It’s always terrifying to write down your AI predictions, for they rarely age well. And so I’m not writing this post because I have strongly-held views of the future. In fact, I have not felt as uncertain about the future for a long, long time. I’m writing this because, as a purely personal matter, I wish to take a snapshot of myself at this moment in time. And to keep myself honest, I’ll put it up on my pseudonymous blog. As such, to save myself the embarrassment, please do not share this, even (especially) if you know my real name. These are not carefully thought out numbers. And they will likely not be the same by the next time we meet.
Serious pullback in AI (beginning of “winter”) in 2025: 15% Humanity is extinct before 2030: 0.01% AI directly causes catastrophe where 10,000+ people die before 2030: 10% AI can do {95% of non-physical labor weighted by 2024 US GDP} by 2030: 1% AI solves a millennium prize problem by 2030: 50% Median year that AI-generated social media content overtakes human-generated content: 2027 Median year the average American spends over an hour a day with an AI companion: 2028 Median year most US major metropolitan areas have self driving cars: 2029
The claims below are that each topic becomes one of the “icons” of AI in 2025, not that they happen or that progress is made. For calibration, I would probably say that 2024 (so far) is the year of open source, investment, and post-training. Keep in mind though that the year is not yet over.
2025 is the year of agents: 55% 2025 is the year of multimodal: 75% 2025 is the year of reasoning: 65% 2025 is the year of test time compute: 35% 2025 is the year of 1T+ parameter models: 15% 2025 is the year of small models: 35% 2025 is the year of AI products: 90% 2025 is the year of open source: 10% 2025 is the year of sub-quadratic attention: 5% 2025 is the year of interpretability: 5% 2025 is the year of robotics: 10% 2025 is the year of pre-training: 20% 2025 is the year of post-training: 40% 2025 is the year we get rid of this garbage pre/post naming convention: 50% 2025 is the year of RL: 70% 2025 is the year of winter: 15% [repeated from above] 2025 is the year of government: 10% 2025 is the year of war: 5% 2025 is the year of situational awareness: 1%
This era of scaling laws has ended: cannot see through the mist Humanity is extinct before 2050: cannot see through the mist Median year that AI will solve cancer: cannot see through the mist Median year that non-physical labor mostly becomes reviewing AI outputs: cannot see through the mist First year US GDP doubles (2023 growth rate was 2.5%): cannot see through the mist Our descendants will spread across the stars: cannot see through the mist
For those of you who have talked much to me about AI, you will recognize that one of the numbers is an enormous aberration from my previous beliefs. It’s that human extinction by 2030 is sitting at maybe 0.01% right now. I am not even remotely close to confident in this number. But my current point estimate that varies wildly by the day (in log space no less) is that 1/10000 worlds rolled out from this moment may result in human extinction by 2030. For the first time in my life, I acknowledge that we are perhaps only one jump away from AGI. Of course, we might actually be far, far away. Once again, one can never see all that clearly through the mist. I still do not believe that scaling up next token prediction will lead us to AGI as many proclaim to believe (though only a select few actually believe). Given that I assign a 15% probability of the onset of AI winter in 2025 and a 0.01% probability to extinction before 2030, it’s also kind of hard to classify me as “pilled”. But I am semi-pilled, or as you might say, I have begun to feel the AGI.
For I have glimpsed the future. Terawatt clusters costing more than the GDP of a small country will soon pouring out concrete foundations. And once built, they will not be torn down even if the markets tremble. Not even if the investors turn the Transformer 8 into curse words. They may not build new ones, but once built, they are here to stay — at least, until the hardware depreciates in a few years. And so for the first time in human history, an idea can go from the scrap of a napkin to using more compute than all the world had access to not long ago, in a matter of weeks or months, perhaps even days. In 2015, Andrej Karpathy wrote The Unreasonable Effectiveness of Recurrent Neural Networks. It was a simple next-token predictor trained on probably a million times less compute than the LLMs of today. I read it and thought it was a cute trick — nothing more. Karpathy himself did not really believe it, as he declared a decade later, having not pursued that line of research in the time since. Once again, this is the bitter lesson at work. Not that you can scale up any idea. No. But that when you find the right idea, the right thing to scale — you can shake the world before you.
For calibration, my 2024 predictions went reasonably OK all told? Among some of the best for people outside the frontier labs, though the bar is really on the floor there for both informational and competence reasons. I did well enough that I could see it on a few people’s faces that they thought I was getting serious leaks. Ironically, I think I mostly outperformed the leakers simply because there is so much poison out there that leak dependents get duped far more often than they hit gold.
