Diff's recent activity
-
Comment on Google removes some of its AI summaries after users’ health put at risk in ~tech
-
Comment on Google removes some of its AI summaries after users’ health put at risk in ~tech
Diff Link ParentHealth is one narrow well that isn't completely poisoned. There's a few of those left, programing is another, but they are fairly narrow and still very clogged.Health is one narrow well that isn't completely poisoned. There's a few of those left, programing is another, but they are fairly narrow and still very clogged.
-
Comment on Google removes some of its AI summaries after users’ health put at risk in ~tech
Diff Link ParentAnd it's poisoned the entire well of the internet. There's always been SEO-optimized nonsense sites clogging search results, but it's nigh impossible to find any good sources of information...And it's poisoned the entire well of the internet. There's always been SEO-optimized nonsense sites clogging search results, but it's nigh impossible to find any good sources of information nowadays that aren't from one of the mainstream social media sites.
-
Comment on Google removes some of its AI summaries after users’ health put at risk in ~tech
Diff Link ParentThere's just too much potential to go wrong. It's whack-a-mole with one hammer and a million parallel moles to hit. Someone asking how to tight to tighten their car's lug nuts, asking how long a...There's just too much potential to go wrong. It's whack-a-mole with one hammer and a million parallel moles to hit. Someone asking how to tight to tighten their car's lug nuts, asking how long a food stays good in the fridge, asking whether two medications interact. There are so many ways this can go wrong and cause serious harm. When it's human-authored, a human can be held accountable. These systems continue to prove they can not be held accountable, but are continuing to prove they have massive potential for harm.
AI usage at work vent
And people are relying on them more and more, and they trust them against continuing evidence not to. Today at work there were three separate instances where my boss just trusted the AI overview answer, even in one case where the AI's answer was internally inconsistent, one where it was just plain disconnected from reality, and one where it gave incorrect information on a niche USPS policy topic that would have cost us a good deal of money and time to get wrong on an order that's been cooking for weeks.It could be worse, that's certainly true. But there's just no way to do this right. It's a rotten idea from the start.
-
Comment on Mac advice for a long time Windows user in ~tech
Diff Link ParentNitpicky note on both of the above comments, but we're talking about standards compliance so it's probably one of the few places it's warranted. This article speaks on UNIX compliance, which is...Nitpicky note on both of the above comments, but we're talking about standards compliance so it's probably one of the few places it's warranted.
This article speaks on UNIX compliance, which is different from POSIX compliance. macOS is UNIX certified, Linux is POSIX compliant, and (as I understand it) UNIX is a superset of POSIX, so macOS is likely more POSIX compliant than it is UNIX compliant.
100% agreed on your conclusion though. I migrated to macOS from daily driving Linux, and I still feel as comfortable and at home in the terminal on my MacBook as I do when SSH'd into any of my Linux servers.
-
Comment on Is YouTube's use of AI upscaling for Shorts unethical? in ~tech
Diff (edited )Link ParentFully, yes. People's attention was drawn to this not by an announcement (it was an entirely silent launch), but by the actual visual effects it has. The technique they're doing is not just...Fully, yes. People's attention was drawn to this not by an announcement (it was an entirely silent launch), but by the actual visual effects it has. The technique they're doing is not just upscaling, it's also adjusting contrast, vibrance, sharpness, and it has a mild but noticeable impact on content that was not intended. For most content, this can be noticed as just an obnoxious oversharpening effect combined with a smeariness of unclear detail. For some content, this highlights edges that were supposed to be subtle in a way that actually reduces visual clarity or just negatively impacts the aesthetics. And although I said it wasn't "just" upscaling just now, it also doesn't really appear to actually be any kind of upscaling at all, as it's donating obnoxious generative-AI-esque smears, smudges, and nonsensical detail to already-high-resolution videos.
I follow one artist, SavannahXYZ, who posts claymation-style 3D models that often have subtle fingerprint details in the materials. This AI upscaling, particularly the oversharpening and edge contrast boosting, makes that subtle detail leap out at you. The effect is honestly (mildly) unsettling.
I've noticed for some types of animation, particularly animation that's primarily line art, where the edge highlighting ironically makes it quite difficult to actually even see the lines from a distance, as the brightening effect around the darker line art (again, from a distance) averages out into blending right in with the background color.
So, yes, people noticed it in the first place because it was affecting their content in ways that no other video host does. In ways that no longer faithfully represent their artistic vision. Some videos, styles, and lighting scenarios are affected worse than others by this automatically-applied and impossible-to-remove technique. The AI is just the shit cherry on top for many of those artists.
-
Comment on A new era of intelligence with Gemini 3 in ~tech
Diff Link ParentThey are primed to do so, like stu said it's part of the package. When LLMs imitate a person it is because people make up the whole of their training data. They are trained to predict likely text...They are primed to do so, like stu said it's part of the package. When LLMs imitate a person it is because people make up the whole of their training data. They are trained to predict likely text based on a huge variety of human text, with emotions included. Part of their instructions could be to keep a neutral tone, but that doesn't eliminate that massive signal from the training data.
