sfRattan

> The idea of "contradictory data" is a projection that you're imposing.

No. Using the topic in this subthread: the methods described as "whole language reading instruction" contradict the methods of learning to read supported by science. That is an observation. If you think it a projection or a hallucination, then maybe there is nothing left to say because we are talking past each other and you are being extremely pedantic in a way that is unproductive to communicating a problem that, yes, exists in the minds of humans. Humans who use LLMs without understanding LLMs, and walk away with information that is incorrect, useless, or worse-than-useless in the real world.

If you genuinely think computer-literate people are projecting the notions of contradition and/or sycophancy into the workings of an LLM, I'm not sure how to persuade you otherwise. But I suspect most people on this forum using phrases like "sycophancy" understand what an LLM is (and why it is not technically a literal sycophant) and simultaneously understand that the phrase "sycophancy" provides a useful cautionary, metaphoric image for average people who are using LLMs absent deeper understanding of the technology.

I think what those observations missed is that the best readers quickly build up a large baseline vocabulary of known words, whether by phonics or something else, and therefore appear to be guessing often or, when they're guessing, have a much better understanding of the surrounding prose than a median learner who has only been taught to guess at every word.

And yes, something similar is happening with common core math. Curricula are trying to teach the tricks, shortcuts, and mental re-arranging of numbers that are only possible to comprehend once the basics are memorized and internalized. It was miserable memorizing times tables, but I remember being able to start to see the ways you could combine and recombine numbers after I had successfully memorized them. Trying to teach those things without any memorization is putting the cart before the horse.

Yes, the podcast goes into exactly that detail.

IIRC it was a teacher in New Zealand[1] who had teacherly charisma, was earnestly committed to her method, and was so profoundly wrong that she became the first domino of a regression in reading ability which we are still feeling today across the English-speaking world. And teachers and education professionals became fans of both the method and the growing personas behind it. There were Americans who turned whole language hokum into a product for sale... using the cult-like sales tactics of Amway and other MLMs but applied to a target audience of largely woman-attended teachers' development conferences and fairs instead of Mormon housewives.

Sadly, humans were capable of producing plausible sounding nonsense long before the advent of LLMs. And promoting it persuasively.

[1]: https://en.wikipedia.org/wiki/Marie_Clay

> There’s no contradiction from the point of view of the model and its responses, and no need to invoke sycophancy or flattery as a reason for the behavior.

Sycophancy and flattery are words that relate to how humans experience LLMs. It's appropriate to use words that could anthropomorphize as descriptors for a problem when that problem exists in the space of human experience. It doesn't matter that an LLM is merely a predictive model. It matters that humans who use the model perceive it as a conversation partner and walk away feeling 'flattered' by the model's predictive responses to the humans' input, which the model predicts without regard for factual accuracy, which happens in turn because (as you point out) "there’s no contradiction from the point of view of the model."

We could call the problem "LLM predictive output which tends to have the effect of flattering the user's opinionated assumptions without regard for factual accuracy" if you prefer that much of a mouthful, but "sycophancy" describes the problem well and is one word instead of twenty.

> One prompt will elicit a response based on one set of weights that are most closely related to the prompt. A different prompt will elicit a response from a different set of weights, for the same reason.

Yes or, in other words: there's sufficient contradictory data which were used to train the model's weights that it will predict (a series of tokens which when read in order in English convey) accurate information for some prompts and (a series of tokens which when read in order in English convey) inaccurate information for others. That the contradiction is beyond the LLM's scope is part of the problem. The two identical parentheticals above are (or ought to be) as unneeded as pointing out that a statistical model cannot technically be sycophantic.

> That seems like an unfair prompt. You literally told it to do a specific task and it did as instructed. For sycophancy...

You redefine syncophancy too narrowly. Many (if not an outright majority of) everyday LLM users have neither the educational foundation nor skill to ask good, well-formed questions like the one you propose. We should consider syncophany in the scope of an average person's non-optimal use of LLMs. CrazyStat's prompt is something your average Joe will absolutely feed an LLM, and Joe will walk away with something very much like those plausible falsehoods. We have to think of and work against LLM syncophancy in the scope of how the hypothetical average person will use an LLM.

