk, an active supporter of regulation. He consistently shows great fear of AI getting out of control while simultaneously striving to build his own AI empire.
On the other hand, LeCun, a proponent of AI progress, is also a strong advocate for open source and sharing.
LeCun's stance: Embrace openness, resist regulation
LeCun tweeted against the backdrop of increasing calls for stronger AI regulation amidst a significant exodus from OpenAI's AI safety team.
LeCun's stance sharply contrasts with views expressed by former OpenAI board members.
The latter have consistently advocated for stricter AI regulation, believing that taming market forces is essential for the well-being of all humanity.
Even OpenAI's embattled CEO, Altman, expressed concerns about the "subtle misalignments" that can make AI dangerous. He stated that "killer robots" are unlikely, but the potential risks of "misalignments" should be taken seriously.
Furthermore, Google's CEO, Pichai, also agrees with Altman and Musk on the importance of proper AI regulation.
However, LeCun's post indicates that the AI community is far from reaching a consensus on this issue.
Just like the significant divide on the open-source vs. closed-source debate.
LeCun expressed approval under a comment from a netizen suggesting, "Understanding how the AI we design and build works might be a good idea."
"Yes, that's why I have always strongly advocated for open research and open-source," he replied.
Musk: Regulation is necessary, but competition is key
Musk has always been a proponent of regulation.
He has repeatedly warned of the dangers of artificial intelligence, stating that "AI is more dangerous than poorly maintained aircraft design or production, or bad car production. In a sense, it has the potential, no matter how small people think that possibility is, it is not insignificant - it has the potential for civilization destruction."
This includes his famous open letter calling for a 6-month pause in the "uncontrolled" AI development race.
He has also attended closed-door meetings with legislators as a top tech leader and publicly supported government regulation of artificial intelligence.
While Musk acknowledges that "being regulated is not fun," he believes that once AI "might control everything," it may be too late to set regulations.
However, this has not stopped Musk from taking action or deep involvement in AI development.
Through investments in his vast corporate empire, Musk has engaged in a broader AI arms race.
He is working tirelessly to create a product that can rival AI offerings from tech giants like Microsoft and Google.
Musk describes it as "AI that seeks truth to the maximum extent," and "cares about understanding the universe."
At the same time, he also sponsors other AI projects.
This has led to mockery from netizens:
LeCun mocks Musk, gets blasted by netizens.
Meanwhile, under Musk's recent retweet of a job posting from his own xAI company, LeCun starts throwing shade again.
Elon Musk's caption:
LeCun's comments are simply astonishing.
If you can tolerate such a boss, join xAI:
Claim that the work you are doing will be solved next year (no pressure)
Claim that what you are doing will kill everyone, must be stopped or paused (6 months of vacation)
Claim to want to "strictly pursue the truth to the maximum extent," but spread crazy conspiracy theories on your own social platform
LeCun made this statement because last week Musk publicly stated that AGI will arrive next year, and he estimated the possibility of it destroying humanity to be 10%-20%.
The comments section is all firing shots at LeCun. Some netizens are asking, why so harsh lately?
Someone also made a meme of Zuckerberg assigning tasks to LeCun, aiming to divert everyone's attention.
Others have also expressed, "The post you made seems like you want to work for Musk, but it's better to work for LeCun."
LeCun stated, "I'm not anyone's boss, even if you join FAIR, you won't be working for me."
In the following longer tweet, netizens take sides with Musk and engage in a heated debate with LeCun.
Musk responded, "That's nothing, you're getting weaker, push harder!"
LeCun suddenly felt amused, "Sounds like you're my boss."
After seeing Karas' father, you need to have a "cage fight" to settle the score.
In just 10 months, Musk's xAI is valued at one-third of OpenAI's.
Information once revealed that Musk will build the world's largest "supercomputer factory," equipped with 100,000 H100 chips to train the next generation Grok model.
Just yesterday, Musk announced that xAI has completed a staggering $6 billion Series B financing.
Reportedly, the company's valuation has reached $24 billion.
Investors in Series B funding include Andreessen Horowitz, Sequoia Capital, Valor Equity Partners, Vy Capital, and Fidelity Management & Research, among 8 investors.
The largest Series B funding in history
Last December, xAI completed its first round of funding, raising a total of $134.7 million.
Established in July 2023, xAI has garnered significant favor from the capital market within just 10 months since its inception.
Currently, xAI's scale has reached one-third of OpenAI's. Data from February indicates that OpenAI's valuation has soared to $80 billion.
This latest round of funding marks a further escalation in the AI arms race.
In recent years, leading AI startups have raised billions of dollars, which are being poured into the computational power needed to train advanced large models.
Currently, OpenAI has secured a $13 billion funding commitment from Microsoft, while Anthropic's funding has exceeded $6 billion.
xAI has been able to attract such a large amount of funding due to the efforts of its team, their research achievements, and their mission.
Let's first take a look at the strong backgrounds of xAI's research team members, who come from prominent companies such as Google, DeepMind, Microsoft, Meta, OpenAI, and others.
On the other hand, xAI has achieved significant results over the past six months.
From the release of Grok-1, to the iterative upgrades of Grok-1.5 and Grok-1.5V, as well as the open-sourcing of Grok-1 and the PromptIDE tool.
In November last year, the groundbreaking Grok-1 model made its debut, aiming to target OpenAI's ChatGPT.
Empowered by the new model, the chatbot Grok can multitask, run multiple conversations concurrently, and seamlessly switch between different dialogues.
In various benchmark tests, Grok-1 has shown advantages over GPT-3.5 in mathematics, programming, and MMLU.
It is worth mentioning that achieving such a breakthrough was the result of intense training by 11 researchers at xAI over a period of 2 months.
In March of this year, Grok-1, based on the MoE architecture, was fully open-sourced with a staggering parameter count of 314 billion, making it the largest open-source model at the time.
Grok-1 is also released under the Apache 2.0 license via magnet link.
Shortly after, in less than half a month, Musk unveiled an upgraded version of Grok-1.5, with the key feature being its support for 128k context and significantly enhanced reasoning capabilities.
The benchmark results show that Grok-1.5 has seen comprehensive performance improvements across all aspects, bringing it closer to the world's leading large models.
Just last month, the multimodal version of Grok-1.5V was born, claiming to bridge the physical and digital worlds.
In addition to its basic text capabilities, Grok-1.5V can also handle visual information such as documents, screenshots, and photos.
For example, it can generate Python code from an image.
In performance testing, Grok-1.5V is on par with GPT-4V, and in some aspects, it even surpasses GPT-4V.
Since its inception, Musk's vision for xAI has been to seek answers to life, the universe, and everything.
Moreover, it is an AI company dedicated to exploring the "ultimate truth" and the "essence of the universe."
The official account stated that this round of financing will be used to launch the first batch of xAI products into the market, to build advanced infrastructure, and to accelerate the research and development of future technologies.
Reference:
Editor: Gao Xuedan_NT5097
