Elon Musk on how Grok is rebuilding the X algorithm

The algorithm on X has been aggressively promoting content based on minimal user interaction. If a user likes or even just pauses on a post, the platform has been delivering a flood of similar content. Elon Musk explained this was a bug where the algorithm's "gain" was turned up too high on any engagement.

It had too much gain on it, just turned up the gain way too high on any interaction. You would then get a torrent of that. It's like, oh, you had a taste of it. We're going to give you three helpings.

This issue was compounded by another bug that prevented posts from accounts you follow from appearing in your feed. According to Elon, fixing one problem exposed another, as the system's legacy heuristics were interconnected in complex ways. As they delete old code, they find one bug was often masking another.

The platform is now moving towards a system curated by its AI, Grok. One upcoming feature is a curated "Following" tab. Grok will analyze the posts from people you follow and highlight the most interesting ones. The broader vision is for Grok to read and understand the roughly 100 million posts shared on X every day. It will start by processing the top 10 million posts and eventually scale up to all of them. The goal is to use this understanding to surface the most relevant and interesting content for each individual user, a task impossible for humans to perform. This effort will be computationally intensive, requiring an estimated 50,000 H100 GPUs.

Grokopedia is an AI-driven alternative to Wikipedia

Three years after acquiring Twitter, now X, the mission has evolved beyond just preserving free speech. The platform's data is now a valuable asset for xAI. A key development is Grok, an AI tool that can be used on any X post to analyze its truthfulness and provide context, helping to pierce through propaganda.

This same AI has been used to create Grokopedia, a direct alternative to Wikipedia. The process involved training a version of Grok to be "maximally true seeking" and skilled at critical thinking. This AI then analyzed the million most popular Wikipedia articles, using the entire public internet to correct errors and add significant context. The speakers note that propaganda often isn't about outright lies, but about presenting technically true facts that misrepresent the full picture of a person or event. One speaker observed their own Grokopedia page was far more accurate and five to six times longer than their Wikipedia entry, which they felt had been skewed over time by people with a grudge.

Sometimes the nature of the propaganda is that facts are stated that are technically true, but do not properly represent a picture of the individual or event.

Even in its early stages, Grokopedia is considered to provide a more realistic and fleshed-out description of people, events, and even technical subjects like physics. A remaining hurdle is getting it to rank higher than Wikipedia in Google search. The path forward is through organic adoption: as people click, share, and cite Grokopedia links, its search ranking will naturally rise. Elon notes he even texted Google's CEO after discovering that searches for "Grokopedia" were initially being auto-corrected to "Wikipedia."

Building a superior source of truth with Grok and X

Grokopedia aims to be a fundamentally superior product to Wikipedia, which is described as having information that is sparse, wrong, and outdated. Elon Musk believes Grokopedia will succeed by being more comprehensive, accurate, and neutral. It will also incorporate images and AI-generated explanatory videos using Grok Imagine, covering topics from how to tie a bow tie to complex chemical reactions.

It's not going to be a little bit better than Wikipedia. It's going to be a hundred times better than Wikipedia.

However, simply presenting the truth may not be enough to change people's minds due to confirmation bias and tribalism. People often believe things simply because it aligns with their in-group's views, regardless of evidence. Elon mentions an example where individuals who believe certain negative things about Trump refuse to even watch videos that directly contradict their beliefs.

It is remarkable how much people believe things simply because it is the belief of their in group, whatever their sort of political or ideological tribe is... no matter what the evidence could be staring them in the face and they're just going to be a flat Earther.

On the platform X, new features are being implemented to combat this. The ability to invoke Grok in a reply has become a "truth-seeking missile." Users can tap the Grok symbol on nearly any post to get an analysis and research the topic further. This tool, combined with Community Notes, helps correct misinformation. Community Notes is based on consensus from people who have historically disagreed, and its code and data are open source for transparency. The ultimate goal is for X to become the best and most reliable source of truth.

Recalling the waste and absurdity at Twitter headquarters

On the third anniversary of the Twitter acquisition, the participants recalled the early days. Elon was staying at a friend's house when the deal was closing. The day before going to the Twitter office, his team requested a sink for what would become the famous "let that sink in" meme. His security team confused a local hardware store by asking for "any kind of sink," as it was purely for meme purposes and not for actual plumbing.