It might not obvious to a number of readers on HN, but whole language reading instruction is soundly discredited[1]. For those who prefer listening, the America Public Media podcast Sold a Story[2] covers the history of reading instruction in the English-speaking world and the rise and (slow) fall of whole language hokum in extensive depth.

And CrazyStat's implication strikes me as still largely true, so I'll make it explicit: even frontier LLMs will still resort to sycophancy when there's sufficient contradictory information in their weights to reproduce only some of it in response to a user query. And the harnesses that LLM providers silently, invisibly integrate behind-the-scenes to counteract hallucination and syncophancy are definitionally patch-jobs at best. The latest models may be better when explicitly instructed or configured by default to search the web and pull sources into their context, but the root problem remains unsolved and can still rear its flattering head if you're not paying close attention.

[1]: https://journals.sagepub.com/doi/10.1177/1529100618772271

[2]: https://features.apmreports.org/sold-a-story/

> ...imagine if your dashboard controls and icons, had occasional tweaks in size and shape that made them slightly harder/easier to use, but over time ended up with controls you found more intuitive and easier to use.

When even a single button disappears from where I expect to find it in an application (and reappears somewhere else), that has never resulted in the application feeling more intuitive or easy to use. It has only ever been frustrating in the most literal sense of that word, and only every caused me to start looking into other programs.

A more subtle undermining of the user's trust that their tools will be where they left them on the screen the next time they open the application is arguably more insidious, not less. Both because it is more maddening to the end-user than a full redesign and because the damage of lost trust and user ire is harder to measure than the A/B engagement data until things progress too far for any easy, non-painful remediation.

A/B testing can be helpful as a part of an actual user study on users who are aware of what is happening. But unleashed on an unsuspecting user population out in the real world with real things to do on real deadlines, A/B testing inevitably becomes a harmful dark pattern.

> A/B testing users without their knowledge and enthusiastic consent is unethical.

Yep, and A/B testing as experienced by uninformed, unaware end-users is a dark pattern.

It undermines the perception of (and trust in) continuity which is necessary to make effective use of a tool. The best way I can describe it to the skeptical is: imagine the dials on your car's dashboard rearrange themselves occasionally overnight, and on some commutes to work you suddenly can't work the radio or the AC while moving at ≥35mph. Of course, since the widespread use of touchscreens, that example became very literal.

So the car manufacturer has figured out the "optimal" arrangement of dials and buttons on their dashboard for their preferred levels of user engagement. Great. How many of those users now associate their car's brand with inconsistency? "I can't trust the damn buttons to be in the same place the next time I drive."

Humble Bundle frequently has nice bundles of genre fiction or comics by specific authors. I've picked up large sets of Ebooks from them over the last few years by numerous authors: Ursula K. Le Guin, Robert Jordan, comics by Moebius, and recently Timothy Zahn and various scifi/fantasy authors with careers spanning decades.

It's a great resource!

KOReader is good, as others have mentioned.

If you want a cross platform option that's pretty good across Android/iOS/Mac/Windows/Linux, Readest[1] Is another great option. It supports syncing a library of ebooks and is licensed under the AGPL[2].

[1]: https://readest.com/

[2]: https://github.com/readest/readest

My homebase has the around 30TB usable storage, but the only workloads I've found really need to run on the same machine (rather than the same local network) are media transcoding for Jellyfin and ML classification of footage in frigate. With those two happening, and with ZFS (appropriately) using a majority of the RAM as a cache, it makes sense to put the rest of services on another machine.

My prod (now homeprod, thanks for the name!) has no HDDs, just SSDs. It's a proxmox cluster that runs opnsense in a VM for the whole network, Home Assistant also in a VM[1], and most other services in containers. In the age of expensive storage, it has just enough SSD space for VMs and containers. VM backups live on homebase, except for opnsense which also needs a backup on the same machine (same machine, not just same cluster) if the network goes down.

So my conceptual distinction is mostly that the only services/containers I put on homebase are doing work that really benefits being that close to the data. Oh, and I'm in the jolly position of needing to replace the hard drives this or next year. Yaaaaay.

