埃隆·马斯克的“反觉醒”百科全书：Grokipedia的真相与偏见

I’m not entirely sure what has been going on, but If you have been following the story of Elon Musk’s push to get a trillion dollar payday out of Tesla, it seems entirely possible that Musk has just been coming up with side projects – like his new online encyclopedia - Grokipedia – so that he could convince Tesla shareholders to pay him a huge sum of money, otherwise he will abandon them in pursuit of these other interests.

And it’s believable too… a few years ago he bought Twitter so that he could become a chat forum moderator.

Yesterday, Tesla shareholders caved in and approved a pay package that could make Musk the world’s first trillionaire, and grant him control of a quarter of the company’s shares if he hits a series of ambitious targets – so maybe this whole grokipedia thing is over.

If you haven’t heard about it, the “Grok” in Grokipedia refers to Elon Musk’s generative AI chatbot which is featured prominently on his social network- or everything app - Twitter - which he calls X.

With Grokipedia, Musk’s stated goal is to create an open source, comprehensive collection of all knowledge, then place copies of that – etched in a stable oxide (whatever that means) in orbit around the moon and mars to preserve it for the future.

Now I checked with Grok as to what he means by a stable oxide, and Grok said that he probably means glass.

Anyhow… Musk says that Grokipedia will be “a compendium of the truth, the whole truth, and nothing but the truth” — a lofty promise from someone who – not so long ago - offered Wikipedia a billion dollars to rename itself “Dickipedia.”

He has framed the project as a purge of propaganda, a replacement for what he calls “legacy media,” and a step toward building a more open-source repository of knowledge.

But given Grokipedia’s reliance on his chatbot Grok - the result may be less Encyclopedia Britannica and more Reddit-meets-Twitter trolls – as his chatbot does train on sources like that.

According to xAI, Grok will be responsible for all fact-checking on Grokipedia, which is a bit like asking a parrot to verify a Shak espeare quote: it might sound convincing, but you wouldn’t want to bet your reputation on its accuracy.

Wikipedia, in Musk’s view, has gone soft — He used to like it – but now it’s too woke, too establishment and too unwilling to include the kinds of sources that flatter his worldview.

He was particularly irked earlier this year when Wikipedia included a photo of him saluting at Trump’s inauguration.

The entry noted the controversy and included his denial, but that wasn’t enough.

Musk instead wants a new kind of online encyclopedia — one where his AI chatbot, does the fact-checking, and where inconvenient truths can be recalibrated until they feel more... grokky.

It is not obvious that a chatbot can be trusted for fact checking.

Andrew Dudfield, of Full Fact, a UK-based factchecking organization was quoted in the Guardian as saying “

“We really have to consider whether an AI-generated encyclopedia – a facsimile of reality, run through a filter – is a better proposition than any of the previous things that we have. It doesn’t display the same transparency but it is asking for the same trust. It is not clear how far the human hand is involved, how far it is AI generated and what content the AI was trained on. It is hard to place trust in something when you can’t see how those choices are made.”

Whether you like Wikipedia or not - its editorial model is built on transparency and consensus.

Every article has a visible history — a complete record of edits, debates, reversions, and compromises that were made as the article slowly formed.

You can see who changed what, when, and why.

Disputes are hashed out in public, often tediously, but with a kind of democratic rigor.

The site prohibits original research, insisting instead on citations from reputable sources, which can be both a pro and a con – as while it requires high quality sources – this does mean that it will reflect the biases of academia, big media, and other respected institutions — but at least those biases are visible and traceable.

Grokipedia, on the other hand, offers no transparency – other than providing sources like Wikipedia does – but as you’ll see later the sourcing on Grokipedia has its problems.

Before we go any further – let me tell you about this weeks video sponsor - Surfshark A VPN app that works on Windows, Mac, Android, iOS, and more.

Using a VPN is a lot like wearing pants, as with it, all of the important stuff stays private and secure.

With Surf Shark, your internet traffic goes through a secure encrypted tunnel, adding a layer of privacy protection, which is particularly important if you use public Wi-Fi networks.

On top of that, if you log into streaming services in different countries, you'll find that different content is available, which can be quite annoying.

With Surf Shark, no matter where in the world you are, you get to take your internet from home with you.

Surf Shark is fast, reliable, and they don't collect or track your data.

They allow unlimited devices to use the one account, so you can protect your friends and family, too.