A friend who visited the headquarters found it shockingly empty. One of the two buildings was completely vacant, and the other had only about 5% occupancy. The cafeteria highlighted the inefficiency; there were more people making food than eating it. Elon calculated that because of the low occupancy, the real cost of a $20 lunch was effectively $400.

In an early meeting, they found all the whiteboard markers had dried out from years of disuse. However, the empty building was spotless, as the cleaning crew had continued their duties for years. The most bizarre finding was in the men's bathroom of the empty building, where a fresh box of tampons was restocked weekly. They joked about the string of improbable events that would need to occur for someone to use one.

You'd have to be a burglar who is a trans man burglar who's unwilling to use the woman's bathroom that also has tampons. You've broken into the building, and at that moment, you have a period. You're more likely to be struck by a meteor than need that tampon.

Absurd corporate waste was discovered in an empty office

An entire room filled with merchandise, including t-shirts, scarves, and hoodies with the hashtag "Stay Woke," was discovered at an office. There were also large magnetic buttons that said, "I am an engineer." This was seen as absurd.

Look, if you're an engineer, you don't need a button. Who's the button for? Who are you telling that to? You could just ship code. We would know.

The discovery was a sign that "the barbarians are fully within the gates." Beyond the cultural implications, this pointed to significant financial waste within the publicly traded company. A review found they were paying for numerous unused SaaS products, some of which had never even been installed despite being paid for for two years. The most extreme example was software that analyzed desk usage in an office where nobody worked. Another piece of software used cameras to analyze pedestrian traffic to alleviate jams in the completely empty building.

The Orwellian nature of 'Trust and Safety' and shadow banning

The most significant legacy of the Twitter acquisition is its impact on free speech, as people may have short memories of how restrictive things were previously. Figures as varied as President Trump, Jordan Peterson, and Andrew Tate were all banned. When their accounts were reinstated, it was like freeing the "bad boys of free speech."

Beyond outright bans, there was the practice of shadow banning. For years, Twitter denied this, dismissing it as a paranoid conspiracy theory. Ironically, this was carried out by the "Trust and Safety" group. Elon Musk calls this name Orwellian, comparing it to a Ministry of Truth.

I just think we shouldn't have a group called Trust and Safety. I mean, this is an Orwellian name if you ever... If there ever was one. I'm from the trust department. Oh, really? We want to talk to you about your tweets. Can we see your DMs?

It was revealed that Twitter had built an elaborate set of tools specifically to deboost accounts. This practice is not isolated; other social media companies, including Google, still do it. Suppressing search results by pushing them down the page is a subtle but effective form of censorship.

What's the best place to hide a dead body? The second page of Google search results. Because nobody ever goes to the second page of Google search results. So you could hide a dead body there and nobody would find it.

The Twitter files revealed extensive government collusion

The Twitter files exposed extensive collusion between the government and the platform's previous management. Unfettered access was given to investigative reporters, who discovered that the FBI had 80 agents submitting takedown requests. These agents were deeply involved in banning, shadow banning, and censorship. This was confirmed through internal emails, Slack channels, and database records, which later became public record after U.S. House of Representatives hearings.

Elon Musk explains that the current policy is to simply follow the law in any given country. He notes that pushing for free speech in a country that doesn't have such laws is not feasible, as the platform would simply be blocked if it did not adhere to local regulations. The goal is to avoid putting a thumb on the scale and going beyond what the law requires.

Our policy at this point is to follow the law. Now, the laws are obviously different in different countries. So sometimes I get criticized for, like, why don't I push free speech in XYZ country that doesn't have free speech laws? I'm like, because that's not the law there. And if we don't obey the law, we'll simply be blocked in that country.

This policy can lead to complex situations. For example, in Brazil, a judge ordered the company to break Brazilian law by banning certain accounts and issued a gag order to prevent them from speaking about it. This created a dilemma where a judge was instructing them to violate the country's own laws, which ultimately led to the platform being temporarily banned in Brazil.

The global movement to suppress free speech

Censorship on social media was a rapidly growing trend, expanding beyond COVID to include topics like gender and even climate change. "Content moderation" was an Orwellian euphemism for this censorship, as the definition of hate speech continually broadened, leading to more people being banned or shadowbanned. This trend would likely have continued if Elon Musk had not purchased Twitter and opened it up.