-
Comment on A new era of intelligence with Gemini 3 in ~tech
Diff Link ParentThis matches my messy, anecdotal experience in this VS Code ripoff they simultaneously released. Gemini 3's having so many API issues it's difficult to get through a single prompt cycle, but it's...This matches my messy, anecdotal experience in this VS Code ripoff they simultaneously released. Gemini 3's having so many API issues it's difficult to get through a single prompt cycle, but it's actually capable of navigating a messy, old codebase with quirky interactions, researching existing examples, and synthesizing new-ish behavior.
I don't want to praise it too much, when I informed it that it was stuffing its own files in the examples folder rather than the src folder, it deleted all of its own files before trying to move them, got briefly confused when that failed due to the files no longer existing, and then had to rewrite them all out of its context window (which was still intact at this point), and it still assumes connections between issues where there are none.
But that's far, far better than the Sonnet and GPT-OSS that are also available in Antigravity. Using either of those models results in an immediate, irredeemably nonsensical crash and burn. There's an awful lot of handholding going on, awful lot of silly mistakes, but there's also a lot more capability present than I've ever seen in my past attempts at giving this AI thing a fair shot. I don't see much point to Antigravity though.
-
Comment on AGI and Fermi's Paradox in ~science
Diff Link ParentI think training it to emulate people is a dead end for AGI. There's only so much data that humans produce, and it's all incredibly lossy. LLMs have already scraped the whole of the internet and...I think training it to emulate people is a dead end for AGI. There's only so much data that humans produce, and it's all incredibly lossy. LLMs have already scraped the whole of the internet and every book and resource they could pirate, and are stalling out hard. They're working backwards through too many lossy layers of indirection and abstraction to ever be able to reach AGI, in my opinion. It's like building a console emulator based on only game reviews, screenshots, and HD footage of games played on it. Not only is straightforward emulation incredibly resource-intensive, it's just not enough info to work from.
If AGI is ever going to exist, I think it's a lot more likely to happen starting with a small, independent intelligence and figuring out how to make it more complex. I don't think that will have any relation or likeness to people or the way we operate.
-
Comment on Part of me wishes it wasn't true but: AI coding is legit in ~tech
Diff Link ParentIt's not how stigma works, but I don't think my statement is inaccurate. The stigma would not evaporate from the face of the earth, but it'd be hugely diminished if it was more capable. People...It's not how stigma works, but I don't think my statement is inaccurate. The stigma would not evaporate from the face of the earth, but it'd be hugely diminished if it was more capable. People roll their eyes when an AI authored PR comes in because everyone can make a reliable, highly accurate assumption that it will not be worth the time it takes to finish reading the emoji-riddled description, let alone the PR itself. The stigma would be a lot less if that wasn't the case.
And even if I'm incorrect about all of the above, I think we can both agree that there's more than enough support, excitement, and bubble to insulate anyone who would want to defect and embrace the AI lifestyle, even under a pseudonym, while they build up and advertise enough inarguable evidence to convince outsiders of the way, the truth, and the light.
-
Comment on Part of me wishes it wasn't true but: AI coding is legit in ~tech
Diff Link ParentFor some people and some projects, but that's very far from universal. It's not getting anyone blacklisted from the industry if they allow AI PRs. If AI is the future, some eager maintainer would...For some people and some projects, but that's very far from universal. It's not getting anyone blacklisted from the industry if they allow AI PRs. If AI is the future, some eager maintainer would give it a try, find it true, and start advertising that fact. There wouldn't be a stigma if it could back up its words with action.
-
Comment on Part of me wishes it wasn't true but: AI coding is legit in ~tech
Diff Link ParentIs the statement "This is not arguable" an argument? No facts or statements were put forth as proof or evidence. I think it's a tool. It has uses, and I make use of it. But like the 10x engineer...Is the statement "This is not arguable" an argument? No facts or statements were put forth as proof or evidence.
I think it's a tool. It has uses, and I make use of it. But like the 10x engineer myth before it, I don't see any indications that multipliers are a thing that exist outside of games. The studies that have been published, while preliminary, hint at AI reducing productivity in the short term, and having the effect of loss of skill in the not-as-long-as-you-might-think term.
-
Comment on Part of me wishes it wasn't true but: AI coding is legit in ~tech
Diff Link ParentOf course it's up for debate. It's an ongoing debate everywhere right now. The studies aren't in, the polls have barely been sent out, the conclusion is not done and written in stone.Of course it's up for debate. It's an ongoing debate everywhere right now. The studies aren't in, the polls have barely been sent out, the conclusion is not done and written in stone.