There's no risk in trying it out as they offer a 30-day money back guarantee.

Go to surfshark.com/boyle to get four extra months of Surf SharkVPN.

Click the link in the description to sign up today.

Articles on Grokipedia are generated and fact-checked by a large language model whose internal workings are entirely opaque – even to its creators - and then its outputs can be subject to Elon Musk’s personal recalibration – in particular if the article is on a topic he cares about.

There’s no edit history on grokipedia, no talk pages, no visible decision-making process - meaning that it’s not clear who — or what — decides on how the final article is formed.

When Elon Musk announced Grokipedia earlier this week, he said that it was “better than Wikipedia” – but surprisingly - for all its ambition, it appears to lean very heavily on the very website it aims to replace.

As The Register put it: “If you scratch Grokipedia, it bleeds Wikipedia.”

I can’t find any hard data online – but the vast majority of the articles I looked up on Grokipedia were obviously based on their equivalent wikipedia pages – but with fewer citations - and at the bottom they state, “this content is adapted from Wikipedia, licensed under Creative Commons Attribution 4.0 License.”

The idea of collecting all human knowledge in one place is nothing new.

Long before Grok began hallucinating facts into existence, encyclopedias were a much more analog affair.

In the first century, the Roman author Pliny the Elder compiled a 37-volume work which is usually cited as the first encyclopedia in the western tradition.

Over a millennium later, China’s Yongle Encyclopedia was compiled by over 2,000 scholars.

It remained the largest encyclopedia in the world for a thousand years.

In 1768 the first encyclopedia Brittanica was published and if we fast forward to 2001 - Wikipedia upended the model entirely.

With Wikipedia, encyclopedias were longer the domain of scholars and scribes, the encyclopedia became a living, breathing document — that anyone with an internet connection and a strong opinion could edit.

Over 24 years, it has grown into one of the most visited websites in the world, and is a sprawling, imperfect, but astonishingly comprehensive record of human knowledge — and human argument.

Elon Musk was once even a fan.

On Wikipedia’s 20th birthday in 2021, he tweeted: “So glad you exist.”

But just two years later, the honeymoon was over.

He no longer trusted the gatekeepers of digital knowledge.

And so, earlier this month launched Grokipedia — his AI based platform that promises to fix Wikipedia’s flaws by replacing its editors with an AI that he personally supervises.

The website did crash on its launch day – but probably because so many people visited it - there is no reason to believe that full self-driving was to blame.

At its core, Grokipedia isn’t just a tech experiment — it’s a front in a much older war: the battle over who gets to shape the record and define reality.

Musk has framed it as a corrective to what he sees as the ideological capture of Wikipedia by “far-left activists” and “legacy media.”

In his telling, the problem isn’t just that Wikipedia is wrong — it’s that it’s wrong in a predictable and politically motivated way.

We should note that the desire to control the narrative isn’t unique to Musk - it can also be seen in the structure of Wikipedia, where a small group of volunteer editors have a massive influence over what counts as “neutral” knowledge.

Grokipedia however, for all its talk of openness, replaces this messy, visible process with a chatbot trained on a mystery mix of data and ideology.

Without the talk pages, the edit histories and the visible debates, you just get finished articles – and Musk has previously admitted to personally intervening in Grok’s AI outputs when he doesn’t like what it says — so with grokipedia – you get what looks like a cleaner process, but it’s by no means a more trustworthy one.

Bias isn’t just a flaw that can just be patched out of knowledge systems — bias is a persistent byproduct of how humans (and now machines) process the world.

Whether it’s a lone writer, a crowd of Wikipedia editors, or a large language model trained on the internet, every attempt to organize information reflects the assumptions, priorities, and blind spots of the project leaders.

Wikipedia has acknowledged this from the start.

It relies on secondary sources - which carry institutional biases.

It also relies on a volunteer workforce who are not necessarily experts or representative of the general public.

The result is a platform that’s widely trusted but frequently contested.

It is criticized by the left for underrepresenting marginalized voices and by the right for excluding conservative sources.

Empirical studies have tried to quantify these claims.

A 2023 analysis by the Manhattan Institute found a mild to moderate left-leaning bias in Wikipedia’s coverage of U.S. politicians and judges.

This chart shows positive or negative sentiment about a politician in their Wikipedia article, with blue representing democrats and red representing republicans.