Once Twitter broke ranks, it became obvious what other social media companies were doing. This forced them to reduce their censorship policies, although some shadowbanning still occurs. In response to this new openness, however, a counter-reaction has emerged from governments. There is now a global movement to suppress free speech under the guise of stopping hate speech.

Your freedom of speech only matters if people are allowed to say things that you don't like, or even that things that you hate.

Elon argues that if you suppress speech you dislike, it is only a matter of time before the tables turn and you are suppressed. This movement to codify speech suppression into law is particularly strong in Europe, Australia, the UK, and Germany. In the UK, thousands are reportedly in prison for social media posts. According to Elon, the situation is so extreme that violent criminals have been released to make room for people who simply made posts online.

They have in a lot of cases released people who have committed violent crimes in order to imprison people who have simply made posts on social media.

This situation highlights why the founders of the United States made the First Amendment a priority. They came from places where you could be imprisoned or killed for what you said.

Elon Musk on corporate governance, activist investors, and the OpenAI lawsuit

There is a fundamental issue with corporate governance in publicly traded companies. About half of the stock market is controlled by passive index funds, which often outsource their voting decisions to advisory firms like Glass Lewis and ISS. Elon Musk refers to these firms as "corporate ISIS," arguing they have been infiltrated by far-left activists.

These firms own no stock in the companies they advise on, creating a breakdown of fiduciary responsibility. Index funds have a duty to vote in a way that maximizes shareholder returns, especially for people relying on them for retirement savings like 401(k)s. However, when this duty is outsourced to activist-influenced firms, votes may not align with shareholders' best interests.

Elon connects this issue to his role at Tesla, particularly regarding the development of the Optimus robot. He feels he needs a 25% voting stake to have a strong influence over the project's safety and prevent a "Terminator scenario."

My concern would be creating this army of robots and then being fired for political reasons because of ISS and Glass Lewis. Fire me effectively, or the activists at those bombs fire me even though I've done everything right. That's my concern. And then I cannot ensure the safety of the robots.

The conversation also turned to the corporate governance at OpenAI. Elon confirmed his lawsuit against the company is likely heading to a jury trial. He states there is a mountain of evidence showing OpenAI was created as an open-source nonprofit, with incorporation documents explicitly stating that no officer would benefit financially. He claims they violated this principle when they realized how much money could be made.

I came up with the idea for the company, named it, provided the A, B, and C rounds of funding, recruited the critical personnel and told them everything I know. If that had been a commercial corporation, I'd probably own half the company. But I created as a nonprofit for the world, an open source nonprofit for the world.

Elon argues that OpenAI should open-source its models, as that was its original purpose. He criticizes the company for releasing broken, non-working versions of their models as a "fig leaf" while noting that some of the best open-source models now ironically come from China.

Elon Musk on the ironic history of OpenAI's founding

When considering the pace of job displacement from artificial intelligence, Elon Musk describes AI as a "supersonic tsunami"—a giant wall of water moving faster than the speed of sound. He explains that he tried to slow down AI's development and created OpenAI to serve as a counterweight to Google, which at the time had unilateral power in the field.

Elon recounts a concerning conversation with Google's co-founder, Larry Page, who he felt was not taking AI safety seriously. This concern was amplified when Page called him a "speciesist" for prioritizing human survival over machine intelligence.

I was like, Larry, we need to make sure that the AI doesn't destroy all the humans. And then he called me a speciesist, like racist or something for being pro-human intelligence instead of machine intelligence. I'm like, well, Larry, what side are you on?

Given this situation, with Google holding a near-monopoly on AI, OpenAI was established as its opposite: an open-source, non-profit organization. However, Elon now ironically suggests its name should be changed to "Closed for maximum Profit AI." He describes the company's transformation as a comical, "Bond villain level flip," moving from a mission-driven entity to one that is now ravenously pursuing revenue.

What is the most ironic outcome for a company that was created to do open source nonprofit AI is it's super closed source. The OpenAI source code is locked up tight in Fort Knox. And they are going for maximum profit.

Tesla's plan to turn its cars into an AI network

Tesla has a plan to connect its entire fleet of vehicles during their downtime to offer up AI inference compute. Elon Musk explained that if the Tesla fleet reaches 100 million vehicles, each equipped with a kilowatt of inference compute, it would create a distributed supercomputer.