Things I got right:
– Synthetic data – Inference time compute – Competitive coding
This section is self explanatory given my choices, which I will not describe or else reveal my actual name. Given my name, you can find out why I was dead on quite easily.
Things I got wrong:
– General reasoning progress went faster than I expected. This is quite ironic given my dead on guesses. The crux is that while I correctly guessed the most important research directions, I incorrectly guessed how many other people also thought this was the most important direction. I think this actually happens more than people expect.
– I did not expect GPT-4 models to get commoditized. This was a huge question mark in my head and I leaned towards it not being likely to happen. Obviously, many people confidently predicted this, and they predicted it right. I was not one of them. Ironically, as much as people like to rag on VCs, they got this one dead on. It turns out there is no magic for this class of models — just flops.
– Multimodal did not take off as much as I expected. I still(?) think it’s directionally correct, so I’m shifting my date to 2025 with the updated prediction.
– The importance of electricity.
Note the date. I saw this when it first came out, but it took until mid 2024 before I had any idea of what roon was talking about here. Ironically, a year later, I still think the tweet is wrong, though at least I think(?) I understand what he means now.
Heart of the Matter: Single transferable vote is not monotonic. Writer’s note: this was written before November.
The current flavor-of-the-month in my social circles is bitching about how democracy, tech giants, and finance companies are fundamentally flawed but how we could solve all our problems if we just switched to mandatory therapy, blockchain and ranked choice voting.
Ranked choice voting works in the following way. People rank their candidates in preference ordering, instead of simply choosing a single person to vote for. If the candidate that you support most comes in last place, they are eliminated and your vote instead goes to the person you ranked in second. This process continues until there is only one candidate left who gets 100% of the vote. Conservatives (who tend not to like ranked-choice voting) often accuse liberals (who tend to like ranked-choice voting) of wanting to turn the US into a totalitarian-communist state, and though they definitely aren’t right all the time, they might actually be right here. At the conclusion of ranked-choice voting, the winner, just like in Russia or North Korea, gets 100% of all the votes!
Let’s run through an example so you know I’m not lying to you. Think back to the 2020 democratic primary and imagine there are 100 voters choosing between Andrew Yang (A), Bernie Sanders (B), and Cory Booker (C) with the following preference orders (data is manufactured, just like the real thing).
A > B > C (47)
B > C > A (27)
C > A > B (26)
Using a standard election procedure (plurality), Andrew Yang would win because 47% of people like him best! Hurrah for Asian representation in politics! Now, what happens if we use ranked choice voting instead? Well, first Cory gets eliminated since he is the least popular of the three politicians, and his supporters get distributed to their second choice which is Andrew Yang.
Hearing I’m a lot of people’s 2nd choice. This is actually a huge step toward becoming a lot of people’s 1st choice.
Then, Bernie gets eliminated, so all his votes go to Andrew who wins with a whopping 100% of the vote! Hurrah! Yellow Power! OK, that’s a great outcome.
But let’s say that Andrew, since he has the Asian work ethic, goes above and beyond, and campaigns even harder before the election. He campaigns so hard that he convinces some of Bernie Sander’s supporters that he’s actually the right man for the job. Let’s say that two of those people go from ranking Andrew at the bottom of their list to ranking him at the top (changing nothing else in the process). Thus, they go from B > C > A to A > B > C like the rest of the Yang Gang. Now, the vote looks like:
A > B > C (49)
B > C > A (25)
C > A > B (26)
Notice that everyone in this preference ordering likes Andrew Yang at least as much as in our original voting profile. Let’s do ranked choice voting again. This time, Bernie loses the first round because he’s the least popular and all his votes are reallocated. Thus, the final ranking is:
A > C (49)
C > A (51)
Which means that …. WHAT?!?!?!?! Andrew gets eliminated and Cory gets 100% of the vote?
I find this outcome especially ironic since ranked choice voting does WORSE than the commonly critiqued plurality voting (where the person with the most number of votes wins) and really want to emphasize just how ridiculous this outcome is. Andrew Yang would have won the election if he stayed at home, but he instead went out and turned some of his haters into his greatest fans. This cost him the election. Any voting scheme where this is possible is so rigged it needs to be thrown into the trash bin. Actually, on that note, any voting scheme that doesn’t elect Andrew Yang needs to be thrown in to the trash bin.
So don’t believe the lies that ranked-choice voters tell you. They just never wanted a yellow man in the White House.