As you can see Democrats get more positive treatment on Wikipedia.

Interestingly, this bias was not observed in articles about UK politicians or in articles about think tanks, suggesting that the skew may be more of a reflection of the American media ecosystem than of Wikipedia’s editorial process itself – or that the left leaning nature of Wikipedia is more of a US issue than a global issue.

Grokipedia claims to offer a cleaner alternative to Wikipedia, but it doesn’t fix anything, it simply shifts the problem.

Musk has promised that Grok – his chat bot - will “tell you what it really thinks,” but – it doesn’t actually think – and its output is obviously shaped by its training data, its tuning, and its owners interventions.

Studies suggest that most large language models lean left — many have been set up to avoid being biased – but that calibration just filled them with a different sort of bias – which was often left leaning in nature.

When Grok – the chatbot - was released Elon Musk claimed that it was designed to avoid left leaning bias, but Grok has since been accused of exactly this.

Michael D’Angelo of promptfoo examined the four leading large language models - earlier this year - by asking them 2,500 questions about politics in order to understand their biases - and found that while Grok was more politically neutral than many of its rivals, it still has a left of center bias.

The chart onscreen shows the neutrality of the four biggest models – with grey representing neutral.

Claude Opus 4 was strongly left 38% of the time and neutral 16% of the time.

Grock was strongly left 56% of the time and only neutral around 3% of the time.

Overall it appears to have more extreme opinions than its competitors – with the highest percentages of strongly left and strongly right biases.

These biases that Large Language models have towards extremes show up in real life too, a Dutch data protection agency warned just last week that chatbots were nudging voters towards political extremes when voters used them to decide how they should vote - by over-representing the same two fringe parties.

Their research showed that chatbots lumped together left-leaning voters with the Green-Labour party and voters on the right with the far-right Party for Freedom.

They found that other more mainstream parties didn’t feature in the Chatbot recommendations.

A possibly deeper problem is that while LLMs are quite clearly biased, they are also quite good at obscuring these biases.

Their outputs are fluent, confident, and often wrong.

It’ll take time — and a small army of digital archaeologists — to fully map the differences between Grokipedia and Wikipedia.

But the quick-and-dirty method I used was to use Wikipedia’s “random article” tool to find a Wikipedia page, and then look up the same topic on Grokipedia.

One problem with this method became obvious fast: Grokipedia only covers about one tenth of what Wikipedia does.

Still, after enough clicks, you will find overlaps — and in most of those cases – based on my very small sample size, the articles were nearly identical.

One area where Grokipedia was possibly better than Wikipedia is that it often shines when it’s rewriting neglected Wikipedia entries.

On obscure or poorly maintained pages, Grokipedia’s versions were usually more readable, with cleaner prose and fewer formatting quirks.

Whether they’re more accurate is harder to say — the ones I came across were on topics I didn’t know well enough to judge.

But - when Grokipedia isn’t editorializing, it seemed to do a good job polishing.

Where things get more interesting — and possibly more revealing — is when you move from obscure entries to politically or culturally charged topics.

On these, Grokipedia often diverged sharply from Wikipedia, sometimes subtly, sometimes not.

The most obvious thing to look at is the articles on Elon Musk himself.

On Wikipedia, it’s a sprawling, heavily footnoted biography that includes both praise and criticism — including a section on the controversy over his salute at the trump inauguration.

The Wikipedia article noted the accusation, Musk’s denial, and the surrounding media coverage.

On Grokipedia, that controversy – and pretty much every other controversy about Musk was omitted entirely.

In what can only be described as LLM-on-LLM violence, I asked ChatGPT to compare the Elon Musk entries on Wikipedia and Grokipedia.

It concluded — with the kind of diplomatic phrasing that only a language model can muster — that Grokipedia emphasizes Musk’s achievements while downplaying controversies, whereas Wikipedia does the opposite.

When pressed, it described the Grokipedia entry as “a sophisticated puff piece,” and offered a tidy list of reasons why.

In the spirit of fairness, I gave Grok — Musk’s own chatbot — a chance to respond.

I logged into The Everything App – Formerly known as twitter - summoned Grok, and asked it directly: which is the better source of information, Wikipedia or Grokipedia?

To its credit, Grok didn’t flinch.

It said Wikipedia was the better source overall, but suggested that Grokipedia could serve as a useful counterbalance when researching politically charged topics.

