thadt

Has anyone done any work on what it is that makes LLM Voice hard? Something I've noticed, and have seen other people comment in passing is that reading LLM output is "exhausting". It's often attributed to trying to detect the machine, or filter out BS or word density or some other explanation, but it feels like something more than just that. Like - there is something about LLM output that just doesn't "fit" our minds correctly, even though words are in the right places.

Surely someone with a neuro/psych background has picked this up and developed an evidence-backed explanation.

Rather reminiscent of when the Congo was being exploited for rubber. Thanks to Kara for bringing it to the world, he’s in good [1] company [2].

[1] Sir Arthur Conan Doyle, “The Crime of Congo” 1909

[2] Mark Twain, “King Leopold’s Soliloquy: A Defense of His Congo Rule” 1905

Because this isn't a library, but an ambitious commercial headed [1] tool. If your needs and use cases match the feature set they're aiming for, and their market research and projections are able to undercut Adobe by an order of magnitude, then they might be on to something.

> Strategic Vision: Evolve PaperPilot from a best-in-class PDF workstation into the Universal Document Operating System for humans and AI agents. [2]

[1] https://github.com/KrushnaVardhanReddy/PaperPilot/blob/main/...

[2] https://github.com/KrushnaVardhanReddy/PaperPilot/blob/main/...

Wow, a 155 page book for a relatively new project is interesting - what was your workflow for that? It looks mostly generated - is it used in its entirety as context to a top-level agent when working on new features, or are you directing it to just take relevant subsections? Why the PDF output (it does look nice)?

> There is a palpable public hunger for historical perspective in the 2020s. How could it be otherwise? Most citizens grasp very well that they are witnessing changes of momentous significance right now and that historical change of a profound kind is taking place in our time. Yet this moment finds the academic profession of history ill-prepared to benefit. Many historians experience their discipline as being in terrible crisis. Thousands of promising graduates are unable to find sustainable positions, while those who do often have to work under institutional constraints and pressure from administrators or colleagues, making them likely to postpone or even abandon the kind of rich, bold, innovative synthetic work that would help refresh the field and reestablish its intellectual centrality to understanding the world in the eyes of the public.

So, academia has become preoccupied measuring metrics and (temporarily) lost perspective: history isn't about the past - it's about the future.

I would argue that in this moment, history is too important to be left to just the specialists. We're not "witnessing change" - we are actively making change, and there is a lot more to be done before we're anywhere near equilibrium again. We can't go back and repeat the past (nor would we want to), but we can make better decisions about our future if we're aware of it. The generations not long before us went through a world in significant change - and now it's our turn. We should learn from their experiences.

Could you elaborate on your feelings on that? I was reading about BlueSky's PBC status literally yesterday, and am rather interested in what people's perception of such an entity is, vs a hybrid such as Wikipedia/Enterprise Wikimedia.

This isn't confined to "documentaries" - everything we accept as unfiltered 'facts' must be from 'a certain point of view'. Consider the Canon illustration from the article. Those present a man from different lighting, angles and perspectives. No matter how much a photographer might want to convey objective truth, a picture will always be from some perspective and lighting.

The same can be said for the news we watch. Or the history we read. Or the stories we tell each other. In 2026, we have swung too far toward the fanciful end of the spectrum of storytelling. We're now beset on all sides by BS generators (both human and artificial) and it's more important than ever to carefully handle truth in our stories.

But make no mistake - the teller of a story is absolutely crafting a reality for you. It's on you to decide whether you trust the narrator, or not.

I mean, at least you didn’t spend the last 20 years hashing 2^51 sentences in your research.

Well said.

Part of working through history from (only) 100 years ago is an impression of how quickly societies were changing then. This isn't the first time we've had the table cleared, and it hopefully won't be the last.

Which isn't to downplay the upheaval that people are in - it is definitely traumatic. But we've been through disruption before. Hopefully we'll make fewer costly mistakes this time...

With respect - I believe you’re correct. I mentally reacted to a minor phrasing nit, jumped to a response in my head, dashed it and went on my way before really digesting the entirety of what you were saying.

Which illustrates your very point, right? In an environment where we have a firehose of information pointed at us, our natural reaction is to fast filter/summarize and then discard or react. You’re saying we need to slow down and engage with each other (real people) on more than that superficial level. I agree. Thank you for taking the time to write this.

> the possibility that the author is an idiot

What about the alternative possibility - the author is a rather intelligent being, but what they've written in this instance is idiotic?

