I'll be concerned if these limits last longer than a couple of days. I doubt the GCP stuff is related (even the article you link connects it to Twitter Trust & Safety while this outage is affecting Twitter's core infra). FWIW Twitter seems fine for me now. It was severely degraded for a few hours on both app and web earlier but looks to have improved. I suspect someone fucked up deploying the new login-wall and they're running damage control, and possibly using the situation to run some experiments (or as leverage for their negotiations with API customers). This is the first major service interruption since Musk's takeover and (unless it persists!) I really think you're catastrophizing too much. I mean, Reddit was completely unusable for several days just a couple of weeks ago (though that was due to managerial incompetence rather than technical); 12 hours of degraded service is a bad look for a major tech company but hardly apocalyptic like you seem to imply.
The egg thing always bothered me, as someone who's had family members that raised chickens. Modern hens produce a shit ton of eggs and will literally go insane ("broody") and starve themselves to death sitting on unfertilized eggs if you don't collect them. What's the vegan rationale for refusing to eat them? I can see being against factory farming or whatever (I don't agree, but I can acknowledge that there exists a consistent ethical system to be against it), but just flat out refusing to consume all animal products regardless of context seems overly dogmatic.
The closest thing to a steelman I can come up with is something like, the chickens didn't consent to be your pet so it's unethical to raise them in the first place. But given that the majority of vegans I've met have pet cats, I don't think that's the logic. If it's ethical to raise and provide for an animal (with conditions superior to what can be found in the wild!) and to do so you need to perform some caretaking task that creates something usable as a byproduct (eggs from chicken, wool from modern sheep) it certainly seems as though you could reconcile eating eggs and wearing wool with being vegan, unless you're willing to bite the bullet and just admit that the modern domesticated breeds of these animals are unfit to survive and should go extinct, which is... a take.
I'm reminded of a recent viral story about OpenAI's use of Kenyan labor to train ChatGPT to avoid generating offensive content: https://time.com/6247678/openai-chatgpt-kenya-workers/
I thought about making a top-level post at the time because the "outrage" was actually infuriating to read. Somehow we live in a society where paying people to read some vulgar words is conflated with literal slavery. I couldn't summon the necessary restraint to write a neutral summary so I refrained from posting at the time, but man, I want off this ride where it is apparently commonly accepted that written words are equivocal (or, you know, in some cases, worse) than actual violence.
I was watching a streamer play the Japanese dub and the voice is normal there, so the only "tell" that the character is intended to be trans is that they're just kind of masculine and ugly.
Trans characters in general seem rather difficult to just toss in for extra inclusivity points. As you mentioned, the aspiration of most trans women is to be indistinguishable from an actual biological woman, not clearly identifiable as some weird third neither-man-nor-woman thing that disgusts normies. But if you include a trans character that obviously passes, then you need to be really hamfisted with the dialogue to make sure whoever's playing understands that yes, this character is trans and it's still likely they'll miss it. So if you want to inject your political views and make sure they actually get across you need to make it obvious through e.g., voice or appearance.
There is a dark humor here, where apparently even in Harry Potter's world of magic and wonders, trans women still look and sound like men in dresses. I don't know that I'd be happy with this depiction if I was a trans activist.
The most impressed I've been with ChatGPT has been when I've pasted in 100-200 lines of my own (uncommented, with not particularly descriptive variable names) code and had it accurately explain precisely what it does, and if prompted, offer reasonable-sounding suggestions for improvement, as well as answer more abstract questions about it. I had a somewhat lengthy Jenkins file, activated by a github webhook, which pulled code, ran a formatter, compiled the Rust code, zipped the resulting executable up, copied the zip file into a directory on a web server, grabbed the last 20 git commits with git log, wrote them into an .html file to provide a quick list of what the newest build contains, and then finally sent a message via webhook to a Discord server to notify users that a new build was available.
ChatGPT had no trouble at all recognizing all of this and even proactively recommended that contributors take care not to leak credentials in git commit messages, and could perform simple transformations on the Jenkins file (like adding a rule to send a different message if the build failed) with no errors. I also had success asking it to rewrite .bat build files to use a Makefile and clang instead of MSVC. I could see ChatGPT in its current state easily shaving 5-10 minutes off simple (but annoying) tasks like this, and also helping with boring API wrangling that is not technically difficult but requires tediously scouring docs to find the appropriate functions. Asking it to write more complex code whole cloth was less impressive and for meaningful contributions to a larger codebase I'd expect you'd quickly run into issues with its limited context window. I'd say it's roughly on the level of a novice (maybe 10th percentile?) programmer with access to Stack Overflow, but it provides solutions instantly. It's certainly more competent than some people I've had the misfortune of working with, though that probably says more about my former coworkers than it does about ChatGPT's capabilities.