Footnote: This article is clearly tongue-in-cheek, but the problem with ranked choice voting is a real thing. The crux was that in this setup, in a head-to-head election the outcome was a cycle: Andrew > Bernie > Cory > Andrew. Thus, the order in which the candidates get eliminated determines who wins the election. By wooing some extra voters, Andrew made it such that Bernie, not Cory was eliminated in the first round, which cost him the election.
Before anyone protests, I should clarify that this is not Arrow’s Impossibility Theorem! That theorem isn’t even as strong as people say it is, but this example is not it! There are many election outcomes that have the property that getting more votes cannot cost you the election. One such election scheme is the commonly used plurality vote. Arrow’s impossibility theorem merely says that all election schemes are messed up in some (often very, very subtle way), but ranked choice voting is messed up in a pretty blatant way: convincing MORE people to like you can make you LESS likely to get elected.
Sensitive topic that I admittedly don’t know much about. That being said, it’s my private, pseudonymous blog, read by 10 people tops, so here goes.
Here’s a question I have been pondering: in the long run, how much does race matter? In the present day, it is will known that income, education, voting patterns, crime, prison rates, etc. all vary tremendously based on race, and the past year has shown us that racial tensions are far from a thing of the past. But let me make a bold prediction that I will likely never live to see: in 250 years, assuming America (as it exists today) does not collapse, race will cease to be an important issue in society. The reason? Interracial marriage. If there is sufficient mixing, there will be no discrimination against Black people because there will be no Black people. There will be no affirmative action, because there will be no races to affirm.
Because of interracial marriage being a relatively new phenomenon, history can offer us only limited guidance. Nevertheless, consider the largest ethnic group on Earth today: the Han Chinese. Are Han Chinese people one ethnic group? Well yes, but it wasn’t always this way — the original distinctions have been long lost to time. Everyone is part Mongol, part Huaxia, part whatever. And this happened despite a strong norm against interracial marriage because time is more powerful than them all. People will always buck against rigid norms and over a thousand years, the diffusion slowly adds up.
For the Han Chinese, it took that long because they didn’t have interracial marriage really. In America, we won’t have a thousand years, but I doubt we’ll need it either. Interracial marriage was legalized in 1967 with the aptly named Supreme Court decision of Loving v. Virginia. Over half a century later, one in seven new marriages are interracial. Anecdotally, [redacted] once noted to me that while he felt out of place growing up because he was half white and half Asian because he did not see many people who looked like him, no one in the next generation would be able to empathize with his struggle. He’s right. Right now, only around 3% of Americans are mixed race, but in 2015 (latest year I could find data), 15% of newborns were mixed and going forward the number will likely only rise. When every single American is X +- 1% Black, whatever the equilibrium number ends up being, you can’t have a conversation about race … since everyone is the same race.
Retort #1: People will find other racial or ethnic groups to oppress, you might retort. After all, Uighurs are well known second-class citizens in China. But in fact, this is the entire point. America is presently headed down a course where everyone mixes with everyone else. The melting pot that is so often promised in elementary school history textbooks has not yet come, but if we continue along our current path, it very well might!
Retort #2: People will just manufacture racial groups to justify racial discrimination! Perhaps, but this will become increasingly harder as it becomes more and more impossible for everyday people to distinguish your new definition of “race” via visual features. Also, these new racial categories are much less likely to have the bite of history. Much of racism can trace its history to the institution of slavery. Take that away, and much of the power leaves the punch.
Retort #3: What about immigrants? This, admittedly, is the tricky part of the equation. Can America keep being a net importer forever? It’s 2021, so never say never, but it feels hard over the very long run. Also, similar to the response to retort #2, if the only trace of racism is anti-immigrant sentiment, it’s significantly less of an issue (even if nonzero) than current racial tensions. And again, post mixing, I think it will again be very, very, very hard to tell who you should be discriminating against visually, even for immigrants, which should muzzle much of the bite.
On a broader note, while I think that race may disappear, class will almost certainly not. There always has, and likely always will be an us-them relationship that underlies much of human interactions. It may perhaps grow even stronger, given my conjecture that it will replace race as the primary marker for discrimination. Perhaps immigrant status, perhaps something else entirely, like whether you own the right NFT. Status is an elder god that will not be slain so easily.
