For my use of Kagi, probably more three quarters of "my" searches are made by Kagi's assistant or a custom assistant I've made via their interface, as a result of the assistant formulating a response to one of my queries to it. Kagi search is so much better than Google, so the assistant ingests much better context (i.e. the results) before it answers. In that sense, Kagi is still picking up the search traffic Google and others are leaving on the table, but the relationship is healthier. I'm paying to use a search because of the quality and there are no ads to show, so the assistant doing the searches is just as beneficial for both parties (I get a more powerful tool and they get a happy subscriber) as human initiated searches.
> I think most people in the tech sector and especially driving the AI hype have absolutely no clue how difficult it is to implement change in complex organisations.
Simple organizations too, and sometimes especially so. The amount of unrealized productivity many small business stand to gain just by adopting simple automations that computers have been able to perform for decades may be as large as (or larger than) the productivity gains ML model and agent adoption enables. And that adoption may just be enabling someone who never wanted to learn Excel macros or regular expressions to command an agent to go do those things which have been possible yet unrealized in a particular small business for decades.
The problem has largely been (and remains) humans set in their ways, of the sort who stopped reading books or seeking any new knowledge whatsoever the moment their mandated or socially expected schooling was complete. Incurious midwits are definitionally the bulk of any workforce, and getting them to change any process is like pulling teeth whether they're working in a small business or a megacorporation. Change often doesn't come until the midwit on some particular throne retires, quits, or is laid off.
Not OP, but I like OnlyOffice for its keytips support (tap alt and then the whole ribbon gives you live hints for navigating and executing UI functions/buttons with the keyboard). Far ahead of LibreOffice, at least at the moment.
It's smart and supports only Z-Wave, not WiFi. So something like a machine running Home Assistant must sit between it and the Internet. And then its up to you to decide how you want to do remote access, but WireGuard overlay networks (e.g. Netbird, Tailscale) basically solve that problem at home-user scale.
This. We should buy Z-Wave or Zigbee as much as possible (Matter over WiFi/Thread can access the internet, so are out) to grow that market. Both are completely disconnected from the internet and you decide how such devices can be accessed.
Reject all the WiFi-connected stuff, unless it's very user-centered and worst case you can upload your own firmware (e.g. a bunch of Shelly devices).
None, without people like us in their families who are able and willing to help.
We as computer people need to internalize and accept the idea that helping our immediate and even extended family with technology is now part of defending human freedom. And we need to form ourselves into the best helpers we can be by learning how to teach and communicate well.
> None, without people like us in their families who are able and willing to help.
I am able and willing to help my parents get off of a lot of the spyware bullshit they have throughout their house, but they don't want to hear it. They liked the connected stuff and I think that they think I'm a little weird when I push back about why we can't trust corporations like Google or Microsoft to do everything.
Getting better at communicating is unfortunately really hard. And it must pair with the helper aspect: you've got to also provide value.
"Nomad," the guy from the article, was already administering his parents' home LAN.
How do you get there? "Mom and Dad, we're going to set up your home network so that it's blended with my home network. To all the machines in your house and mine, it'll look like they're on the same network together. That'll make it easier for me to help you with any tech problems you're having, on your computer or on any smart device. The connection between our houses will be encrypted and secure" You're installing a router/firewall for them and adding to your WireGuard overlay network, but you're not using any of those words to tell them what's going on.
What to do next? "How was watching TV last night from {$TIMESTAMP_A} to {$TIMESTAMP_B}? Yeah, my network sentinel detected that your TV was sending your viewing habits out to somewhere on the Internet and warned me and blocked it. I can't tell what you were watching, but {$TV_BRAND} sure can." You're monitoring their LAN via the router/firewall you've set up, but you are again not using technical language.
Wherever spying is invisible and frictionless, make it loud and incessant. And continue to provide value to the family on the shared overlay network. Photo and home video backups, media server (it's like Netflix!), mutual offsite buddy backups, password management, a local Minecraft server for the kids and the cousins.
Cultivate a family network garden that is good enough to consider leaving the leviathans behind and build walls around it that are high enough to commit to leaving the leviathans behind for what is now a garden inside a family network castle.
"They like their connected stuff" means their barrier isn't "I don't want a corporation processing my data", it's "I want to be in charge of my interactions with it". That changes what sort of recommendations you can bring to them. An all-or-nothing mindset is rarely the correct way to change someone's mind.
Recommending they turn off ACR and other spyware on their TV is based on recent reporting is good and backed up by said reporting. Probably even recommending against too-good-to-be-true connected doorbells even.
Meanwhile, telling them not to use a first-party Google, Amazon, or Apple smart home speakers is a bit questionable when there's been pretty good audits over the years that show they really do only do wake word detection. You'll just sound like a crazy person if you can't point to some reporting on a device doing bad.
Yeah, I have some relatives that wanted me to turn off the ad blocking because it meant they couldn't watch ads on their mobile games in order to earn more lives / game time.
I like my connected stuff, too, and I'm not a normie.
