avidiax

The copyright owner must make new copies of the work available; the price must be no greater than the original price (not inflation adjusted). And if they fail to do so, anyone may produce copies and escrow the original price (less the cost of production) for collection by the copyright holder.

That means that orphan works are effectively in the public domain. Calculus professors can ask students to get the cheaper 2nd edition, not the latest 22nd edition. And a company like Google could make scanned works available in their entirety for a small amount of money for each work. And the copyright holder still gets their end, without having to arrange a printing or hold stock.

* LLM Chatbots are currently perceived as a safe space where suicidal thoughts can be discussed, without fear that a welfare check and involuntary commitment is going to be triggered. Attach liability to this and the safest/easiest thing for the AI companies to do is always refuse the prompt if suicide is discussed. If they are required to inform the authorities, your chat account will need location services.

* Involuntary holds are a good solution to some kinds of suicide risk, particularly those that are either acute or treatable with medication. But they actually make some patients worse off by destroying trust, disturbing life patterns, and saddling with debt.

* Mental health care is famously expensive, and not so accessible. LLMs are available 24/7 and are cheap. There's both a huge danger to unqualified advice and opportunity here for accessibility.

* Even if we claim that LLMs should not act as a therapist, they still need training/evaluation on mental issues. AI psychosis will keep happening unless we start training and evaluating AI on intervening appropriately.

* There are probably more subtle forms of AI psychosis that are not being detected yet. AI sycophancy plus lack of context means AI can be highly reinforcing in areas in which someone would actually benefit from being challenged.

1. Underpriced AI products (tokens too cheap)

2. Bottlenecked AI spend (can't get enough chips, can't build enough DCs)

So it's possible that prices will rise until demand meets the bottlenecked production, potentially with no trough or collapse (excepting that part of the financial model that is demanding outsize growth).

That sensor would be much better if it was integrated with the camera and LIDAR, so that you could "paint" a scene with thermal data by moving the phone around.

There are fixed or per-vehicle costs:

* signage, lighting, snow removal, etc.

* weather-related wear and tear from freeze/thaw, thermal expansion, etc.

* capacity (more lanes needed, driven by passenger vehicles, not trucks)

So yeah, semi's should probably be paying $100k/year in road tax based on wear, but that doesn't meant that your car's road tax should be $1/year.

There's also a great argument that all the suburbs are not paying enough in property taxes. Those paved roads have high costs to maintain over time, and service just the traffic to/from some large houses. Some have gone so far as to call suburbia a scam, on the basis that it has to perpetually grow to pay for maintenance.

Road damage scales with axle weight to the fourth power (w^4).

So if a 500lbs motorcycle is the unit of damage, a 4,000 lbs sedan does 4,096 units of damage (4096x as much), a 6,000 lbs SUV/pickup does 20,736x, a 14,000 lbs delivery van does 614,000x, and the 80,000 lbs US federal limit 5 axle semi-trailer does 90,800,000x the damage of the motorcycle.

So probably shipping stuff in trucks doesn't cost enough, and SUVs/pickups should cost 5x for road tax what sedans do.

A velocipede at best saves you a bit of energy going flat or downhill, with relatively little increase in speed, with a relatively high capital expense. Compared with walking and horses, it doesn't help much.

You have to get all the way to the safety bicycle (with pneumatic tires) before you have something that beats walking for someone that is not a cycling enthusiast.

It's a bit like the question of why the steam engine wasn't invented earlier. It was invented to pump water out of coal mines. Being 2% efficient (for the original model) didn't matter, since it was only used to mine its own fuel. I don't think we'd ever get to modern steam and later internal combustion engines without coal deposits making the initial prototype feasible, and then funding continual improvements.

By that sort of logic, we are lucky that European aristocrats were bored and enjoyed scary proto-bicycles, or we'd still not have them today. The final safety bicycle is practically a Segway by comparison with four wheel carriages.

Unfortunately for most engineers, their entire lives and schooling have never exposed them to the sorts of problems being discussed, namely, problems where there is insufficient information coupled with limited ability to gather more, or where some of the information you think you know is incorrect, or where you are misinterpreting the information you have, etc.

These kinds of problems require you to go over everything multiple times. Really consider what you actually know. Doubt yourself. Make educated assumptions. Squeeze everything you've observed dry, even the seemingly meaningless, random, or ordinary. And then you'll get a theory, that's hopefully testable, and you'll test it.

It is a problem to have a full SSD from a performance perspective. Modern SSDs use pSLC (psuedo-single-level-cells) because they can write much faster. When you write to an SSD, you generally write in pSLC mode first, and once the data goes cold, it will get written back to MLC/TLC storage.

Once you are very full, pSLC gets disabled or becomes less effective, affecting write speeds.

There is also an effect known as write amplification that is exacerbated when a disk is nearly full. You want to write 4kb, but the available erasure zones are all full (either valid or invalidated sectors), so now the SSD has to erase a 4MB zone and compact the data there (along with your new 4kb) to write your 4kb.

This gets even worse when you have a filesystem that isn't SSD optimized, such as EXT4. It will cause further write amplication.

Keep your SSD's <80% full if you want good performance.