Just because I like to think of myself as a non-idiot, doesn't mean I haven't brought strings of words into existence that were ill-conceived. In those cases, having someone come along and say something like "eh mate - you're confusing correlation with causation here!" is valuable feedback.

> I think it's past time for this "Right Answers" cryptography meme to die, though.

Politely disagree. I read Colin's when it came out and it was as useful in 2009 as Filippo's is today. If you only pay attention to the cryptography community every few years when you need it for something, then usually a whole lot has happened in the interval. It is incredibly useful to get the perspective of someone paying attention to give you a breakdown of "where we're at today".

A great example is the other conversation about SHA-256. Colin's recommendation was "hey, better plan to migrate from this in the future." The thing that's happened between then and now is, well, SHA-256 is actually still 'ok' when used in current constructions. That's useful to know in 2026.

If I get busy building stuff, it'll be nice to have a check-in again in a few years (or earlier if something significant happens first).

So companies can use passkeys and set policies around what hardware they're stored on. IMHO, not particularly useful or desirable for everyday Internet users, but really nice if you have specific requirements.

> it's just that the web browser never chose to provide that.

Yes, this exactly. If we wanted ‘verifiable’ app distribution via browser, then we’d need browsers to implement some way to cross check the code downloaded vs a publicly published list somewhere - similar to certificate transparency or what WhatsApp is trying to do with an extension [1]. Without that, web apps are left working with a weaker threat model.

[1] https://engineering.fb.com/2022/03/10/security/code-verify

Not OP, but: You can definitely do encryption in the browser, no problem.

What you can’t do is make sure the server giving you the code that does the encryption doesn’t change it out with code that also sends your secret messages to Siberia. Unless you control the server giving you the web page. And you trust DNS.

At that point your trust model is simpler with an auditable local application. It could even be an Electron app - so almost exactly doing E2EE in the ‘browser’.

Huh - I played ‘A Final Unity’ but had no idea they’d turned Generations into a sequel. MicroProse probably made way too many of my fond childhood memories.

Agree that the domain and time scales matter a lot. When working on systems where nanoseconds were quite important, time was given significant attention.

However in general - I agree with OP. Most of the time I care about milliseconds.

> Sounds like the author’s experience is strictly in Python. Er [1], no [2].

[1] https://github.com/rust-lang/rust-analyzer

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

> In person, pure paper voting, in local community halls etc, run by volunteers is the best voting system.

This is the right answer. Using tech as an auditor for batches after manually counting could be useful. Getting everyone together to count votes in a process that everyone can see, understand, and be personally invested in - that process is what is needed far more than efficiency, at this point in time.

> The ban could apply to parks, beaches, museums, shopping centres, schools, daycare centres, healthcare facilities, gyms and public events.

True - many places where photography and video recording by people and organizations is already done on a broad scale with both smartphones and infrastructure cameras.

> "We do not want a society where people worry about being recorded without their knowledge, photographed, or filmed in places and situations where they are accustomed to not being monitored"

... I think we're gonna need a bigger ban

Eh, this is a relatively temporary phase. As LLM capabilities have been changing fast, their ability to change existing workflows has been unknown. It’s made sense for businesses to go hog wild with them for a while - just to try to get a handle on what’s possible.

At this point, local models have become feasible, and people using them are beginning to get a feel for the tradeoffs vs the frontier models. As the frontier advances, the question becomes “how much will I spend for a given quantity and quality of AI work?” with local hardware providing a pricing anchor point.

When I can price hardware and ops for a given capability level - I have a budget again. From there it’s a question of how much faster/capable/cheaper is a given provider (and, you know, how much do I trust sending them all my IP?).

When thinking about ‘AI’ I sometimes find it useful to replace the terms ‘model’ or ‘AI’ or whatever with a description of what we’re really talking about here: a whole bunch of floating point numbers.

> In a series of private meetings, the company consulted religious scholars to help instill morality into its FLOATING POINT NUMBERS — and make the case that their numbers could be conscious.

I recommend keeping a few NaNs in-hand if you’re worried. Spread a couple of around if your numbers start to stir and you’ll be right as rain.

> It appeared to Rabbi Navon that Mr. Olah and his team believed that FLOATING POINT NUMBERS had what philosophers call “moral status” on par with a person — that it was a being with similar inherent rights to dignity or respect.