My homelab is currently a cluster of 4GB Pis that I had lying around from before even the Raspberry Pi shortage. Whenever I upgrade my Framework 13 mainboard, I'll probably use the old one as a homelab instead of the Pis.

[1]: Home Assistant's plug-ins are themselves containers, so it's a good idea to give it either a VM or its own machine.

I very much like how the name "homeprod" contrasts against "homelab."

But I'm not as convinced as the article's author that "NAS" is outdated as a separate concept, just that as new people learn more about networking they rediscover why just having storage on in its own box can be a good setup. And when new people are brand new, having everything on one box is conceptually simpler, though not always optimal.

I think, in the family/household context, I like the names:

Homelab: the box/cluster you use to experiment and test new stuff.

Homeprod: the box/cluster you use to deploy stable services for family. Runs only things that have graduated from at least a yearlong successful deployment or so on homelab.

Homebase: the box/cluster you use to make ZFS pools and databases available to both homelab and homeprod (on the same network or overlay network). Basically a NAS in the older sense of the word, before homelabs started expanding its scope in casual use.

Famadmin: the sysadmin for the family (i.e. "you" in the three items above). The person to whom, as you say, "providing services that support your family's larger existence, independent of the surveillance industry," is a matter of deep principle. And well put, BTW.

On the flipside, if I fall asleep reading my ereader, I almost never lose my page. It just goes to sleep too, with the last open page ready to resume.

> I realised it doesn't improve my life, to care about folder structure and naming, so much that I insist on retaining it and hobbling the media browsing frontend.

There's no good reason for any folder structure to hobble a media browsing frontend. There are multiple frontends I've tried for my personal library of ebooks (mostly Humble Bundles of software/hardware books), and a number of those frontends have an option that's basically one book per folder (i.e. all identically named files in that folder treated as different versions of the same book... epub, pdf, etc) and folders above that can be whatever you want. BookOrbit[1] supports arbitrary folder hierarchies and Readest[2] supports at least one level of intermediate hierarchy.

In the context of existing software counterexamples, Jellyfin's delicate sensitivity to folder meta-structure is a regression from the baseline expectation. But I'd agree that ultimately folder naming requirements for the media items themselves makes good sense.

> In which case, enjoy your UI being a series of nested folders.

When the Jellyfin UI reflects those nested folders, I can tell it to only rescan a specific subfolder when I add a movie, rather than checking the whole library for new content. Folders and hierarchical structure are objectively useful, though our search-addled brains make that less obvious than it used to be.

[1]: https://bookorbit.app/library-file-structure/#grouping-folde...

[2]: https://readest.com/docs/library#import

Not OP, but I've run into similar instances of Jellyfin "going nuts."

Mixed media type libraries are depreciated and "broken" according to the documentation[1], which is fair enough, but sometimes this results in weirdness if a TV-show is also part of a franchise and has associated TV-movies (e.g. the Battlestar Galactica reboot, Futurama, or Breaking Bad). If you want to keep those movies on your file system in a folder with the other stuff from the franchise, Jellyfin often can't handle it. And it's not obvious to new users whether these TV-movies belong in a TV-shows library or a movies library.

I've also noticed it cannot handle additional layers of folders (i.e. for Genres) in TV show libraries, but it can handle that layer in movie libraries. Failure mode (the "going nuts") looks like those genre folders recognized as bizarre or obscure television shows, and nothing scanned the next layer down.

I haven't looked into how their scanning algorithms work, but it's definitely frustrating that there's not a button or feature to say: "No, Jellyfin, this is a folder, not a show or movie. Try scanning again on the subtree starting here."

Honestly, any software that either silently/automatically reorganizes a user's files and folders (not Jellyfin, but there are others that do this... even FOSS projects) or demands a specific meta-hierarchy on the file system is frustrating to anyone who deliberately keeps their files and folders well-organized.

[1]: https://jellyfin.org/docs/general/server/media/mixed-movies-...

The details you mention are all hammered out to death and back as a regular part of the legal process, before and during a trial, and by both prosecution and defense. One can only imagine a huge potential for abuse by suspending any and all familiarity with that legal process.

> But it does AFAIK care about plausible deniability. Whether or not any given instance of denial is plausible must of course ultimately be determined by a judge.

