Did I Predict 2024?
It’s been a while since my last post - sadly, this blog is not my top priority in life and starting a full-time job has made it more difficult to spend lots of time writing. I still plan to write more, but I can’t promise there won’t be future long pauses as well - I would rather not publish anything than publish a post I’m not satisfied with.
A year ago, I published a post called Predictions for 2024. I made 12 predictions for the upcoming year, aiming for 75% (or 9 of them) to come true. Before I reveal my predictions for 2025 (coming in a different post), let’s review what I thought would happen in 2024.
2024 Predictions:
Artificial Intelligence
Prediction: OpenAI will not lose their overall lead on having the best AI tools, to Gemini or Claude or Grok or anyone. This means that ChatGPT or a successor will remain the most used AI product, the one that the average person views as basically the same thing as just saying “AI”. No company will beat them on language benchmarks by significant margins without cheating. They might get clearly surpassed on image models or text-to-speech or something, but at the end of 2024, when you want to use AI to complete a language-based reasoning task, you’ll turn to an OpenAI model.
Result: Correct. There was increased competition in 2024, from Google, Anthropic, and others. Anthropic’s Claude in particular is especially impressive and I often find myself preferring its style. But no one surpassed OpenAI on language-based reasoning tasks, and no one is far ahead of them on benchmarks.
Prediction: Someone will implement an LLM into their smart home devices. I think Google is best positioned to do this, as one of the leaders in both smart home devices and AI. But Apple or Amazon could do it as well. There hasn’t really been any innovation in the smart speaker market since they were introduced (source: I can’t name a single new capability that my Google Home has gained in the last five years). LLMs provide a great way to add a bunch of new capabilities.
Result: Wrong. I don’t think this was a crazy prediction, and there were some attempts at it, but they didn’t get as far as I anticipated. (“Remarkable Alexa” is the closest maybe but wasn’t released, Google tried something but it didn’t really succeed.)
Prediction: “AI girlfriends/boyfriends” will not become significantly more relevant in 2024. The possibility of getting people addicted to chatting with an AI is there, and this will likely become more important later, but the current best models are not good enough to provide a legitimate partner experience. Eventually someone will figure it out, but it will mostly be a hypothetical in 2024.
Result: Correct. There were some people who got addicted to character.ai, and some people who use ChatGPT for this, but it wasn’t much more than in 2023.
Prediction: Deepfakes will continue to be talked about in the news, but despite the availability of high-quality image models to produce them, no deepfake will have many significant consequences in 2024. This is due to the fact that knowledge of deepfakes is relatively widespread, so they won’t be a surprise, and I don’t think any trusted source would use a deepfake or be fooled by one. People will probably try, but should be unsuccessful. There will probably be a few deepfakes that fool me, mostly of the “Pope wearing Balenciaga” flavor with no meaningful consequences.
Result: Correct. I think this was one of my less obvious predictions, but while deepfakes continue to be spread and AI images have taken over Facebook, there weren’t really examples of deepfakes that had significant implications, such as fooling a major news source or swaying the election results.
Politics
Prediction: Both Donald Trump and Joe Biden will be the presidential nominees for their respective parties in 2024. Trump won’t be removed from the ballot in a way that significantly affects his chances, and Biden won’t step down to allow a younger Democrat to take his place. There is no serious challenger to either - Nikki Haley comes closest, but polling at 11% (compared to Trump’s 61%) is not a recipe for a successful primary challenge. We’re heading towards a 2020 rematch, and it’s looking like it’s going to be close again, similar to 2020. I would slightly favor Biden because I think general economic perception seems to be improving, but I am not an expert on politics or economics, so take that with a grain of salt.
Result: Wrong. Biden certainly attempted to not step down, but eventually did. Pretty clearly wrong here. A lot of other people were also wrong about this, so I don’t feel quite as bad - I think predicting this at less than 50% would have sounded crazy in January, if you weren’t on the Biden campaign team.