If ultimately there's a Tesla fleet that is 100 million vehicles, which I think we probably will get to at some point... and they have mostly state-of-the-art inference computers in them that each say are a kilowatt of inference compute and they have built-in power and cooling and connect to the Wi-Fi... you'd have 100 gigawatts of inference compute.

When considering the power efficiency of AI, the human brain serves as a clear benchmark. Our brains use only about 10 watts for higher functions, yet this has been enough to build civilization. Musk points out that today's AI supercomputers use megawatts or even gigawatts and still cannot do everything a human can. This vast difference highlights a massive opportunity to improve the power efficiency of AI compute.

We've managed to build civilization with 10 watts of a biological computer... Given that humans are capable of inventing general relativity and quantum mechanics... with a 10-watt meat computer essentially, then there's clearly a massive opportunity for improving the efficiency of AI compute because it's currently many orders of magnitude away from that.

All current Teslas are built to be robotaxis, equipped with advanced AI, software, and cameras that are subtly integrated into the car's design. In addition, Tesla is launching a dedicated product, the Cyber Cab, which has no steering wheel or pedals. Production is slated to begin in Q2 of next year, with a goal of making millions per year. Despite requests, Musk is firm that a steering wheel will not be added. He argues that people don't truly want to drive, using an analogy:

How many times have you been, say, in an Uber or Lyft and you said, 'You know what, I wish I could take over from the driver. And I wish I could get off my phone and take over from the Uber driver and drive to my destination.' How many times have you thought that to yourself?

Elon Musk on the cautious rollout of self-driving cars

The stakes for self-driving technology are very high, leading to an extremely cautious approach. Elon Musk explains that while the car is currently very capable, the company remains paranoid about safety. This is because even a single accident would create worldwide headlines, especially for a brand like Tesla.

The car is actually extremely capable right now. But we are being extremely cautious and we're being paranoid about it because... even one accident would be headline news.

Elon feels some in the press are particularly antagonistic, creating additional pressure. Despite calls to deploy the technology faster, the use of a human safety monitor is considered the correct decision, even if it draws criticism. The plan is to eventually remove the monitor.

We do expect that the cars will be driving around without any safety monitor before the end of the year. So sometime in December in Austin.

The real-world challenges of managing a robot car fleet

The initial deployment of robot cars in Austin has gone pretty smoothly. A significant part of the learning curve involves fleet management. This requires writing proprietary ride-hailing software, similar to what Uber uses, but for summoning robot cars instead of human-driven ones. As the fleet scales up to potentially a thousand cars in the Bay Area and over 500 in Austin, new challenges emerge.

For instance, the system must prevent all the cars from going to the same supercharger or congregating at the same intersection. It also needs to manage the vehicles during periods of low and high demand, deciding whether they should circle the block or find a parking spot. This leads to many oddball corner cases. A car might identify a faded disabled parking space as available or squeeze into a spot that's technically possible but leaves no room for passengers to enter.

It sees a space to park, and it's like, ridiculously tight, but I can get in there with three inches on either side. Bad computer. But nobody else will be able to get in the car if you do that.

Dealing with varying regulations across different cities and airports is another hurdle. Elon explains it's a lot of tedious work that just takes time. For example, to drop off passengers at the San Jose airport, the robot car itself must make a remote call to the airport's servers to pay the required drop-off fee. Despite these quirky challenges, the sight of cars driving around with no one inside will soon become extremely normal.

Elon Musk on Bill Gates, scientific reasoning, and the 50-year climate timeline

Elon Musk recounted a conversation with Bill Gates that left him surprised by Gates's grasp of science. Despite founding one of the world's largest technology companies, Musk feels Gates is not strong in the sciences. He shared an anecdote about a visit Gates made to the Tesla Gigafactory in Austin.

He was telling me that it's impossible to have a long range semi truck. And I was like, well, but we literally have them and you can drive them and Pepsi is literally using them right now.

Musk tried to break down the problem into its core components, asking Gates which specific metric he disagreed with, such as the energy density of the battery pack or the truck's energy efficiency. Gates did not know any of the numbers, leading Musk to conclude that it was premature for Gates to declare the truck's impossibility without understanding the underlying data.

