There’s also a Base64URL variant that is a little more friendly in the modern world where the +/= often need escape sequences.
The first two are replaced with more sensible characters and the third is just removed entirely - do you really need padding?
I don’t think anyone knows. I’m assuming they didn’t have a good reason and are embarrassed to admit that.
Sure, that’s possible.
But Microsoft never actually signed an employment contract with Sam and it doesn’t look like they ever will. Just because someone says they plan to do something doesn’t mean it will happen.
It’s a non-profit. There are no investors.
Microsoft gave them some money in return for IP rights… and they will potentially one day get their money back (and more) if OpenAI is ever able to pay them, but they’re not real investors. The amount of money Microsoft might get back is limited.
Sticky tape.
It’s not a myth - I just fired up the install of Windows I have in a virtual machine. It’s a clean install, downloaded direct from Microsoft with a license key the gave me through their Developer Program… absolutely nothing has ever been installed on it, and the start menu has ads for:
They are definitely ads - when you click on them it takes you to the Microsoft store page… except for Office 365 which I assume is part of OneDrive - I can forgive that one, since it’s part of their free cloud storage service and probably should be integrated into the OS. If you’re not doing cloud storage of some kind, you should be.
It shouldn’t show you as online in discord/slack, but it should be downloading messages/etc so that when you do come online you don’t have to wait for it to sync with all your cloud services.
Also - consider those cloud services might not necessarily be available when you come online - maybe you open your laptop on a train in an underground tunnel or something.
Macs do a good job at this. They have “high efficiency” CPU cores which are still very fast (like, very fast*) but draw about half as much power as the regular cores. Software is also able to schedule background tasks based on various things like power level, network connectivity, how often the user actually launches your app on this device (maybe you have an app installed on all your devices but only actually use it on your phone…).
Background tasks like checking emails, backing up your computer, installing security patches, etc will all run while your Mac is sleeping.
Anti-theft features run even fully powered off. So unplug the battery, and never plug it back in, if you’re going to steal anything with an Apple logo… the fact you can never turn it on does hurt the resale value, but that’s better than going to jail. It’ll phone home as soon as you boot it up too, and even after a full factory reset is still probably tied to the actual owner. You’ll need the owner or Apple to deregister it - and Apple is likely to call the cops unless you’ve got a good story.
(* to give you an idea how fast the “Efficiency Cores” are on a Mac — in Game Mode the “Performance” cores are powered down, because the efficiency ones are more than fast enough and generate less heat - which allows the GPU to be pushed to the limit of the cooling system. The “efficiency” mostly comes from reducing features like speculative execution… though they do also run at a lower clock speed - as in ~3Ghz instead of ~4Ghz)
The Trump government shut down automated mail sorting machines, cut overtime for workers (so if they weren’t keeping up with the workload, they’d just stop delivering mail instead of working a longer shift), replaced a bunch of air mail delivery routes with road ones, added delays to re-delivery attempts when a letter couldn’t be delivered and removed mail collection boxes.
Supposedly all of this would “improve the efficiency” of the postal service. Yeah right.
for the “Rate I currently pay”, which is 0.08c/kwh
How did you get that rate? We pay 33 cents, and it was 24 cents just a few months ago… wouldn’t be surprised if it goes up again next year and the year after since even 33 cents is government subsidised (so - there’s no cheaper option available).
otherwise I would have not have pulled the trigger on a 50,000$ project
Ooof. Why’d you do that? We simply put (a bit over) 5kW of panels on the roof, and a good 5kW inverter. One day of sun generates about as much power as we use in a week, and even if it’s overcast we still come out ahead.
We’re basically only paying for overnight power and pretty easy to keep that to a minimum (with good insulation, efficient overnight appliances, avoiding unnecessary overnight power consumption - such as putting the beer fridge and hot water heater on a timer).
I paid $5k recently without the battery - it’s not just affordable, it’s cheaper than drawing power from the grid. Pay off on the upfront investment will be about 7 years and it has an expected life of 30+ years (we paid extra for long lasting panels).