Prediction: There will be no new wars in 2024 which provoke significant global response. The Russia-Ukraine conflict would have counted for 2022, and the Israel-Palestine conflict for 2023, but I think that was mostly just bad luck - there weren’t a lot of wars started in the previous years. China won’t invade Taiwan, North Korea won’t attack anyone, and India and Pakistan won’t escalate their conflict into a war. There might be some unrest somewhere of the sort that would make it into the NYT News Quiz, but nothing that will, for example, provoke large protests in the US.
Result: Correct. I don’t claim to have any special knowledge of global politics; my analysis here was mostly just looking at the number of conflicts started over the past decade and guessing that the Russia-Ukraine and Israel-Palestine wars were an outlier instead of signifying a large increase in violence.
Prediction: TikTok will not be banned in the US in 2024, but there will be increasing calls to ban TikTok based off of the misinformation on the platform and its influence over young Americans. Especially in the lead-up to the election, this will gain more prominence in the news. TikTok will be defended mostly by young, very left-leaning people, and Biden will not move to ban it because it would hurt him in the election. On TikTok itself, there will be essentially no popular arguments in favor of banning TikTok, or selling the US division to a US company, instead there will be memes about how dumb US politicians are and how great TikTok’s CEO is.
Result: Mostly correct. Proud of this prediction, even though it came pretty close to being wrong. There were increasing calls to ban TikTok (in the form of a bill forcing a sale/ban of the app), and it didn’t get banned in 2024. I can confirm as a TikTok user that it was full of anti-TikTok-ban memes. The only incorrect part was that I didn’t think Biden would try to ban it, but he did ensure that it wouldn’t get banned before the election, which had a similar effect.
Prediction: Despite the Supreme Court’s ruling to ban affirmative action, this will have approximately no impact on the racial demographics of the Ivy League’s incoming class, largely due to their new focus on personal essays (including essay prompts like “How will you contribute to the diversity of our school”) and recommendations instead of test scores. Legacy admissions will remain strong and no new major schools will ban it.
Result: Correct. A quick search led me to this article saying that the racial demographics of Yale and Princeton effectively didn’t change, and the percentage of Asian-American students actually declined slightly, which is unlikely to be a result of banning affirmative action. The article does mention that MIT had some more significant changes, though MIT is not part of the Ivy League.
Technology
Prediction: Twitter/X will remain the global “public square,” and people mostly will not leave. Threads will not surpass it in popularity in 2024. There will probably be more changes made that make the user experience worse, as there were in 2023 (unless you’re a fan of the blue check purchasers being constantly prioritized). But there was no mass migration away in 2023, and there also won’t be one in 2024.
Result: Mostly correct. Threads didn’t surpass Twitter in popularity, and neither did Bluesky, though both gained users in 2024 while Twitter lost users. Elon’s changes remain frustrating as someone who doesn’t want to see racist posts/replies on my timeline, but nothing else seems better yet, and a 10% decline isn’t a mass migration away.
Prediction: MrBeast will pass T-Series to become the most subscribed channel on YouTube. This one is mostly just an extrapolation of projections, but what MrBeast has done on YouTube - optimizing every second of his content to get as many views and subscribers as possible - is incredibly impressive. He’s also going to become the third-most-followed TikTok user, and could challenge the top two soon enough. Whoever is the second best person in the world at getting people to watch their content, MrBeast has to be at least twice as good as them.
Result: Correct. This happened, though it was just an extrapolation of trends and was probably more likely than 75% in reality. MrBeast not only passed T-Series but extended his lead by around 60 million, and now has more subscribers than the population of the US. He also became the third-most-followed TikTok user, and his success remains difficult to replicate - no other active YouTube creator who isn’t a kids channel has more than 20% of MrBeast’s subscribers.
Prediction: Self-driving cars will continue to expand their reach. Waymo might add LA and Austin to the cities they are operational in (currently SF and Phoenix), and Cruise will likely relaunch and continue to log miles, though they will be less aggressive about expansion than they were last year. However, self-driving cars will continue to not be available in any city without year-round generally good weather (so, not New York, Chicago, Portland, Seattle, etc), and no self-driving car will be available for inter-city trips (trips from, for example, SF to LA) by the end of 2024.