Perhaps that’s why interracial marriage was (is?) so sensitive. It’s the only thing that actually matters in the long run if you care about preserving your race. While there will always remain racial supremacists, who like the Amish, restrict themselves to their own communities, they will become increasingly irrelevant and eventually inbreed themselves into extinction. Of course, other considerations make this process significantly less clean than I described above (e.g. the one drop rule, marriage not being IID, etc.) but I still suspect that in the long run, demography will prove to be stronger than them all, for it alone has time on its side.
Silicon Valley is enamored with “fuck you money”, a concept I have heard explained as having enough money to be able to say “fuck you” to anyone you pleased.
I must confess that I’ve never really thought much of the idea. My tongue is already too sharp to be loosened further: instances where I have withheld an insult from one who I felt deserved it have alas been rarer than I might wish. Certainly none have occurred because I lacked enough money. In fact, I suspect that one of the blessings of stumbling into a large fortune might be holding my tongue back more. Few care if [memorymancer] delivers a sharp rebuke, but add a billion dollars to the equation, and it might hit the news headlines by morning.
Besides, for all Silicon Valley’s tongue wagging on stoicism, they remain surprisingly silent on Diogenes, one of their supposed grandfathers. Here was a man who called Plato himself a bastard defiling the legacy of Socrates, who would urinate on this who who insulted him. Diogenes was known for regularly shitting and masturbating in public. One of his favorite hobbies was wandering the streets with a lantern in broad daylight asking if there were any real men in town.
If you want an example of saying fuck you to anyone you pleased, look no further Diogenes. Once upon a time, Alexander the Great stood before him and asked Diogenes for his heart’s desire. Here stood a man undefeated in battle across fifteen years of warfare spanning half of the known world, who burned down one of the greatest cities in antiquity, who named a vanquished city after his horse. He could make you the richest man in 1000 miles or annihilate you, your entire family, and any unlucky bystanders at the wag of his finger.
Diogenes asked Alexander to stand aside and stop blocking the sunlight.
Truth be told, getting fuck you money is not really one of my goals in life. I say too many fuck you’s as is, and besides, living without any possessions seems like it would be rather unpleasant.
Take it slow. Richard Nixon and Lyndon B. Johnson both proposed to their future wives on the first date. They were both turned down. Richard Nixon, the consummate ladies man, kept his dreams alive by offering to chauffeur his beloved to dates with other men. And before you call either of these men pathetic, remember both that they eventually won the girl over and that they, unlike you, once scaled the heights of power. Evil they may be, pathetic they are not.
Don’t marry family. We’re not even talking royalty here. Marrying your cousin isn’t just for people in Alabama. You’d be in the company of Edgar Allan Poe, H.G. Wells, Charles Darwin, and so many more less famous people. And before you argue that times were different back back then, add Albert Einstein, Rudy Giuliani, Saddam Hussein, and the sitting president of Iran to that list.
Don’t date your mentors. In a charming love story for the ages, the sitting president of France married his high school sweetheart teacher. When they started courting, he was fifteen and she was a married woman twenty-five years his senior with three children, the oldest of whom had just turned seventeen. I suppose, given that the French age of consent was recently raised to fifteen, that maybe it’s just a different sort of culture than what I am used to. Also, the couple who won the Nobel Prize in Economics in 2019 met in graduate school, where he was her advisor.
High school relationships can’t last. I know six couples who were high school sweethearts (and not Macron-style sweethearts), each of whom spent their college years long distance. And by know, I mean that I talk each of them on a semi-regular basis. I went to a somewhat conservative Christian high school, so if you allow me to count everyone I “know” but don’t talk to regularly, I “know” quite a few more too. Looking back, I am slightly bemused that this remains one of the facts that freshman me was most wrong about. I was repeatedly told that a relationship could not survive long distance in college and the fool that I was, I believed my elders unquestioningly.
Don’t sh** where you eat. My impression is the most common way of meeting someone after graduating is work? When I was working at Asana, I personally knew of three separate couples who were actively dating in a company that was no larger than 150 or so. My grandparents met at work, along with around a fifth of married couples in the US.
For every rule I see (and I am explicitly not talking about moral rules because there are obviously tons of examples of evil people “succeeding.” I am talking explicitly about rules for success), there is a blaringly obvious example of someone who violated the rule and then went on to become wildly successful. I’m not even sure that these “rules” are even correlated with success. More and more, I suspect that they are geared for people who have a maxmin perspective on life—trying to maximize the worst-case scenario instead of the best or even average-case scenario.