Battery prices will come down - in the mean time it’s still better to just get that power from the grid.
Facebook and Twitter have become an excellent opportunity and tool to get important causes in from of peoples’ eyes
CNN/Fox are biased, for sure - but that’s nothing compared to straight up lies pushed by large sections of the internet. And those lies tend to perform better than facts on algorithmic timelines that optimise for engagement. For example articles showing “proof” that covid-19 killed various celebrities who are, in fact, very much alive and healthy, with the clear intent to create fear among large sections of society. A tactic that seems to be far too effective.
I think the world needs to go back to human moderation. Like we have on (well run) fediverse communities.
It’d be a lot easier to covertly record someone with a phone - those don’t have a flashing status light.
GPT-4 is already kinda slow - it works best as a “conversational” tool where you ask follow up questions and clarify things that have already been said. That’s painful when you have to wait 10 seconds for a response. I couldn’t imagine it being useful if it was minutes.
There are comparable models to GPT 3.5 “Turbo”, which is faster and 30x cheaper than GPT 4 (if you pay OpenAI’s regular API prices).
I suspect that’s because GPT-4 needs 30x more memory than 3.5.
I’m not aware of any other model that performs as well as GPT-4. In fact I suspect even 3.5 Turbo is the second best model.
To put some numbers on it - RAM runs at tens of gigabytes per second (bytes, not bits). High Bandwidth Memory runs at several hundred or sometimes terabytes per second (OpenAI is likely using the latter, and that memory isn’t just expensive it’s also supply constrained, so the prices are astronomically high right now).
You can buy HBM, and you can use it as your main system RAM, but it’s painfully expensive. The actual amount of bandwidth also scales linearly with with the amount of memory you buy as well. So a 500GB is 10x faster than 50GB - because it write to all of the chips simultaneously (and then read from all of them when you access the data back).
It’s pretty standard on high end GPUs these days. Apple also uses it on all their computers (if you buy a Mac with 64GB of RAM, it’ll run at 800MB/s - which isn’t quite as fast as a high end GPU but it’s close and it is HBM). It’s part of why Macs are so expensive (and also why the cheaper ones have very little RAM).
ChatGPT 4 is estimated to use 700GB of “High Bandwidth Memory”.
… which will set you back about half a million dollars at current prices (which are high, because the manufacturers can’t keep up with demand). Or, you could just pay 20 bucks a month.
They made a $40b profit last year. More than half their profits is a “drop in the bucket”?!
The M1 is around 25% more efficient than the M2.
No that’s not right. The M2 is far more efficient. Third party tests report he M2 MacBook Air lasts up to 3 hours longer than the M1.
Yes, it draws more power under peak load… but it more than makes up for that with better performance allowing it to return to an idle state more quickly. Give an M1 and an M2 the same task, and the M2 will draw less power to get the task done.
I’m on an M1 MacBook Air - Anandtech measured between 11 and 17 watts with an M1 Mac Mini.
However, the Mac Mini has an excessively large cooling system for the chipset it runs (before Apple Silicon, they sold the same Mac with an Intel i7 that turbo boosted to 4.6Ghz).
The MacBook Air has basically no cooling at all and it definitely throttles under high load. It’s still fast enough to get 60fps with good graphics settings while throttled for the games I play - I’d say it’s about on par with my gaming PC that has an entry level Nvidia GPU, but there’s no way it’s drawing as much power as in Anandtech’s testing on an actively cooled chip.
Based on the battery life I’m getting, I’d guess it’s drawing somewhere around 8 watts on average while playing games. It’s a very efficient chip… it draws 0.2 watts while idle according to Anandtech testing. Remember, this family of chips started life on devices with a 10Wh battery and the MacBook Air isn’t much faster than an iPhone.
They might want to, but it’s illegal.
The “quiet period” is a reference to an SEC law that forces any company to be radio silent for a strict 40 day period during the IPO process. Reddit is in that period now and therefore they cannot say a word.
JPMorgan was fined almost a billion dollars for answering questions on a phone call during their quiet period.