Result: Mostly correct. Waymo did add LA (and kind of Austin), while also announcing plans to expand to Atlanta and Miami. They didn’t add any of the cities with bad weather that I mentioned, and aren’t available for inter-city trips. However, Cruise didn’t relaunch, and in fact was shut down late in the year (which I’d say is being less aggressive about expansion!) I think there will soon be a year where self-driving cars really catch on nationwide, but it wasn’t 2024 and I doubt it will be 2025 either.
Prediction: The Apple Vision Pro will launch in 2024, and while it will not achieve mass adoption due to its high price point, it will be generally seen as a success. Apple will announce a lower-cost model to be released in the future. Apple does not often launch major new product lines, and when they do (iPhone, iPad, Apple Watch, AirPods) they often dominate the market. Meta will remain the market leader as $500 is a much more accessible price than $3500, but Apple will begin laying the groundwork for eventual mainstream appeal. Hopefully I get to try one this year because they look really cool.
Result: Wrong. The Vision Pro doesn’t feel like a successful product right now, it seems like it’s in limbo with a crazy high price point and few real uses. I am still optimistic that it has a positive long-term future, but it definitely didn’t happen in 2024. It looks like they are planning a lower-cost version for 2026, so it doesn’t seem like they’ve given up, but 2024 was not the year of the Vision Pro. I think this was probably my worst prediction of the year, and I was pretty biased by personally thinking the Vision Pro seemed like a cool piece of technology. (I still think this, after trying the demo, but admittedly I’ve only used it for half an hour.)
Final Results
Correct: 6/12
Mostly correct: 3/12
Wrong: 3/12
Overall, I think I’m pretty happy with these results. I had 3 clearly wrong predictions, but if you count “mostly correct” as being, like, 80% correct, that gives me a final score of 8.4/12 or 70% correct - pretty close to the 75% accuracy rate I was going for.
Why did I go for a 75% accuracy rate, and not 100%? Am I just trying to make myself look better? Surely if I were really a good predictor, I’d be able to aim for 100% accuracy! However, I believe that’s based on a flawed concept of what it means to make a prediction.
A common issue that people have when they think about making predictions and forecasts for the future is that they think of it as being “all-or-nothing” - you should predict everything with 100% confidence, and then score yourself based on how many you got right. I think this is not a great approach for a couple reasons.
Firstly, no one is ever 100% confident about predictions. Being 100% confident means that you believe there is zero chance that you’re going to be wrong, which is almost never true about anything that would be remotely interesting to predict. Consider sports as an example: any given game probably has a favorite and underdog, but it would be ridiculous to claim that the favorite is literally 100% likely to win. Even the Harlem Globetrotters don’t have a 100% chance to win.
Secondly, the impressiveness of predictions isn’t just based on how often you’re right about them, but also how hard the thing was to predict in the first place. You could easily go 366/366 on predictions in 2024 by predicting “The sun will rise today” every day of the year, but no one would say that was impressive. On the other hand, if you can predict with 60% accuracy whether the stock market will go up or down tomorrow, that would be extremely impressive and you would quickly become very wealthy, even though you’re only doing 10% better than flipping a coin every day.
And finally, most things that are interesting to predict are fairly uncertain. It would be great if I could come up with a list of 12 interesting predictions that I thought were 99% likely, while the general public didn’t, but that’s incredibly hard to do. (If you think you can do it, please send your predictions to me!) So instead, I ended up going with things that I thought were 75% likely, but also things that I could easily imagine people thinking were more or less likely than 75%.
Importantly, this means that going OVER 75% would also have been bad. If I’d gotten all 12 of my predictions right, then that might sound good - 100% accuracy - but it would mean I was underconfident, or predicting things that were too easy. That’s an easier problem to solve, of course, but it’s important to be able to distinguish between 75% and 95%, for example.
I could write about how annoying it is when people treat predictions as needing to be 100% certain for a while (see: Problem 1 in my post about presidential debates), but I’ll stop here and save it for a possible future blog post. But if you disagree I would again love to see your 100% confident predictions.
I’ll try to release my 2025 predictions soon. I plan to have more than I did in 2024, and probably I won’t make them all at 75%, either. If you’ve got any ideas for topics I could cover with those predictions, I would be happy to hear them. See you soon!