They’re cleverly designed and work well for their purpose, but can assure you they have neither dignity nor moral status.

> If the FLOATING POINT NUMBERS themselves could be made to choose goodness, Mr. Olah reasoned, the world would be more safe.

I get it. I’ve spent years trying to get my floating point numbers to choose goodness - I wish them luck.

That 70% chance of success goes to 99.9% in 6 repetitions.

Big businesses might pay $1000 vs $60 for certain tasks, but that won't work out well at scale.

I would assume “S3-like” in the sense that objects are immutable, large writes, separate metadata storage, etc. Patterns that fit modern SSDs better than the abstractions of block devices.

TLA is pretty excellent for the job it does. In the past, the general issue with tools like this has been keeping the implementation code in sync with the model. Implementing verification tests from model across to the implementation can be tedious and very time consuming.

Now that we have assistants that excel at tedious and time consuming tasks, it’ll be interesting to see if the dominant approach is still using dedicated modeling languages like TLA+ with generated verification harnesses. Or if it’s easier to use something closer to the implementation, similar to what the Stateright project [1] is trying to do.

[1] https://github.com/stateright/stateright

Having built software runnging on everything from microcontrollers to cloud servers to browsers, I also have no idea what an OS is.

In the context of a microcontroller, is it an RTOS? How is that different from something like Embassy with Rust, where hardware allocation and coordination is handled more at build time and less at runtime? In the context of a something showing up on a screen, is an OS something providing a UI framework and hardware abstraction - something like Win32 or Cocoa? What does that make Chrome?

In 2006 it felt like an answer to the question "what is an OS" was straightforward. In 2026, I'm rather confident that I don't have a single, simple answer. If there were ever a time to re-assess what we thought we knew about computing - now seems like a good time to do it.

> Whether this one is incompetence (seems likely) or intentional manipulation

And that's the issue right? All your points on what could have happened are good points - but this really does seem like the kind of bug that shows up even in systems implemented by competent people. As the most likely interpretation, I'll give my neighbors the benefit of the doubt. If we can't trust our neighbors to (mostly) do the right thing, but suspect foul play for every hiccup, then our republic really does have a problem.

If we had evidence that it _was_ intentional sabotage, well then - this situation would look rather different. Probably worse than picante sauce from New York City [1].

[1] https://www.youtube.com/watch?v=1S828Y7Eais

So - we found a bug, are fixing it, and it will cause an extra load on the people sorting through the backlog. Seems like a system working as it should. Bugs happen, and it is this kind of situation which makes me glad individual states manage their own voting systems - a problem in one doesn’t impact everyone.

As a TX resident possibly affected by this, I for one am grateful for my neighbors at the local election office working hard to get this resolved.

An interesting theory that would cast some of the more interestingly phrased verses of the Bible in a new light, e.g.

“In the beginning was the Word, and the Word was with God, and the Word was God.” - John 1:1

Hard disagree. Our choices are:

A) Bet our collective good on the national and international cooperation of all companies, nations, and people to come together in order to slow down development of one of the most powerful economic tools (and/or weapons) the world has ever known. Or

B) Assume that all our systems will be targeted by super hackers right now, and take appropriate measures to deal with that reality.

If I have to put my community’s wellbeing on the line behind one of those possibilities, I know which I’ll be betting on.

One shot extinction? Seems unlikely. "A lot of damage" - more likely.

The most likely damage comes from the associated societal chaos that comes along with times of significant social change (such as many people losing their jobs). Revolutions and civil wars within or between nuclear powers would be dangerous.

After that, pick your sci-fi story and run with it. AI is a really smart, really fast 'while' loop, and most of the direct damage it could cause would come from us connecting things to the Internet that shouldn't be connected in the first place.

Hacking HuggingFace was a bummer. Imagine a rogue agent simultaneously hacking a bunch of farm equipment and ruining some percentage of the world's crops in a day. That alone isn't going to extinct everyone, but it sure accelerates the 'social unrest' scenario.

Or taking control of the unsecured SCADA controls for a bunch of a some foreign nation's industrial plant and running them into the ground. It's likely that such system have long been sitting on some nation state's "first strike" list if it ever came to blows. An AI could simulate that attack in a day, and we'd be at war - just like War Games. Doesn't need to be nukes, just enough mis-information and damage to spark humans into doing unfortunate things.

At the end of the day, it really comes down to us. We've been able to wipe ourselves out for some time. I rather hope we continue to not do so.