That, plus he's quick enough to come up with good follow-up questions on the spot. It's so frustrating listening to interviews where the interviewer simply glosses over interesting/controversial statements because they either don't care, or don't know enough to identify a statement as controversial.
In contrast, Dwarkesh is incredible at this. 9/10 times when I'm confused about a statement that a guest makes on his show he will immediately follow up by asking for clarification or pushing back. It's so refreshing.
Not the person you're responding to, but I think the salary paid to the researchers / research-engineers at all the major labs very much counts as eye-watering.
What happened at meta is ludicrous, but labs are clearly willing to pay top-dollar for actual research talent, presumably because they feel like it's still a bottleneck.
Having the experience to build a frontier model is still a scare commodity, hence the salaries, but to advance AI you need new ideas and architectures which isn't what you are buying there.
A human-level AI wouldn't help unless it also had the experience of these LLM whisperers, so how would it gain that knowledge (not in the training data)? Maybe a human would train it? Couldn't the human train another developer if that really was the bottleneck?
People like Sholto Douglas have said that the actual bottleneck for development speed is compute, not people.
All of those tools are garbage. There is no reliable automated way to detect ai generated text. In 2023 OpenAI had a tool for this as well and they eventually took it down because it wasn't accurate enough. The major AI labs are probably best positioned to make such a tool work. If even they can't, then some random company with access to a fraction of a data and a fraction of the compute almost certainly also cannot.
Agree those tools are unreliable. Unless you have a massive amount of ML models trained on individuals' writing[0], the best you can do is vibe-checking[1].
I do this too. It's great. The term I've seen used to describe this is 'Immersion Reading'. It seems to be quite a popular way for neurodivergent people to get into reading.
They also tried to heal the damage, to partial avail. Besides, it's science: you need to test your hypotheses empirically. Also, to draw attention to the issue among researchers, performing a study and sharing your results is possibly the best way.
Yeah I mean I get that, but surely we have research like this already. "Garbage in, garbage out" is basically the catchphrase of the entire ml field. I guess the contribution here is that "brainrot"-like text is garbage which, even though it seems obvious, does warrant scientific investigation. But then that's what the paper should focus on. Not that "LLMs can get 'brain rot'".
I guess I don't actually have an issue with this research paper existing, but I do have an issue with its clickbait-y title that gets it a bunch of attention, even though the actual research is really not that interesting.
The idea of brain rot is that if you take a good brain and give it bad data it becomes bad. Obviously if you give a baby (blank brain) bad data it will become bad. This is about the rot, though.
And insisting on building their own mediocre ui helps them achieve that how exactly?
Maybe I'm just super out of touch on how people use their cars, but my cars infotainment system has not made its manufacturer any additional money as far as I can tell.
Once you have the data the money can be figured out later.
You can figure out how people use the car and what features matter to them and upsell or upcharge for those features in the next line.
You can cross-advertise (oh you listen to music when driving here's Spotify deal through us and behind-the-scenes we get a cut for lead generation to Spotify).
You can use the information to defend in lawsuits. Oh our car is faulty leading to accidents? But all these people were fiddling with the unit before crashing.
Also if you control the platform you can sell integration spots to companies. I know my old BMW had a specific separate path to connect Spotify on your phone to car, no other audio app.
There's surely other ways I haven't thought of. The investment pays off later IF you get the data, but CarPlay and Android Auto have really mucked that gambit up for the car makers
Surprisingly, it isn't. You can change the language in your google account and it will take that into account for what to translate and into what language, but you can't turn it off completely.
I don't know who thought this was a good user experience, because it's one of the most frustrating features I've ever had to deal with. I'm german, but almost all of the things I watch are in english. So usually I will just ignore recommendations with german titles. Except I can't do that anymore, because there's no guarantee that youtube didn't randomly decide to translate the title of an english video into german. And recently, they've added auto-translated audio, which is even worse, because now I'm opening an english video and a terrible robotic german voice is talking to me and I manually need to switch to the original source.
It's also not consistent behavior. It's not like all videos on the front page are looking like they're in german. It's just some of them and afaik there's no way to tell.
And you genuinely can't turn it off completely. Incredibly frustrating and I'm just puzzled by the thought-process that lead to this decision. This would be a pretty cool feature if it was consistently applied and freely configurable.
Moreover I could still prefer movies in my native language *but properly dubbed by some voice actors*, not by some random AI that's going to mess up all the context.
Which is insane to me. Silicon Valley is filled to the brim with multi-lingual people. And yet so many decisions that are coming out have no understanding of languages
Even for the monolingual Google employees it's not uncommon for them to travel to other locales, even as part of their job, so they would be on the receiving end of this "experience" too. We've had 2 decades of experience with this being an issue. One would think that they'd incorporate this "edge case" into their design process matrix by now.
I wouldn't be surprised at all if they one day decided that to have an experience of having no user content automatically translated, you have to pay for the privilege. Call it a multilinguality tax.
I'm in France, but my Google, browsers and devices languages are English. So Youtube randomly auto-dubs (and auto-translates the title of) some French videos into English, and some English videos into French. But they're never the same videos depending on the devices or the browsers. However, the automatic subtitles during the preview remain in the original langage.
Do note that when rolling out features like these, they geoblock them, even on a per run basis, so it might be happening a lot throughout the world but it just hasn't reached your country. For an example, mobile YouTube in the US lets me minimize the video and multitask while still seeing a picture-in-picture window and the audio, while as soon as one lands on France that feature gets immediately disabled.
I've heard about different features in different regions, but GP is also in France.
I am also not connected to my account when I browse in Edge (it's my work PC, it also uses a separate IP), so I don't think it's related to the feature being rolled out on a per-account basis.