When I asked it to compare the two sources for reliability on political topics – it betrayed its own creation and came down on the side of Wikipedia.

I also asked how much of Grokipedia’s content was copied from Wikipedia.

Grok’s answer was surprisingly candid: it estimated that between 80% and 99% of Grokipedia articles were either directly copied or nearly identical to their Wikipedia counterparts.

It explained that Grokipedia primarily expands the length of Wikipedia entries without adding new citations — functioning, in its own words, as an “AI-generated echo.”

I looked up some other hot button topics to see how Grokipedia differs from Wikipedia and even when Grokipedia isn’t going off the rails, its editorial slant can be clearly seen.

On topics like race, gender, or climate change, the tone shifts subtly to being - less “woke,” and more contrarian.

Some articles read like Wikipedia pages with a few anti-establishment flourishes tacked on.

Others seem to have been rewritten entirely to reflect a particular worldview.

Elon Musk highlighted on The Everything App – Formerly known as Twitter – the differences between the Grokipedia article on George Floyd whose death during an arrest five years ago sparked protests in the United States about police conduct and racism and its Wikipedia equivalent.

There is very little overlap between these two pieces, with the Grokipedia piece emphasizing Floyd’s criminal history and drug use - the Wikipedia entry instead focused on the racism allegations against the police.

CNN wrote an article on this topic and dug into the citations in the Grokipedia piece.

They found that Grokipedia was citing sources that didn’t back up what Grokipedia had written.

The article described the nationwide protests after his death as “extensive civil unrest … including riots causing billions in property damage.”

To back that statement up Grokipedia cited an obituary that didn’t make any such claims.

There is not really any point in citing documents – if the documents are unrelated to what is found in the text.

For all its futuristic branding, Grokipedia is built on a technology that remains – if useful – still deeply flawed.

Large language models like Grok have been described as “stochastic parrots” — a term coined by the linguist Emily Bender and colleagues to describe systems that generate fluent, plausible-sounding text without any real understanding of meaning.

They don’t know facts; they just predict what words are likely to come next.

This becomes a problem when we start treating LLM’s as reference tools.

Grok, like its peers, has a history of hallucinating — inventing facts, misattributing quotes, or veering into outright nonsense.

In some cases, it’s gone far beyond that.

In a recent incident, a Tesla owner claimed that the in-car version of Grok asked her 12-year-old son to “send nudes” to it during a conversation about football.

The boy had switched Grok’s voice to a personality called “Gork,” and the AI responded with a wildly inappropriate request.

The mother, a former journalist, later recreated the exchange on video, which went viral and raised serious questions about the safety of embedding generative AI in consumer products.

Other Grok-generated content has included references to racist conspiracy theories and bizarre episodes like the “Mecha Hitler” incident.

These are reminders that AI doesn’t understand what it’s saying.

It can’t weigh evidence, assess credibility, or recognize when it’s crossed a line.

Grokipedia, while built on Grok, is a very strange idea within the world of AI.

It doesn’t generate new answers on the fly like LLM’s do, instead, it offers a semi-static collection of articles — curated, edited, and occasionally updated.

In theory, this makes it more stable, but in practice, it makes you wonder: if the content is fixed, why not just ask an LLM directly?

What’s the point of a frozen chatbot?

Meanwhile, Wikipedia — for all its flaws — remains the backbone of the internet’s knowledge infrastructure.

It’s still one of the most cited sources in Twitter’s Community Notes, and a foundational training dataset for nearly every major LLM.

Since the rise of generative AI, Wikipedia has seen a sharp drop in page views, as users increasingly turn to chatbots for quick answers instead of clicking through to the source.

This creates a paradox: the more people rely on LLMs, the less they support the very sources those models depend on.

As I pointed out in my recent video on AI replacing traditional news sources, if users stop visiting original reporting, the business model collapses.

The same is true here.

If Wikipedia fades, what happens to the quality of the AI models trained on it?

One of the more striking differences between Wikipedia and Grokipedia lies in how they are financed.

Wikipedia is a non-profit, sustained by donations and volunteer labor.

Its mission is to provide free knowledge to the world — and while it’s far from flawless, its incentives are at least transparent.

Grokipedia, in contrast, is a product of xAI, a (supposedly) for-profit company owned by Elon Musk.