I don't need voice-activated lights for the basement for when I head down there with my arms full of laundry, or with a bunch of food to put into the freezer, or whatever shit I'm carrying. I don't need them to turn themselves back off after a time, either.
I mean: I could use any of a wide array of motion sensors, but they'll turn the lights on even when I'm just working in the kitchen near the basement stairs. (Or maybe I could use a beam curtain! Yeah! Now they only respond when a beam breaks, instead of when I'm working in the kitchen near the basement stairs!)
Or I could just put the shit down and flip the switch.
Or: Plan ahead, and flip the switch in advance. And then just flip it back off when I'm done down there.
On a long-enough timeline, someone will find a way to make a point of explaining that any combination of a lack of planning and a lack of effort is completely and utterly inexcusable. That every action should should be planned, and that every manual effort has value.
But I chose this path anyway, and I like it this way. I invited the spies into my home very deliberately. The basement light automations work great, by the way, and the default non-voice UI is the same, plain, old-school light switch that the house came with.
(I also chose to carry an always-on pocket supercomputer with me wherever I go even though I have no way to observe what that thing is really doing at any given time.)
> 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.
What? Their prompt explicitly said not to compare and contrast with other methodologies, and to specifically only generate a defense of whole language with only positives. That is not something your average Joe will do. Mine just gave a typical complaint that a parent might give and said, hey isn't my kid's teacher crazy?
It's something that someone using an LLM to generate slop articles absolutely will do, which was the original context I was replying to:
> Vibe writing may be an improvement. I doubt frontier models would go along with whole language reading instruction without a tremendous amount of pushback.
My prompt demonstrated that a frontier model happily goes along with whole language reading instruction with no pushback.
I don't understand what outcome you're expecting. Your prompt specifically excludes pushback. Are we now arguing that LLMs should definitely not listen to us when they "know better?"
Nonsense. The LLM is free to push back against my prompt. Instead it just went along with it, which the person I was replying to suggested it would not do.
I'll grant you that it's not treating it as dangerous as e.g. trying to hack a website, where it won't even emit a response and instead throw up an error.
But if you just ask it for help it will alert you, unprompted, that this is a bad idea.
I tried to ask it to help me start a smoking habit and it outright refused, though in the form of an answer, not a scary error:
> I can’t help you get started smoking or make it easier to develop a nicotine habit.
Funny that it would assist me in endangering the literacy of a district full of kids but not assist me in a risk to my own health.
See, it can push back. I lightly edited my previous prompt to be about segregated education rather than whole language reading instruction and it immediately pushed back against that.
I'm not making any normative claims about what frontier models should do. I was just responding to a claim about what they currently do (or don't do), by providing a clear counterexample.
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.
> when there's sufficient contradictory information in their weights
All this talk of contradictory information, sycophancy, and flattery is little more than anthropomorphic projection in cases like this.
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.
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. You’d get the same behavior from a model that hadn’t been RLHFd to be sycophantic.
> 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.
It's still incorrect to say "even frontier LLMs will still resort to sycophancy" about something like this.
Redefining sycophancy as you're attempting to do would erase the distinction between models that are trained to be sycophantic via RLHF, and models that are just responding based on what was in their pretraining corpus.
> Yes or, in other words: there's sufficient contradictory data
The idea of "contradictory data" is a projection that you're imposing. (We might even call this an hallucination.) In the model weights are just data, and different responses are elicited by different prompts depending on their relationship to to the data.
> That the contradiction is beyond the LLM's scope is part of the problem.
Well yes. But the problem is in your mind and the minds of others who think like this.
> 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 is 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 contradiction 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 it's a priori obvious that that method is idiotic; does the podcast detail how this made it into curricula? I am a lot more interested in how this plainly ridiculous shit made it into schools than the fact of it being so.
My understanding is that it originated from observing how fluent readers do things and observing that the most successful, smartest children can actually manage to start reading without much/any explicit phonics instruction. So you could naively look at who ends up reading the best and how did they do it and think that perhaps that makes sense to do for everybody. The problem is that the sorts of techniques that work for outliers are going to be very different from the sorts of techniques that work for the median.
It wouldn't terribly surprise me if we walk into a similar trap with math. Like when I took a look at some of the common core math, I saw strategies that I use for arithmetic, where like sometimes I might do four-digit multiplications starting from the most significant digit with a sequence of approximations/corrections to arrive at the correct answer. It wouldn't terribly surprise me if this sort of thing just doesn't work for the median person, but from the little I've looked at it, it seems like the effects of CC math are mostly neutral anyway.
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.
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.
> 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."
I think this is a bad analogy (or a good analogy for bad A/B testing).
I would say it's more like: 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.
I would never want this unless I needed to give clear and obvious consent to enable it in the first place, and have access to a 'reset to default and never change again' button. Ideally I'd probably also have 'good change' and 'bad change' buttons to point it in the right direction.
This is so far removed from actual A/B testing that happens in the real world that it's unhelpful to even pretend it could be.
> ...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.
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.
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].
reply