Regarding climate change, Musk finds the discourse polarized between two untrue extremes: those who deny it entirely and those who claim imminent catastrophe. He argues for a more measured view based on observable data. The concentration of CO2 in the atmosphere is steadily increasing. Transferring trillions of tons of carbon from underground into the atmosphere and oceans will inevitably change their chemical composition. The real debate is about the degree and timescale of this change.

The reality is that in my opinion is that we've got at least 50 years before it's a serious issue. I don't think we've got 500 years, but we've probably got 50. It's not five years.

The most reasonable course of action, according to Musk, is to lean toward a sustainable, solar-battery future. He believes massive subsidies for sustainable energy are not necessary, but existing subsidies for the oil and gas industry should be removed. He classifies the long-standing special tax write-offs for oil and gas as subsidies, which prevent newer, cheaper technologies like solar from competing fairly in the market. He also highlighted China's massive solar panel manufacturing capacity, noting they could produce enough panels in 18 months to power the entire United States.

The sun's power makes solar the obvious long-term solution

While nuclear fission is a viable and very safe energy option, it pales in comparison to the sheer power of the sun. Elon Musk notes that despite propaganda and scaremongering, nuclear power has a proven safety record, exemplified by its use on U.S. Navy submarines and aircraft carriers. However, public perception remains a significant hurdle.

Generally people don't welcome a nuclear reactor in their backyard. They're not championing, put it here, put it under my bed. Put it on my roof. If your next door neighbor said, 'Hey, I'm selling my house and they're putting a reactor there,' the typical homeowner response will be negative.

To understand why solar is the ultimate solution, one must appreciate the scale of the sun. The sun constitutes about 99.8% of the mass of the entire solar system, with Jupiter making up 0.1% and everything else fitting into the remaining 0.1%. This means that even if you burned all other matter in the solar system, the total energy produced would still round up to 100% from the sun.

This concept is further illustrated by the Kardashev scale, which classifies civilizations based on their energy consumption. A Kardashev Scale 1 civilization has harnessed most of its planet's energy. A Scale 2 civilization has harnessed most of its sun's energy, and a Scale 3 civilization has harnessed the energy of its galaxy. Humanity is currently only a few percent of the way to becoming a Scale 1 civilization. As soon as you consider a Scale 2 civilization, it becomes clear that solar power is the only thing that truly matters, as the sun produces over a billion times more energy than everything on Earth combined.

Elon Musk on why the sun is the only fusion reactor that matters

The idea of beaming power from a star back to Earth is impractical; it would simply melt the planet. Power generation must be kept local and distributed, leading to a future of numerous solar-powered AI satellites. This raises the question of building fusion reactors on Earth. Elon Musk believes this is not an insurmountably hard problem. He explains that if you scale up a tokamak reactor, the challenge of containing a hot core without melting the walls becomes easier due to an improved surface-to-volume ratio, similar to the principle in rocket engines.

I think just anyone who looks at the math, you can make a reactor that generates more energy than it consumes. And the bigger you make it, the easier it is. And in the limit, you just have a giant gravitationally contained thermonuclear reactor like the sun, which requires no maintenance and it's free.

However, Musk questions the rationale behind such an endeavor. Why create a microscopic sun on Earth when a giant, free, and maintenance-free one already exists in the sky? He views Earth-based fusion reactors as a "fun science project" but considers them "peanuts compared to the sun." The solar energy that reaches Earth is already substantial, at about a gigawatt per square kilometer. With commercially available panels at nearly 26% efficiency and dense packing, one could effectively generate 200 megawatts per square kilometer. Paired with batteries, this provides a continuous power solution. The primary technical challenge isn't the solar panels, which are made of abundant silicon, but the scalability of manufacturing the necessary batteries and sourcing their raw materials.

Earth's most common elements can power the planet

When measured by mass, the most common element on Earth is iron, making up about 32% of the planet. Oxygen is the second most common at 30%. This composition essentially makes our planet a 'rusty ball bearing'. The materials needed for iron phosphate lithium-ion battery cells are all abundant. Iron is the most common element, and phosphorus, carbon, and lithium are also very common. Calculations show that there is no shortage of materials needed to power the entire Earth using solar panels and batteries. The math supporting this has been published on the Tesla website.