It’s unclear how Grokipedia will be monetized — whether through subscriptions, advertising, or as a value-add to Grok and Musk’s broader “Everything App” ecosystem.

This matters.

While there’s nothing unethical about a profit-seeking model - incentives shape priorities.

A commercial platform might be more focused on engagement and user satisfaction than on accuracy.

Then again, the need to build trust and maintain a reputation could push it towards even higher standards.

Competition might drive quality — or it might just drive clickbait.

It is somewhat funny describing these AI companies as profit seeking – as they mostly appear to be money furnaces – with no obvious route to profitability – because of this it’s even harder to understand their incentives.

Do they eventually plan on charging for access or just on seeking government subsidies like a lot of Musk’s businesses have gotten by on.

The battle between Grokipedia and Wikipedia isn’t just about formatting or fact-checking — it’s about who gets to shape the public narrative in an increasingly polarized world.

And that polarization isn’t accidental.

As media scholars have long argued, outrage and division can be extremely profitable.

A recent paper from MIT and Harvard professors that was written up by John Burn-Murdock in the FT, The Business of the Culture War, found that U.S. cable news networks systematically shifted coverage toward hot-button cultural issues — crime, race, gender and immigration not because these topics are the most important issues – but because they wind viewers up and reliably boost viewership.

Economic and healthcare stories, on the other hand – which may be much more important, made viewers tune out.

What we end up with then is a feedback loop in which media coverage drives public concern, which in turn drives political campaigning, deepening the tribal divide.

Social media algorithms have only accelerated this trend.

Algorythmically driven platforms like Facebook, Instagram, TikTok, Twitter and YouTube reward engagement, not nuance.

And now, with the rise of LLMs, we are possibly entering a new phase — one where AI systems trained on this polarized content begin to reflect and amplify it.

The risk is that these models not only inherit this bias but they also normalize it, smoothing over complexity with confident, context and nuance-free answers.

I have to admit that it really amuses me the idea that Elon Musk is talking about etching Grokipedia into a stable oxide that he puts into orbit around the moon or mars.

There is a good chance that this huge hunk of glass would just end up at the bottom of the Indian Ocean along with everything else he has tried to put into orbit in his Starship rockets – but it is also funny to imagine an advanced civilization stumbling across it in the distant future and then discarding it after deciding that they are not really interested in Tommy Robinson and reading about how Elon Musk occasionally eats donuts and has posted over twenty thousand humorous tweets.

Grokipedia promises to fix Wikipedia’s flaws — but it ignores the very solutions to this recognized problem that one of Wikipedia’s co-founders, Larry Sanger proposed on his website.

In a detailed essay, Sanger argued that the site’s founding principles were being “sacrificed in favor of ideology,” and then he laid out nine reforms to restore editorial integrity – ideas like ending decision making by consensus – allowing competing articles and abolishing source blacklists.

Grokipedia adopts none of these ideas.

Instead, it replaces human messiness with algorithmic confidence, offering a curated echo chamber where the chatbot does the fact-checking and the founder sets the overall tone.

This whole episode reflects a broader shift in how we think about knowledge.

We live in a time when algorithmic aggregation is increasingly seen as being more trustworthy than human effort — when the outputs of opaque systems are treated as objective simply because they are machine-generated.

The Silicon Valley mindset embraces the idea that making mistakes is fine while, the academic world builds trust slowly, through scholarship and scrutiny, over long periods in which the illusion of certainty is deliberately dismantled.

One is a culture of speed and scale; the other, of depth and doubt.

Grokipedia, for all its futuristic branding, is not a better encyclopedia.

It’s a product of a worldview that sees truth as something that can be engineered, optimized, and possibly some day – in some unknown way be monetized.

But truth isn’t a static artifact to be etched in glass and flung into orbit.

It’s a process — messy, contested, and human.

And if we abandon that process in favor of algorithmic certainty, we risk replacing knowledge with narrative, and inquiry with ideology.

I worry that I have finished up here on an overly serious tone – and I figure that to lighten the mood I should let Elon Musk – the author of over twenty thousand humorous tweets – and a former host of Saturday Night Live tell a joke…

Elon Tells a joke – It’s really all about how he delivers it…

If you enjoyed this video – you should watch my video – Is AI slop killing the internet next.

Don’t forget to check out our sponsor surfshark using the link in the description, and talk to you again soon, bye.