It seems to me that few people today actually use extensions. They're hard to find in browsers (in Chrome you need to go through several clicks) and have warnings installing them. I'm always surprised to find that many non-technical people I know don't use any extensions at all. But that's probably intended, Google doesn't seem to like them anyway (esp Adblock).
So the question is, do we in future even need specialized models or will GPT4/5 outperform anything else? If general models can perform well in niche domains, it'd massively speed up adoption in companies.
The 100x analytics seems due to columnar storage like with TimescaleDB, but how do you get improvements on the other things? Is it really faster on a database with good indexes?
But it shows here that it performs better than humans who did that in the past. Few annotation tasks were performed by experts. For example, I guess most annotations for animals weren't created by biologists. So tail cases were already handled incorrectly.
Dishwashers were just an unimportant example. Construction, retail and logistics are examples where there's a lot of very simple tasks that robots cannot perform and that pay very poorly.
I would assume some spy agencies already evaluate this. They get too much material to review manually and ChatGPT could really help identifying people they should look at closer.
If it can be used that way, I guess we can expect export restrictions on free usage and code/weights.
Absolutely. We've been great at building AI in software, but hardware is still absolutely lacking. Robots are nowhere near usable for even simple tasks, at least if they don't follow the same pattern each time. Just look at households, there's still no robot that can empty a dishwasher reliably (unless maybe you build a specialized one for just that task), while it's a task much simpler than what GPT-4 can perform on the software side.
At least they opened up the product. It's available for anyone paying $20 per month and soon via API. Historically, most products of that kind were just aimed at large B2B. They announced partnerships with Duolingo, JPMorgan and a few others but still keep their B2C product.
Not defending their actions, but it's not that common that new very valuable products are directly available for retail users to use.
Maybe it wasn't parameter count increase that made any of this possible but they don't want to give that away. By keeping all developments vague, it's harder to determine if they found some novel technique they don't want others to know.
Is there any way to know if the model is "holding back" knowledge? Could it have knowledge that it doesn't reveal to any prompt, and if so, is there any other way to find out? Or can we always assume it will reveal all it's knowledge at some point?
But did you notice how many construction workers are shown there? You can certainly build faster with prefab, but it's still a very manual process, both at the factory and on the construction site. And anything afterwards (wiring, HVAC, painting) is also nearly 100% manual labor.
It seems we've made amazing progress on the "software" side of AI, but are still very far behind on the "hardware" side, ie robots.
I use it actively and feel like recommendations have improved, at least for me. Ads are a bit more annoying but I've faced zero problems with stability. But I'm only reading, not actively tweeting.
Could very well be. Especially in large companies, it's very hard to judge the importance of some roles and people. That's why there are always complains about how promotions work, no one has figured out how to value the work of an employee.
Don't know much about Meta's HR department, maybe they're really good at that. But in the end we're all humans and never unbiased when we make those decisions.
Then you've been at different small companies than I have. Obviously if you're the only person in tech at that company it's hard to do groundbreaking stuff. But small tech companies solve challenges everyday. And being able to directly convince the CEO (who is often technical at these companies) can make decisions much quicker than at large companies.
Is that bad? You can solve interesting challenges in C# with CRUD stuff. Nothing that scales to millions of users, but there are challenges other than scaling.
That's all of social media. I don't like what's happening there but I also don't see how Facebook engineers can change that. Closing Instagram just drives more people to TikTok which is arguably worse.
The best people I've worked with stayed in the same role for 20+ years without any promotion. They did great work, knew exactly what to do and didn't have to play the politics game to get some favorable review for promotion. I wish more companies would encourage that. Even in tech, the world doesn't change as quickly as we sometimes think.
> Then after a month of ChatGPT hype he wakes up and fires everybody. What a clown.
Changing course when you realize you were wrong has nothing to do with being a clown. Not defending Zuckerberg or the lay off, but for large companies it's always been one of the hardest tasks to predict trends. And many that go under do so because they hold on to old technologies and refuse to admit the world has changed.
If Zuckerberg really changed the whole strategy because he saw VR won't go anywhere and ChatGPT has such potential, changing course drastically would be much better than holding on to old projects due to sunk costs.
Government bonds yield 3-4%. I don't think they ever calculated with a margin of just 4%. It's true that liquidity has dried up to some degree and companies have alternatives what to do with their money, but it's not like they though they'd yield only 3% by hiring more people.
Crypto might stay popular, but I also don't see any indication that web3 will live up to the hype. They still haven't found real use cases that need web3.
Crypto, however, will remain popular as long as governments don't ban it.
Construction is actually one of the few sectors that hasn't become more productive per employee over past decades. Buildings have become vastly more complicated but there's comparably little automation.
Aren't technical challenges at smaller companies usually much more interesting? Pay is worse, but smaller companies usually offer much more freedom to implement your ideas and shape the company.
In most big companies I've worked at, I thought I could identify the people who made everyone slower and were a net negative to productivity. However, they never seemed to be the ones threatened to be fired. Often they were really good at playing corporate games so that it was really hard to justify firing them, while the ones fired were junior employees who had great potential but weren't as good at politics yet.
So while FB certainly intended to fire underperformers and increase average performance, I doubt they were able to identify the right people for that.
So far, there seem to be surprisingly few issues. Some glitches here and there, but overall stability looks still quite good. I would've expected major issues much sooner, especially as they did push out new features in the meantime.
When it comes to refinancing at a bank, they certainly are. It's a very specific subject that few ever come in contact with. It's like kernel programming, also a topic that indirectly affects most people but is truly understood by very few.
If there's one thing the US has enough of, it's space. For Europe, that's going to be much harder. But the US has ample space to not only satisfy their own demand with renewables, but also export it in one form or another.
But sometimes it's saved elsewhere. Home office often increases energy usage in data centers through chat software, video conferencing etc. But it saves much more in energy otherwise wasted for commute.
Likewise, other computations in data centers might also save energy used somewhere else. Given all that servers do, 1% of electricity really doesn't seem much.