This isn't like a hybrid car. It's a parallel hybrid, where the gas engine is just big enough for efficient cruise at altitude, and the electric motor/generator provides extra power for takeoff and ascent (or go-around power), and then charges slowly during cruise if needed.

This means that the battery is quite small and light, having only enough charge to take off and get to altitude.

I suspect that this system probably improves safety as well, if architected properly. If one or both of the gas engines fail, so long as they are not seized, that electric motor can still provide some power for diversion.

> It also ensured pressure from governments and plaintiffs, including CSAM victims, who preferred Apple’s more interventionist approaches, which Apple had voluntarily demonstrated it was willing to do.

I feel that Apple open pandora's box with the client-side scanning. It proved that it was technically feasible, and was "privacy preserving". I use scare quotes there because I don't think that political or religious dissidents would find that the same or similar technology used to discover and persecute them is "privacy preserving". And that's really the problem with Apple here. They provided a model for scanning for any kind of message or material while purportedly maintaining privacy.

I wasn't planning to see Dune 3, but that preview made me want to see it.

Can we monitor seat availability on Dune 3 70mm as well? Maybe a beta? I think right now it's just the 4 days of early release tickets, all 100% sold out of course.

Gemini strongly disrecommends buying this product[1], but it's not clear if the opinion is based on Dr. Gough's review in-part.

This, of course, leads to another arms race where listings like this will be "AI optimized". They may have prompt injections, or simply specifically claim that issues like backfeeding are resolved when they are not.

And it also makes the AI companies an arbiter of what products are marketable.

[1] https://share.gemini.google/q2ujmGU7UgeS

ML-KEM -- Module-Lattice-Based Key-Encapsulation Mechanism

ML-DSA -- Module-Lattice-Based Digital Signature Algorithm

solo PQ -- Using post-quantum crypto on its own

ECC+PQ -- Using post-quantum crypto as a layer on top of traditional elliptical curve cryptography (ECC)

So what's at stake here, is that the PQ crypto is not proven yet, and had recent implementation vulnerabilities (Kyberslash 1 & 2).

In the NSA's defense, combining cryptosystems also creates attack surfaces, timing problems, additional complexity, etc. Perhaps they know something we don't. They have sometimes acted to strengthen public cryptography, as with the DES S-boxes and differential cryptanalysis. Of course, they also weakened the key-space...

Good to have stories like this where it enables something that might be impossible or impractical otherwise.

I do think the author has a good understanding of software requirements, which is hopefully something the AI will take over eventually.

The cities are paying exorbitant prices for electricity to pay for safer infrastructure for rural customers (undergrounding).

Some cities have divested from PG&E and enjoy much lower electricity prices as a result.

https://archive.is/yFZjd

----

The complaint here isn't from SpaceX investors. It's from retail investors that are being forced to buy SpaceX stock on an accelerated timeline as part of the ETFs they likely purchased so that they would not be overly exposed to volatile single stock picks. The article isn't explicit on this point, instead vaguely gesturing to the series of mega IPOs coming and pointing out that retail clients need to plan for this.

There is a game being played here where the various indexes (NASDAQ, NYSE, etc.) are trying to sweeten the deal to attract big entries like SpaceX (and later OpenAI, Anthropic). The sugar they are giving is an accelerated timeline and inflated spot in the index (3x float rule). That sugar is paid for by retail investors, who may get squeezed when their index funds pay a high price for a small float of public shares, all on predictable days.

Before you say that retail investors should simply buy SpaceX themselves prior to the 15-day index inclusion, realize that retail investors also don't have access to shares at the IPO offer price. That benefit is reserved for large investors, private equity, etc. While it is possible that some retail investors will take this gamble and win, many will be taking a large risk.

To put it the other way, work that is distasteful in some way, should also pay more, but this is missing the point.

I think the point of the unionization is that the monopsony of a small number of AAA game studios gives them excessive market power to reduce compensation and especially to reduce working conditions.

A union can acquiesce to the passion tax and say that top developers at a AAA should make $150k/year (a bit low), while simultaneously saying that that developer should be able to see their children on nights and weekends. The project management that leads to "perma-crunch" is something that ought to be resolved on the employer's side, not by the employees.

If you give an insta-pot a 6 ft cord, someone will drape it off the counter and a child will pull it.

UL standards actually limit cable length for many appliances.

https://www.intertek.com/standards-updates/ul-1026-electrica...

Is it easier to hold back talented students with a low bar or push untalented ones to a higher bar?

That's the source of hallucinations. A path can be found between A and B, even if A is the 12th century Chinese royal court and B is the Easter bunny.

Interpolation and rote knowledge are still very useful. Most cognitive tasks are like this.

The thing that LLMs are not presently good at is extrapolation. You can train an LLM on pre-1904 literature, but you won't get special relativity from it, at least not without a human to prompt it in just the right way.

You can have an LLM provide a "novel math proof", but you are necessarily discarding 100 or 1,000 "novel math mistakes". The process is more like a guided walk (like the A* algorithm), with human supervision and intervention, not an autonomous math genius.

"They" are, of course, working on it. But the present implementation has some severe structural limitations (such as an inability for new or discovered information to affect model weights) that make LLMs as a human replacement incomplete.