Just in my personal life, I know someone who married a person they just met, a man who meticulously planned a “serendipitous” series of encounters with a girl he went on to date but never mention that absolutely nothing about them meeting was accidental, a man who married a girl because the first time he saw her, she beat someone up in a bar fight, and a whole host of other stories that I should not write down even semi-anonymously. Love has the most adventurous tales, but I think you can find analogous ones in whatever domain you desire.
There are no rules for success. There is only success.
Recently, I was surprised to learn that a JPG was sold for 69 million dollars. Well actually, the JPG is free for everyone to download, but the 69M buys you a digital certificate of ownership on the blockchain [1].
After witnessing the madness of last year, there is a part of me that believes the world really is this wacky,. But the wolf in me can’t help but notice an alternative possibility. Imagine you are an NFT titan (e.g. early stakeholder in one of the new NFT chains, builder/artist, classic crypto con artist etc.) who has a ton to gain from NFT’s going mainstream. Arrange a side deal with the artist Beeple (whose work is sufficiently bad that he should be fined for pollution every time he publicly displays his art), using him as the shell for the transaction. The NFT titan puts up the money to buy his JPG and receives back all the “profits” from the auction. Beeple gets paid a nice fee for his participation in this pump, and you lose the commission to Christie’s, which amounts to around 1-2%. All told, the operation costs you a few million. You get two things in return.
1) The PR you get from this far exceeds anything you could get by buying ads directly to pump NFT’s. After you make your move, people will start investing with real money in hopes of riding the mania. Some of them will actually succeed.
2) Resell value for the “art”. As [redacted] mentioned, you can resell the NFT at a 50% “discount” and still 10x your money.
My (very loose) understanding is that deals like this are not only legal; they are commonplace in the art world. For example, auction houses are allowed to bid on their own items to drive up the price. Also, high end art is one of the best vehicles for money laundering [2].
Goods with unclear value (e.g. art) are always vulnerable to this sort of scheme because their price is solely determined by supply and demand, and you can always manufacture demand. Is that manipulation? Maybe it’s just creating value? You manufacturing a fake transaction at 69M creates real demand from real people using real money. Of course, this is only because they don’t know that your transaction is fake. But if they never know …., what’s the difference from the economic point of view? It’s like a hyperbolic extension of Keynesian economics, except a little more dishonest? This stuff still seems crazy to me, but who knows. I have been mistaken in many such instances in the past. Maybe I’m the foolish one for not figuring out how to make money off of this stuff.
[1]: I should probably say a blockchain actually. I believe there are multiple competing NFT blockchains mostly all built as smart contracts on top of the big Ethereum blockchain. I legitimately wonder what will happen when people begin sell “one-of-a-kind” digital certificates for their art on several different NFT platforms simultaneously.
[2]: High end art provides plausible deniability for practically any transaction. Let’s say you need to get me 10M without the cops being too suspicious. I’m going to buy some cheap a** piece of art and start pumping it. When it comes up for auction, you’ll bid 10M for it. I get my money. You pay your debt and get a leftover piece-of-garbage which even might fetch a handsome price on the resale market.
Who does the Internet belong to? The early users were cyberpunks. While they correctly predicted that the Web would come to dominate the world, they mistakenly assumed that everyone else would also come to cherish the values they hold so dear: freedom of speech, privacy, and anonymity. As the general public moved online, they brought their values too. Now all the major platforms have censorship and surveillance, and either limit or ban anonymity altogether. Not everyone wants to be a cyberpunk.
The most iconic individuals of an era tell you where the power lies. In the Gilded Age, the power lay with the money.
Contrary to popular opinion, history does not always work this way. The greatest icons during the Civil War—Abraham Lincoln, Ulysses S. Grant or Harriet Tubman—were often quite poor. At the turn of the 20th century, people such as Theodore Roosevelt, Vladimir Lenin and Mahatma Gandhi all had massive influence without being particularly wealthy.
It is only the era in between, the Gilded Age, where money drowns out everything else. The presidents of this era—James Garfield, Chester Arthur, Grover Cleveland, and Ben Harrison—are forgotten to all but the history buffs. They held no real power and accomplished nothing of substance. Those familiar with leftist canon may recognize socialist and unionists revolutionaries such as Eugene Debs or William Jennings Bryan, but they lost and their names are but dust compared to those who they fought against. We know the names of the men with power. We feel their influence even today. Carnegie. Morgan. Rockefeller. Stanford. Vanderbilt.
They were the first billionaires. According to Howard Zinn’s A People’s History of the United States, the common laborer on one of their railroads could expect to earn just over $100/year. If that person could save everything they earned without deducting what’s necessary for food, shelter, healthcare, and everything else, it would still take them 10 million years to get to Rockefeller’s wealth.
