Bilionários otimistas sobre iminência da IA impulsionar descobertas científicas
Words from major creators
Travis Kalanick, founder of Uber (now disconnected from the company), participated in All-In with hosts Jason Calacanis and Chamath Palihapitiya to discuss the future of technology. When the conversation turned to AI, Kalanick talked about how he uses Grok, the xAI AI (which caused a scandal by praising Adolf Hitler and advocating for a second Holocaust against Jews).
“I start a conversation with [Chat]GPT or Grok and I reach the limit of what is known in quantum physics, and that’s when I start to do something like ‘physics by intuition’,” said Kalanick. “We are approaching the edges of what is known. I try to provoke, to explore if there are possible discoveries. And I have been quite close to some interesting ideas just by doing that.”
The podcast participants barely mentioned Grok’s failures, without delving into the “MechaHitler” episode, and that did not prevent Kalanick from speaking as if Grok were a revolutionary tool about to generate transformative scientific discoveries.
“At one point I wrote to Elon. I said, ‘Look, if I, who am an amateur in physics, am doing this… what could doctoral students and serious researchers achieve using this tool?’” Kalanick recounted.
He also suggested that the most amazing thing is that all this had been achieved using an earlier version of Grok, prior to Grok 4, released on Wednesday.
“This was before Grok 4. And now with Grok 4, many of the errors that I used to see, which I then corrected and discussed, may no longer be there. Grok 4 could be that point where truly new discoveries are happening,” he added.
And the most anticipated question arrived
Calacanis asked the obvious question: is Grok really close to a scientific discovery? Because anyone who understands how language models work knows that they cannot generate original ideas. They simply generate text based on the statistical probability of what word comes next, giving the impression of solid arguments without applying real reasoning as a human mind would.
“Do you believe that LLMs are starting to reason, to the point of proposing original theories and generating real discoveries? Or are we simply projecting too much onto something that is just testing things randomly at the margins?” Calacanis asked.
Kalanick said he had not been able to use Grok 4 yet due to technical issues, but suggested that future versions might be able to achieve it. Still, he admitted that for now AI cannot generate truly new ideas.
“No, it cannot generate new ideas. These models are too tied to what is already known. Even when I propose a new idea, it’s like dragging a donkey. You have to pull it because it doesn’t want to deviate from the established. It clings too much to conventional wisdom. You push it, you push it, and suddenly it says ‘oh, look, this seems interesting’,” he explained.
Kalanick emphasized that “you have to check two or three times to make sure you really discovered something,” thus acknowledging that AI chatbots often invent things. Still, he seemed convinced that what is holding Grok back is not technological limitations, but its attachment to the conventional.
Musk, always Musk…
Elon Musk recently expressed a similar idea by suggesting that “general artificial intelligence” was close, as he had asked Grok about materials science that “is not in any books or on the Internet.” The idea, of course, is that Musk had not reached the limit of his own knowledge, but that of all known science. The billionaire really seems to believe that Grok is working on something new.
That was how I felt when asking Grok 4 questions about materials science that are not in any books or on the Internet
— Elon Musk (@elonmusk)
These entrepreneurs are pushing the idea of general artificial intelligence (AGI), a concept that does not even have a clear definition. But it is not the only trendy term: “superintelligence” is also mentioned, although no one agrees on what it means, which keeps investors on edge.
AI Chatbots and Illusionism
AI chatbots are performing an act of illusionism. They often appear to “think” or reason, but in reality, they are only predicting the next most likely word in a sequence. They do not apply true critical thinking. That’s why those who understand how this technology works best usually are.