I'm in France, my devices are set to en-GB, I've watched only English videos (plus the odd French one) yet youtube decides to auto translate audio in German and lately in Spanish.
I'd love to have a robotic german voice. All I get is the clickbait MrBeast TikTok voice. I get a real reaction when I hear it. I try so hard to avoid the current social media content. It's unbearable. The shock is even greater when I do stumble across it.
Youtube is really the only website that is straight up unusable for me without a set of Addons (uBlock, sponsorBlock, Unhook).
> I don't know who thought this was a good user experience
Which youtube decision of recent years ever thought about user experience?
It's all "company bets" and "promotion tracks".
When it was a fight against TikTok you got Shorts that you can't get rid of.
Now you probably have to "show commitment to our AI offerings" or something. So you get autotranslated videos by a team which will get 500k bonuses and will move on in a month
> Except I can't do that anymore, because there's no guarantee that youtube didn't randomly decide to translate the title of an english video into german
Sp you clicked when you wouldn't have, somewhere an engagement graph went a notch up, and someone will get a pat on the back.
This is one of those articles where the comments are really interesting to read through. I see a bunch of comments who don't agree with the exact math, which might be warranted, but it seems at least directionally correct to me. However there's also a bunch of people commenting that this lifestyle isn't viable for some reason or another, that mainly just boils down to a personal preference those commenters don't want to live without.
But having read through most of the objections I still find myself enticed by this. If I mentally place myself in this position I think I could quite happily live a few decades without talking to anyone for weeks or even months at a time.
I'd still have my pets to give me companionship. Load my kindle up with a thousand books I want to read and just work my way through it. Pick up writing as a hobby and spend the rest of the time working at a gas station and fixing up the house and/or grow some food to offset the reduced income.
Healthcare is an issue. Doesn't seem like a viable place to grow old. Once you become too frail for physical work it's probably just time to die, which isn't great.
You might also like a documentary called Alone in the Wilderness, which tells the story of Dick Proenneke, a man who moved to the remote Twin Lakes region of Alaska in 1968 and lived alone in the wilderness for 30 years.
Proenneke built his own log cabin entirely by hand, using only simple tools (many of which he made himself), and filmed his daily life, including hunting, fishing, foraging, and crafting everything from scratch
Man, I don't wanna see someone going to live as a hermit in the woods and shit in the bushes for $400. I've got plenty of examples of people living in mud huts in Africa for much less.
I wanna see how it's possible to live a decent life in civilized conditions (roof over your head, running water and sewage system, electricity, heating) on $400.
I think a lot of the complaint I see is people want to have 100% of everything they want, within a 10 minute walk, with zero compromise. I.e. they have unreasonable expectations.
I see people come to London from other parts of the world and ask where they can live that is nice and a 5-10 minute walk from the office and I usually laugh in their faces. If you can afford to live that close in central London (even before you think about "nice"), then you don't need to work. Even single car parking spaces in central London cost more than family houses elsewhere in the UK.
When I got my first place it was on the outskirts of London, cost 20x the average UK salary (and for which I obviously took out a huge mortgage for that basically took 75-80% of my salary at the time to pay), and took over an hour on public transport (which was also not cheap - walking is free but probably about 4 hours each way) to get to central London. And this was just in the 2008-era - it's never been cheap (or even reasonable) to own nice property in the center of a large global metropolis like London where you are 10 minutes walk from your office AND transport AND entertainment AND everything else unless you are ultra wealthy.
People need to lower expectations about locality/proximity and affordability before things become viable. This isn't a new problem.
You can go to not quite as extreme levels. There are thousands of small towns all through Appalachia, the Midwest, and the Northeast with a bit of interesting local culture and a low cost of living. Not this low but they also have hospitals and often colleges and other things.
If you are a minority concerned about the culture of red America, the 2016 and 2024 elections both provide excellent county by county color coded maps for you.
Telework is what really unlocks this if you’re a developer.
The Midwest also has many medium sized cities. Not as cheap as small towns but not as expensive as the coastal real estate cost traps. I live in Cincinnati, which has three universities, four Fortune 500 companies, a small startup scene, and over a dozen small neighborhoods with walkable streets (overall you’d want a car but you could get away with not using it every day).
At some point the only economically rational decision is to leave very high cost of living cities. I tell people for cities like SF that it might be good to go there to launch your career but look at it like a college. If by 30 you are not making — for SF I’d say over $300k — then leave. You will never get above real estate in those places unless you are approaching mid six figures.
It also negatively impacts what you can do. Even if you can earn that, it might be with golden handcuffs to a FAANG. Think about that. If you want to start a startup, one that is not lavishly funded enough to pay that, then leave. If you just don’t want to be golden handcuffed to a monster mortgage, leave.
> However there's also a bunch of people commenting that this lifestyle isn't viable for some reason or another, that mainly just boils down to a personal preference those commenters don't want to live without.
Given humans are also just animals that had lifestyles before money and modern society was invented, this doesn't seem like a useful distinction. Either you're talking about your personal preferences you don't want to lice without or money isn't even part of the picture in the first place. Where that line is personally drawn is just as arbitrary to this point at $1 as $1,000,000.
Overall though I agree the article sheds light where we don't normally tend to think. At the same time it crosses points too often to make that those focus of conversation. That is to say it makes a good point with just a little too much PoV twist inserted so people who don't agree will think about that statement instead. E.g. don't complain about not having the home prospects of a boomer because you can live like your great grandparent - which would be something ~40 years prior to the boomer comparison, even for a zoomer (-> driving pushback against the idea owning a house at all is the same as the lifestyle the article laments hearing about -> driving conversation like the above paragraph).