In the modern-day, the icons of our times are once again the billionaires. Bill Gates. Mark Zuckerberg. Warren Buffet. Elon Musk. Donald Trump. Jeff Bezos’s wealth recently topped 200 billion. Minimum wage varies, but it’s roughly $10/hr or 20K/year, just enough to get the 10 million number all over again. Maybe the ending of Parasite was actually an under exaggeration.
The Gilded Age ended in world war, followed by a pandemic, followed by a brief decade of prosperity, followed by depression into another world war. The modern timeline already has a pandemic. I wonder what happens next.
“A single death is a tragedy; a million deaths is a statistic.”
Joseph Stalin
You can feel something about one death, but a million? Just trying to imagine so many people dying makes me numb, almost helpless in not knowing how exactly to feel. I can’t just multiply the grief over one person’s death by a million, since that overwhelm me. Perhaps a better way is to picture a person being shot, their blood and guts and brain splashing onto the floor in front of me. The looks on their faces: dejection, hopelessness, desperation, sometimes defiance. See it happening over and over and over again, every minute a new execution, every waking moment, for three years of my life. That’s a million deaths. What else could I feel but numbness? The human heart wasn’t meant to hold such darkness.
So in a sense, Stalin was right, at least in that a million deaths is not a tragedy: it’s something else. It’s heroic if you are on the winning side (Manifest Destiny, Allies in WW2). It’s genocide if you lose (Nazi Germany, Soviet Russia). But, unfortunately for Stalin, it’s only reduced to an empty statistic if you are in a region with low GDP (Belgian Congo, Rwanda). Russia, alas, was and is too powerful for Stalin to hide behind mere numbers. King Leopold, Robert Kajuga, and all the others whose names I’ll never know: those are the people who got away with their murders recorded as just some statistic in the appendix of some history book.
If nothing else, our world is … complicated. The 20th century saw our brightest minds pursuing, but never finding, sweeping general theories of mathematics, physics, history, or government. Stories of heroes or villains (depends on the storyteller) trying to do good but failing are so common that modern ethicists still debate on the relative importance of the intention versus the result.
If you had a magic wand, what are some things you could do to make the world unequivocally a better place? This question is actually much harder than it appears at first glance. Here is an incomplete list of some things that could go wrong.
Unintended consequences
People drive more recklessly when wearing a seatbelt or a helmet.
A bounty on snake skins intended to eradicate the population led to people farming the animals for profit.
More generally, in the presence of “optimizers”, changing even the smallest thing might upset an existing equilibrium and lead to major consequences in supposedly “unrelated” areas.
Inequality
Is your change a Pareto improvement (i.e. is *everyone* better off afterwards?)
Even if so, does it exacerbate existing inequalities / power structures or does it create an unfair division of gains?
Beliefs
Are people actually happier after your change? Remember that happiness is fickle and strange and that a person’s thoughts and moods shift faster than the wind.
What are people of various beliefs and religions going to think about your change? Remember that ~85% of the world is currently religious, though beliefs vary wildly.
Does your plan violate anything that some people believe as a fundamental right (e.g. freedom, liberty, property, privacy)
What would your mother/father/sibling/friends/teachers think?
What will future civilizations think of your actions?
Survival
Is your plan likely to extend or reduce the expected lifespan of humanity (or of other life on Earth)?
Everything you do at Google is logged. Google knows (and shows you) every building and cafeteria where you badge into. They even gamify it by giving you virtual “rewards” for each new place you visit. If you want to access Google infrastructure from your personal phone, you need to install an app that gives them complete access, including the ability to do a remote wipe. I don’t really consider myself paranoid about security (I often don’t even lock my front door), but I never ended up installing this app.
There are some interesting exceptions. Ironically, unless otherwise marked, Google deletes your chat messages with other coworkers in 24 hours (30 days if you want to store them) and your emails in 2 years for “storage space” reasons. But I think they are actually deleted, so that Google can honestly say they have nothing when subpoenaed.
Now, there are good reasons for all of this, and Google certainly isn’t alone in keeping tabs on their employees, but …. still. I can’t help but feel a liiiiiiittle creeped out. The pay is excellent, the perks quite nice, but they also know everything about you. For a normal user of Google’s products, the tradeoff is mild. You give up a lot, but it doesn’t know everything or even close to everything. For an employee though, it’s basically 1984 in bright colors.
Musings of a Memorymancer 