There is near zero chance that a judge or jury will decide you have plausible deniability for providing a duress PIN when asked by police for an unlock PIN. It is your phone, you had to configure the duress PIN when setting up the phone, knowing what it does, and provide it to police instead of the unlock PIN.

> I am advocating for civil disobedience in defense of privacy and the right not to self incriminate.

You have a right to silence, not a right to lie, and certainly not a right to lie in a way that is the proximate cause of destruction of evidence during a lawful search. Call the lie civil disobedience all you like, but expect totally avoidable federal felony charges in cases like then one at hand, when all you had to do was remain silent (especially as in this case with a GrapheneOS device).

> I don't believe there's any prohibition against reading people's minds

Common law often doesn't require specific prohibitions (or specific updates in the manner of computer code) because the practice of common law has long functioned on the good faith application of human reason and common sense by jurists. Consider the relevant language of the 5th Amendment: "...nor shall be compelled in any criminal case to be a witness against himself..." I think it's extremely likely that, when mind-reading tech is invented, it will be treated as a form of compelled testimony, and therefore impermissible against a defendant (though potentially permissible against a witness with immunity).

> If the prosection doesn't have to demonstrate that something was valid evidence beyond a shadow of doubt that's rife for all sorts of abuse.

Assuming that is the preceding paragraph you reference, then there is nothing to respond to because the law is clear. As zugi mentions, obstruction of a federal investigation by tampering with evidence (i.e. potentially anything in the scope of a particular search of which you are aware) in just about any sort of way is illegal under 18 USC § 1519 (and again, most states have similar laws with regard to state officials/departments and searches).

This subthread, the general thread, the linked article, and the original event it covers are all within the context of someone aware they are detained or under investigation. In that context, the prosecution doesn't need to demonstrate that what was destroyed is evidence. Just that the suspect obstructed the search by destroying, altering, or tampering with something in the search's lawful scope.

This state of affairs isn't any more or less rife for abuse than the general process for obtaining a warrant or the standards for an officer assessing probably cause.

> In general I don't believe I can catch a (legitimate) felony for locking myself in the bathroom and flushing the toilet during a police raid.

Most juristictions have a duty to preserve or not to spoil evidence once you are clearly detained or a search is clearly imminent (i.e. the police are knocking on your door), or when litigation can be reasonably anticipated. Behavior like locking the door and then burning/shredding papers, wiping a phone, or even flushing a toilet can absolutely be considered destruction-of/tampering-with evidence past that point, even (and especially) if the evidence is impossible to recover.

> I was distressed and got confused and gave the wrong PIN. Anyway there was no evidence on the phone just a few embarrassing photos.

> Also I just so happened to catch a bad case of indigestion right as the police knocked down my door. There were absolutely no drugs in the toilet when I flushed it.

To repeat, for emphasis: the law does not care about "I'm not touching you"-level rationalizations.

> The response described is what everyone ought to do without exception when asked for their PIN by law enforcement. The right not to be forced to provide access to your own devices needs to be staunchly defended.

If by "response described" you mean "remain silent," then absolutely. Not 100% clear from your last paragraph.

> A factory reset done in anticipation of a search is not as different from using a duress feature as you believe it is. Forensics software would have clearly identified the device was recently factory reset. It would provide another defense argument by arguing it was wiped for another reason, but whether that would be believed by a court is unknown.

This is where machine-like personal consistency is deeply important to opsec. Not just for technical reasons, but for legal reasons also.

If you wipe your phone before every national border crossing and restore it after every border crossing as a part of your standard procedure for travel, it will be much harder for a prosecutor to argue that you have ever done so in anticipation of some particular search. If your employer's security policy mandates doing so, that is even more evidence against anticipating a specific search. If a large body of infosec/cybersecurity professionals advise a wipe/reset of your phone before any and all international travel, that is further evidence for defense.

I'd generally agree that the police shouldn't be able to lie to a detained suspect, or at least that such deception should be more constrained than it is in the context but, in general, the law as practiced hinges on the question, "does police deception render a detained suspect's confession or statements involuntary?"

