- Elon Musk highlights rapid AI breakthroughs occurring daily across the industry
- Anthropic launched Claude Fable 5, a powerful model with advanced safety features
- Musk predicts AI smarter than humans within one to two years, hinting at AGI
The pace at which AI is advancing is a "head spinner" according to Tesla and xAI founder Elon Musk. "At this point, when I go to sleep, there's some AI breakthrough, when I wake up, there's some AI breakthrough, and by lunchtime, there's another AI breakthrough," the outspoken billionaire tech founder recently said at an event.
Musk is not exaggerating the pace. On Tuesday Anthropic publicly launched its most powerful model yet - Claude Fable 5 whose intelligence is at par with its Mythos model, the same one that was unveiled a couple of months back. Mythos had sent shockwaves across the world for its ability to expose decades old vulnerabilities in critical systems such as the banking system. The public version has now been launched with guard rails where if a bad actor tries to exploit it to, say build a bioweapon or hack into a critical infrastructure, the system will automatically slip back into the earlier Claude Opus 4.8 model. "Claude Fable 5 handles complex, multi-stage knowledge work with minimal oversight, from deep research and analysis to deliverables ready for your review. Teams can hand off large projects and review completed work rather than supervising every step," Anthropic said in a statement.
According to Musk: "It's pretty obvious that we're going to have AI that is vastly smarter than humans. In some ways, it already is.
It seems like AI unequivocally smarter than humans in every way, including in innovation, is probably not more than a year or two away, I would say."
Musk seemed to be hinting at AGI or artificial general intelligence (a type of AI that matches or surpasses human capabilities across virtually all cognitive tasks) here. The statement is consistent with Musk's AGI timeline. Others such as Google AI chief Demis Hassabis also have said that AGI is knocking on the doors of humanity.
Musk's friend-turned-foe, OpenAI co-founder and CEO, Sam Altman believes AGI could be achieved as early as 2028. During the India AI Impact Summit in February this year, he outlined that AI could achieve "superintelligence" or true AGI-level capabilities within two years. Dario Amodei, co-founder and CEO of Anthropic echoes Altman's timeline and believes it should not be much longer beyond 2027 when "AI systems are better than humans at almost everything...and eventually better than all humans at everything."
However, there are experts such as Yann LeCun - the former chief AI scientist at Meta, often described as one of the 'godfathers of AI' - who believe AGI is overrated and calls the concept "bulls**t". Another key figure - Oxford professor Michael Wooldridge who's also the AI chair for the hallowed university - downplays the AGI hype, noting that it remains a distant prospect.
Are AI Leaders Overpromising?
Not everyone is convinced that AGI is knocking on the doors of humanity though. "I'm not too optimistic on this timeline. Two issues: some issues related to realistic human behaviour can't be demonstrated purely by transformer models... need activity models. Moreover, the volume of original data to train is shrinking due to infusion of large generated data," AI expert and CTO, AiEnsured Srinivas Padmanabhuni told NDTV. Coursera describes transformer models as deep learning systems that process and understand entire sequences of data at once instead of word-by-word, using something called "self-attention" to grasp context and meaning. Activity models meanwhile are designed to understand and predict human behaviour and actions such as walking or running to more complex daily activities.
And Dr Padmanabhuni is not alone when he talks about original data that's used to train AI shrinking. Quick context here: Artificial intelligence relies on vast amounts of data to train itself.
At the beginning of last year, Elon Musk himself said that models have already run out of human-created data, and have turned to AI-generated information to teach itself. "The cumulative sum of human knowledge has been exhausted in AI training," Musk had said. "That happened basically last year (2024)."
The widespread use of synthetic data to train models is obviously dangerous for training models, Musk had warned. It increases the chances of hallucinations or nonsensical content also known as AI slop where AI believes it to be true and will also try to convince you with all its might that it is indeed factual. Over the last year or so AI slop has been a real menace where it has peddled wrong information all over social media and the internet.
It's important to note that Musk's SpaceX of which xAI is a part, Amodei's Anthropic and Altman's OpenAI are all IPO bound and they may be in part playing to galleries in the run up to going public.