-
Comment on Part of me wishes it wasn't true but: AI coding is legit in ~tech
Diff Link ParentThe break in this reasoning is that we have strong CGI advocates. We have people who believe that CGI is that future and who would hold up as shining beacons any good examples. If AI had the...The break in this reasoning is that we have strong CGI advocates. We have people who believe that CGI is that future and who would hold up as shining beacons any good examples. If AI had the capability of Silicon Valley's claims, they'd be right to. It would be revolutionary. It would be obvious, by looking at the rapidly growing collection of claims. We would have entire projects with thousands or millions of downstream users, maintained in whole or in large part by AI. Those projects are nowhere to be found. The only growing cries are from maintainers who are being overloaded in entirely new ways. The existence of bad CGI isn't the problem here, any tool can be misused, it's the lack of good CGI.
-
Comment on Zen browser / chrome alternatives in ~tech
Diff Link ParentThey can on occasion be reasoned with, too. They briefly disabled the built in password manager until it was explained that that'd be A Generally Inadvisable Move..They can on occasion be reasoned with, too. They briefly disabled the built in password manager until it was explained that that'd be A Generally Inadvisable Move..
-
Comment on Zen browser / chrome alternatives in ~tech
Diff LinkI used Zen for a while, built some mods, contributed to upstream, developed some tooling for mods, but Firefox added the features I really wanted, vertical tabs and groups, and had better...I used Zen for a while, built some mods, contributed to upstream, developed some tooling for mods, but Firefox added the features I really wanted, vertical tabs and groups, and had better performance and battery life on my borderline-performing machine due to some of the visual effects Zen used accidentally breaking some performance optimizations under the hood. It still feels like a very young project and I have an on and off relationship with tinkering with my daily driving tools.
-
Comment on I think nobody wants AI in Firefox, Mozilla in ~tech
Diff Link ParentIf you're referring to the terms and conditions update earlier this year, Mozilla never sold user data, and they were never looking to.If you're referring to the terms and conditions update earlier this year, Mozilla never sold user data, and they were never looking to.
-
Comment on Suggestions for a new Steam Deck user looking to make Desktop mode pleasant to use? in ~games
Diff LinkI prefer using the GNOME Desktop on my Deck, I just find Plasma to be subtly janky and worse feeling. I have huge respect for that community and their devs, but there's just something... off....I prefer using the GNOME Desktop on my Deck, I just find Plasma to be subtly janky and worse feeling. I have huge respect for that community and their devs, but there's just something... off. GNOME has had some regressions recently on the Deck with touch support, but still overall is a better experience IMO. The touch screen on the Deck is pretty weak to begin with, anyway.
The other thing that helps is a desktop controller layout tuned for desktop use. For me, the shoulder buttons and top two grips map to Win, Alt, Ctrl, and Shift, Tab on Select, Esc on Start. Left trackpad as a scroll wheel (with click -> middle mouse button), as well as a chord for volume control. I use that layout on my Steam Controller for my actual desktop and I don't use a mouse. Wacom or Steam Controller, but I am mouse free for many years. Sometimes I use the Wacom with the Deck and that works pretty smoothly with GNOME as well.
-
Comment on Does anyone use AppleScript on macOS? in ~comp
Diff Link ParentShortcuts really is something special, I actually have a ton of thoughts about it. Like, how, in the year of our lord 2025, is there not an Apple Intelligence Siri with ridiculous amounts of...Shortcuts really is something special, I actually have a ton of thoughts about it. Like, how, in the year of our lord 2025, is there not an Apple Intelligence Siri with ridiculous amounts of freedom to interact with your device? With LLMs, we could reasonably have a Siri where you can ask, "Hey, can you add the currently playing song to my work out playlist?" and it could stitch together Shortcut blocks to Get Current Song, and Add To Playlist.
And that's a very basic example. It also seems reasonable to be able to share a document to Siri, ask it to extract the first two pages into a new document and save it to a folder in your notes, and if you perform that action multiple times, it could offer to create a dedicated shortcut for that action to bypass Siri entirely next time. It'd also make it an incredibly powerful on-ramp to people taking more control over their phones and experimenting with Shortcuts on their own.
It's just me and my thoughts over here, so maybe I'm overlooking something obvious, but that seems like an actually decent use case for LLMs, and yet there's no movement towards it.
-
Comment on Does anyone use AppleScript on macOS? in ~comp
Diff LinkApple's OSA also supports JavaScript if you'd like something a little more familiar. AppleScript is nifty but that's the route I usually take when I have less-trivial things to automate. You get...Apple's OSA also supports JavaScript if you'd like something a little more familiar. AppleScript is nifty but that's the route I usually take when I have less-trivial things to automate. You get even less documentation though.
Also often overlooked is Apple's Shortcuts app.
Below in my work vent, the internal inconsistency was actually due to exactly this, a bad unit conversion. It actually had (mostly) decent math, but it generated a final answer at the start of the overview, before it ever generated the math question, which was clearly wildly incorrect when units were converted to match the work it showed below.