If yes, then those confessions or statements could become inadmissable (i.e. the suspect was effectively compelled to testify against himself, in violation of the 5th Amendment). When police combine deception with coercion or threats, the answer to that question leans toward yes, but it will depend on the juristiction and the judge and is often assessed on a case-by-case basis. Police in any given juristiction probably have a reasonably good sense (from their own accumulated experience and local training) of what kind of deception results in the produced evidence becoming inadmissable in the courts where the defendant will appear.

> When a house is raided and someone flushes a toilet, can courts assume they flushed drugs, or does that have to be proven?

Game it out at the systems level. If the prosecution had to prove beyond the shadow of a doubt that what was destroyed would have been evidence, then destruction of evidence would be functionally permissible whenever done competently.

> ...either way this definitely is not the "sky is falling", "we're almost a police state" case that folks here are making it out to be. It's a very narrow and novel line.

Agreed. And activists should generally work with lawyers in order to be familiar with areas of the law they are likely to encounter in the course of their activism. Especially activists who travel internationally.

> As far as federal overreach of power goes, that's pretty inexcusable.

Selectively detaining this guy, likely overreach.

Trying to get his phone's unlock code to go on a fishing expedition for whatever they can find, absolutely overreach.

But this guy catching felony charges for giving federal border police a duress PIN to wipe his phone when they asked for an unlock PIN? Not actually overreach.

You have the right to remain silent, not to lie to the police when detained (18 USC § 1001, and many state-level laws to the same effect also exist). Our justice system could not function if people had a right to lie to the police. Once you are detained, whether or not that detention is eventually determined to be lawful, destroying or disposing of your possessions to prevent police from accessing them is also generally a crime.

You fight abuses later, in court. Or, if you're going to use a duress PIN in the moment, you accept the legal consequences.

And yes, giving a duress PIN to wipe a phone when asked for an unlock PIN is a lie which may result in destruction of evidence. And no, the law does not care about "I'm not touching you"-level rationalizations of whether something is a lie or whether it was technically the police who entered the code to wipe the phone. Proximate cause is a longstanding legal concept.

Each one is a bit funny in isolation, and it's also true that the most common feedback is always unhappy feedback, but every company looking for another vein to tap for sustenance hopes observers will write off their particular incision as hyper-specific. Besides providing a place to check before doing business with a company, the Consumer Rights Wiki is about making the cumulative effect of a thousand hyper-specific cuts harder to ignore.

The use of "bug" for engineering defect (a use that predates software by around a century) shares Middle English etymology with bugbear and bugaboo, both of which relate to evil spirits/creatures and annoyances. Khrafstar is not unreasonable as a translation.

Latin in school from 6th grade onward. And a classical education with many holes and gaps that I am continually filling in.

> Migrating to small-scale local LLMs would be a significant improvement over using data centers.

This shift is probably inevitable, but it will vary significantly by region depending on prices for electricity. Look at, for example, the difference in fundamental homelab build recommendations between Germans and just about anyone else. Electricity prices in Germany are so high that even a now expensive Raspberry Pi or other ARM board is often preferred over Intel/AMD builds due to low power draw (especially low idle power draw), an effect that adds up for a machine running all the time over years.

With local LLMs and the GPUs to run it, especially if you want a model available to you all the time and can remote into your local network to use it whenever you want, there's no escaping much higher power draws, even at idle. Wherever electricity is expensive, the electric bill can be a prohibitive barrier.

> Unless your needs are simple, you can relocate somewhere to live very cheaply, and you don't have any dependents.

As a human, your needs are simple[1]. With a $1M net worth, you can relocate somewhere to live very cheaply. And your dependents can live in that same place, also cheaply.

All you need to do is show the Joneses the door out of your mind and never let them back in, nor look toward their lives in envy.

[1]: Healthcare being a notable potential exception, depending on the individual.

> clauses against hosting your own website on residential providers

If their standard residential Internet packages don't offer truly static IP addresses, then it at least makes nominal sense to say: if you want to host your own website from your home, you pay some annual fee for a static IP that is not part of normal residential provision of service. But that can also quickly twist into charging exorbitant rates or pushing people hosting personal websites on their own hardware to pay for business plans they don't truly need.