I got 14/20. After getting properly calibrated it was not too difficult but the sample size is a little too small. Constantly being thrown new examples that shift the entire distribution around is rough early on. Like, someone who would clearly be left-aligned in the Anglo world is apparently right in Finland, okay, which means this guy who is pretty borderline is actually on the right, etc. I think with 50 or so examples most people would be able to achieve 80-90% accuracy.
I'd say it's pretty common knowledge for, well, anyone who's worked in any large software company. The whole "10x engineer" thing is a bit of a meme (though they certainly exist -- competent people who really are just more efficient, produce better code, solve problems more effectively, and singlehandedly do more than entire teams of other programmers -- and without any of the normie cope about how they can only achieve these results by playing loose with standard practice or eschewing maintainability or extensibility) but you don't even need to believe they're out there when the existence of 0.1x engineers is clearly self-evident. "Rest and vest", whole communities based around juggling multiple WFH jobs simultaneously, etc. Then there are the employees who are literally worse than dead weight, the -x engineers, who somehow seem to stick around due to petty political bullshit, and everyone who actually does work wishes they'd just... stop, because their incompetent meddling just creates more work for people who actually know what they're doing.
The question is whether Elon culled the right people, and it's far too soon to know the answer.
Would you be willing to make a few concrete predictions (e.g., where you think Twitter will be in 6 months, a year, 2 years, in terms of userbase, revenue, whether it is still owned by Musk, moderation policy, etc?)
A quick google suggests you can sideload iOS apps by toggling into developer mode (although this apparently doesn't work if Apple revokes Twitter's dev certificate, as they did with Epic & Fortnite). This is actually better than I expected: I had the impression iOS was totally locked down -- looks like it's only mostly locked down and subject to Apple's whims.
Yes, I'm in complete agreement with you there. The existence of scalpers is a strong signal that the original price was too low and I'd much prefer the counterfactual world where Nvidia raised its own prices 30-50%, generating more profits (thus delivering more value to shareholders). I can't blame scalpers, though -- their actions are not wrong any more than picking up a $20 bill from the sidewalk is wrong -- it's Nvidia's pricing error that they're taking advantage of, so it's Nvidia's fault. Personally I find the people complaining about Nvidia's higher prices to be more infuriating than scalpers. As you say, at least when Nvidia charges more, that value is going to R&D, manufacturing, stakeholders, etc.
This is not really correct. Scalpers, by definition, have no interest in the goods they're scalping -- they don't want them. Their only objective is to arbitrage the price people are willing to pay and MSRP. If there wasn't demand for the products at the price the scalpers were asking then they wouldn't sell and they'd be forced to lower their prices. There is no scenario where scalpers are "distorting the market", they correct pricing errors and make the market more efficient.
Voting is primarily a means of electing people who decide how to allocate public funds. I don't see why individuals who live off the government, at great cost to the taxpayer (something like $100k/yr per inmate in California) should have any input into how other people's money is spent. Determining who has the "right" to vote should probably be rolled under the IRS' jurisdiction anyway, with a voter card sent out with your tax return. If a felon gets out of prison and makes enough money to actually pay taxes, then sure, let them vote, just like any other taxpayer.
There's a separate argument, one that I'm partial to, where franchise should be tied to your net tax contribution (and if you're at a deficit, with more benefits received than paid, then you get no vote until you make up the difference), but that's even further outside the overton window, so.
I wrote maybe half a post about TI last week and never sent it because it seemed kind of unfair to criticize a game I've played and (mostly!) enjoyed for a hundred hours. It's deeply flawed and I can't in good conscience recommend it to anyone unless you're already so deep down the GSG rabbit hole you know what you're getting into. Yet it does so many truly novel, interesting things, in a setting that may be poorly written and poorly explored (although, as far as video games go, I don't think it's that bad, it's just not even close to the greats) but still compelling -- I think it's worth playing.
- Prev
- Next
Well. I tried Bing Chat just now and got this.
It is worth noting that the settings besides "Creative" tend to have worse performance for these sorts of tasks. You may want to rerun it on that. Personally I don't have any difficulty believing LLMs can perform some semblance of "reasoning" -- even GPT-3 can perform transformations like refactoring a function into multiple smaller functions with descriptive names and explanatory comments (on a codebase it's never seen before, calling an API that didn't exist when its training data was scraped). It is obviously modeling something more general there, whether you want to call it "reasoning" or not.
More options
Context Copy link