YouTube hasn’t gone down that route yet.
And if they ever do, I’m sure at least 90% of premium users will cancel immediately. I like the quality of the curated channels I subscribe, but I won’t die if I don’t watch YouTube anymore. In the end it’s just the same type of content that could be a blog just as well, but unfortunately most people nowadays don’t read anymore.
The creator is already compensated as of now. They earn more if a premium user watches their video than a free user with YouTube ads.
So the sponsor is giving them more money regardless of whether the user is premium or not, which for them is probably a good deal but for us it feels like being double charged.
It’s just how machine learning has been since ever.
We only know the model’s behavior by testing, hence we only know more or less the behavior in relation to the amount of testing that was done. But the model internals has always been a black box of numbers that individually mean nothing and if tracked which neurons fire here and there it’ll appear just random, because it probably is.
Remember the machine learning models aren’t carefully designed, they’re just brute-force trained for a long time and have the numbers adjusted again and again whenever the results look closer or further away from the desired output.
So how come so many websites simply block VPNs with captchas? There seems to be a range of IPs that popular VPNs use and are widely known.
Doing that would require significantly more compute power, so there’s little economic incentive.
In case people didn’t know what company he was referring to. /s
What’s the experience so far?
I see it more of a limitation, you don’t want your laptop to warm (and it shouldn’t in light use), but you want to cool it for the few times it does.
I think they do have their help, but it’s not nearly as dramatic as some companies earning money from it want us to think. It’s just a tool that helps just like a good IDE has helped in the past.
I mean, if LLMs really make software engineering easier, we should also expect Linux apps to improve dramatically. But I’m not betting on it.
AI evangelists act like it’s already perfect and anybody who dares question the church of LLM is declared a Luddite.
I don’t think that’s the case, though. The only people actively “evangelizing” LLMs are either companies looking for investors or “influencers” looking for attention by tapping on people’s insecurities.
Most people just either find it useful for some use cases or just hate it.
Yeah it is a bit weak on the arguments, as it doesn’t seem to talk about trade offs?
Thanks, I’ll try to use it from title to time.
Why wouldn’t companies have already got their data long ago? Internet archive is nothing new.
The article already mentions it.
You can use something like VirtualBox or VMWare. Won’t be the fastest experience, but also not so bad. It’s good enough to have a feel of something.
Nope it’s because on Search it was summarizing the first results, the “pure Gemini” isn’t doing a search at that time, it’s just answering based on what it knows.
Yeah when you use Gemini, it seems like sometimes it’ll just answer based on its training, and sometimes it’ll cite some source after a search, but it seems like you can’t control that. It’s not like Bing that will always summarize and link where it got that information from.
I also think Gemini probably uses some sort of knowledge graph under the hoods, because it has some very up to date information sometimes.
I don’t even think it’s correct to say it’s querying anything, in the sense of a database. An LLM predicts the next token with no regard for the truth (there’s no sense of factual truth during training to penalize it, since that’s a very hard thing to measure).
Keep in mind that the same characteristic that allows it to learn the language also allows it to sort of come up with facts, it’s just a statistical distribution based on the whole context, which needs a bit randomness so it can be “creative.” So the ability to come up with facts isn’t something LLMs were designed to do, it’s just something we noticed that happens as it learns the language.
So it learned from a specific dataset, but the measure of whether it will learn any information depends on how well represented it is in that dataset. Information that appears repeatedly in the web is quite easy for it to answer as it was reinforced during training. Information that doesn’t show up much is just not gonna be learned consistently.[1]
I don’t currently use a VPN but my impression is that nowadays I’d be greeted with captchas everywhere, is that wrong?