I hope you're not blaming the users. It's understandable they would be confused. The software needs to clarify it for the user. Perhaps, when you try to save it, it should warn you that it looks like you tried to redact text, and that text is still embedded in the document and could be extracted. And then direct you to more information on how to complete the redaction.
We have 30 years direct evidence that the users would ignore that warning, complain about the computer warning them too much, insist that the warning is entirely unnecessary, and then release a document with important information unredacted.
The problem is that the user generally doesn't have a functioning mental model of what's actually going on. They don't think of a PDF as a set of rendering instructions that can overlap. They think it's paper. Because that's what it pretends to be.
The best fix for this in almost any organization is the one that untrained humans will understand: After you redact, you print out and scan back in. You have policy that for redacted documents, they must be scanned in of a physical paper.
The problem is that the user generally doesn't have a functioning mental model of what's actually going on
Sorry, but a professional user not having an operational understanding of the tools they're working with is called culpable negligence in any other profession. A home user not knowing how MS Word works is fine, but we're talking desk clerks whose primary task is document management, and lawyers who were explicitly tasked with data redaction for digital publication. I don't think we should excuse or normalize this level of incompetence.
I don't expect radiologists to have a good understanding of the software involved in the control loops for the equipment they operate. Why should a lawyer have to have a mental model or even understand how the pdf rendering engine works?
Have you ever had to actually react a document in acrobat pro? It's way more fiddly and easy to screw up than one would expect. Im not saying professionals shouldn't learn how to use their tools, but the UI in acrobat is so incredibly poor that I completely understand when reaction gers screwed up. Up thread there's an in complete but very extensive list of this exact thing happening over and over. Clearly there's a tools problem here. Actual life-critical systems aren't developed this way, if a plane keeps crashing due to the same failure we don't blame the pilot. Boeing tried to do that with the max, but they weren't able to successfully convince the industry that that was OK.
if a plane keeps crashing due to the same failure we don't blame the pilot
That's true, we blame the manufacturer and demand that they fix their product under threat of withdrawing the airworthiness certification. So where's the demand for Adobe to fix its software, under pain of losing their cash cow?
Yet, people here are arguing that it is perfectly OK that professionals keep working with tools that are apparently widely known to be inappropriate for their task. Why should we not blame the lawyers that authorized the use of inappropriate tooling for such a sensitive task as legal redaction of documents?
The link in the comment you are replying to has a screenshot of exactly this. it’s a prompt with a checkbox asking you to delete the metadata and hidden info involved with the redaction. you’d have to blaze past that and not read it to make this mistake. It is user error.
I guess if you really want to defend users here you could say people are desensitized so much by popup spam that a popup prompt is gonna just be click through’d so fast the user probably barely recognizes it, but that’s not the software’s problem. For whatever reason some users would prefer to just put black boxes over obfuscated text, so here we are
Professional users doing more than 1 document? Yes, I'm absolutely blaming them.
I agree that affordances are good, but tools are tools, they can have rough edges, it's okay that it occasionally takes more than zero knowledge and attention to use them.
If software developers designed hammers, you'd have to twist the handle before each swing to switch from tack to nail mode. And the two heads would be indistinguishable from each other.
If business MBA's designed them, you'd wind up with the SaaSy Claw 9000, free for the first month then $9.95 in recurring subscription fees, and compatible only with on-brand nails that each have a different little ad imprinted on the head.
But it doesn't matter, because by the end of the year all construction will be vibe-built from a single prompt to Clawde.ai, which will pound non-stop, burning through $1T of investor funds, and confidently hallucinate 70% of the nails until the roof collapses on the datacenter destroying the machine and civilization along with it, and a post-singularity survivor picks up a rock and looks calculatingly at a pointy shard of metal...
The software could do better, sure, but in this case the accountability clearly falls on the lawyers. It's their job - and it's a job that can profoundly impact people's lives, so they need to take it seriously - to redact information properly.
> Folks around here blame incompetence, but I say the frequency of this kind of cock-up is crystal clear telemetry telling you the software tools suck.
Absolutely. They know this is confusing, and they're bound and determined not to fix it. At the least, they need a pop-up to let you know that it's not doing what you might think it's doing.
Apple’s Preview app does exactly that. I discovered this while trying to make a blanked copy of kid #2’s homework worksheet for kid #1 who left his at school after kid #2 already wrote on her copy.
I’m optimistic that because LLMs have brought down the cost of the mere act of typing out code that we will see a shift in focus on certification and verification. Preferably with some legal protection for customers that are sorely lacking today.
Road surface and consequently traction can vary (for example, gravel). It's important to stay well within margin. Once you lose traction at speed, it can be hard to get it back.
TL;DR take it to the track, where the road surface is well maintained.
Technically yes but on any given track day there can still be gravel, weeds, rubber, bits of material from car body kits/aero, and even various liquids that have leaked.
If you do hit gravel/oil, tracks at least have runoff areas or soft barriers, and no oncoming traffic or cliffs to worry about.
Every track day I've attended required the cars to have been inspected for leaks and loose components. And they were quick to clean up any debris or oil.
Not that tracking cars is the safest hobby, but if someone is gonna drive like that regardless its far safer at a track than on public roads.
Like, Apple knows what you're watching within the Apple TV app obviously.
But it's certainly not taking screenshots every second of what it displaying when you use other apps -- which shows and ads you're seeing. Nor does Apple sell personal data.
Other video apps do register what shows you're in the middle of, so they can appear on the top row of your home screen. But again, Apple's not selling that info.
I mean obviously that's what things like environmental and safety regulations are for, as well as things like antitrust. You have to set the "rules of the game."
I just mean that otherwise, usually competition ensures good outcomes for consumers, because the corporations that produce bad outcomes go out of business once consumers catch on.
But there are definitely exceptions, especially around rare events that are difficult to foresee or that can't reasonably be expected to be part of product comparison. The likelihood of your account being shut down without recourse and losing things you've paid for falls into that category perfectly. Predatory surprise fees with things like credit cards and bank accounts, and that change without warning, also fall into that. Also minimum warranties, since consumers can't easily inspect quality on the inside of a product.
> For example, the unaligned model would give uncertain answers in terms of percentages, and the aligned model would use less informative words like "likely" or "unlikely" instead.
Percentages seem too granular and precise to properly express uncertainty.
Seems so, yes, but tests showed that the models were better at predicting the future (or any time past their cutoff date) when they were less aligned and still used percentages.
reply