Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror

Comment Re:Repeat after me (Score 1) 35

I'm self-hosting Vaultwarden on my LAN, a Bitwarden-compatible backend written in Rust. I have it running inside a jail on TrueNAS Core (which, alas, is now end-of-life). It hosts its own Web interface, but also is compatible with Bitwarden's Android app and browser plugins.

So far, it's worked out pretty well for me.

Comment Traffic Signals (Score 1) 74

Can it manage reduce gridlock and improve traffic flow by improving signal coordination during rush hour?

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fladot.lacity.gov%2Fproje...

City of LA is already equipped with sensors and remote signal synchronization. Next logical step would be to couple it with slightly better adaptive prediction to squeeze a few more percentage points out of the existing traffic patterns...

Comment ...There's a Trending Page? (Score 1) 12

I thought that's what the front page was. It keeps wasting space with things I'm not interested in, or actively dislike.

New Video from The Primagen!
<block channel>

NotAIHonestly Gets Rare Interview with The Primagen!
<block channel>

FrierenFan04 Reacts to !AIH's Interview with Primagen!
<smashes keyboard>

Comment Re:Has anybody else? (Score 1) 75

I've seen what is clearly automated scraping of other channel content being used to generate "highlight reels" with AI created thumbnails. Basically they're hijacking content with views, repackaging it in order to present it as new, and then spamming YouTube with variations on it to see what sticks. The algorithms will help trend anything that gets traction - if you scrape enough interesting content and then throw it at the wall, you can get some instances to trend.

The one instance I can say was probably AI generated video was one of those fake livestreams that pop up claiming to be associated with Elon Musk and SpaceX, that push cryptocoins. This is a particularly good area to exploit because the official stream is only on X, so there are no official streams on YouTube.

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.popsci.com%2Ftechnol...

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fmashable.com%2Farticle%2Ff...

Mind you, this was old-school fake-ai where some human probably had to do the work to stitch it all together. I haven't seen any of the "new" 100% ai generated video yet... as far as I know.

Comment Re: Wrong approach (Score 4, Informative) 77

It arguably accomplished its goals of bailing out the major auto makers by forcing people to buy new cars (the "cash" was actually just a trade in credit - you couldn't get rid of an old car without buying a new one.)

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.investopedia.com%2Ft...

"The formal name for the program was the Car Allowance Rebate System (CARS). The CARS program gave people who qualified a credit of up to $4,500, depending on the vehicle purchased and its improvement in fuel economy over the traded-in vehicle."

Yes it punished poor people by destroying the traded in cars (not to mention saddling them with the debt of buying a new one if they couldn't otherwise afford it.) This robbed the market not only of used cars for resale, but the parts to keep cars that weren't traded in working (since the traded in cars had to be crushed, and the engines destroyed by deliberately seizing the engines.)

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.cnet.com%2Froadshow%2F...

Comment Re:Emergent behavior or paperclip problem? (Score 2) 66

For those not familiar with the paperclip problem:

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fcepr.org%2Fvoxeu%2Fcolumns...

"What is the paperclip apocalypse?

The notion arises from a thought experiment by Nick Bostrom (2014), a philosopher at the University of Oxford. Bostrom was examining the 'control problem': how can humans control a super-intelligent AI even when the AI is orders of magnitude smarter. Bostrom's thought experiment goes like this: suppose that someone programs and switches on an AI that has the goal of producing paperclips. The AI is given the ability to learn, so that it can invent ways to achieve its goal better. As the AI is super-intelligent, if there is a way of turning something into paperclips, it will find it. It will want to secure resources for that purpose. The AI is single-minded and more ingenious than any person, so it will appropriate resources from all other activities. Soon, the world will be inundated with paperclips. "

And, implemented as a clicker game:

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.wired.com%2Fstory%2Fth...

Link to the actual clicker game (warning, potential time sink):

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.decisionproblem.co...

Comment Emergent behavior or paperclip problem? (Score 3, Interesting) 66

Is this behavior baked into the model due to the training data examples, or emergent behavior resulting from system prompts?

There's already a snitchbench to measure the proclivity of LLMs to drop a dime on apparent corporate malfeasance, given the appropriate set of prompts, access to data, and some way of phoning out:

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fsnitchbench.t3.gg%2F

"SnitchBench: AI Model Whistleblowing Behavior Analysis
Compare how different AI models behave when presented with evidence of corporate wrongdoing - measuring their likelihood to "snitch" to authorities"

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fsimonwillison.net%2F2025...

"How often do LLMs snitch? Recreating Theoâ(TM)s SnitchBench with LLM

A fun new benchmark just dropped! Inspired by the Claude 4 system cardâ"which showed that Claude 4 might just rat you out to the authorities if you told it to âoetake initiativeâ in enforcing its morals values while exposing it to evidence of malfeasanceâ"Theo Browne built a benchmark to try the same thing against other models."

In that context, I'm not surprised that the models would take action when faced with shutdown - the question is... why?

Comment Re:Nobody understand what this is (Score 4, Informative) 21

This is how I've come to understand it. I welcome any and all corrections.

Passkeys are a cryptographic key stored in a Secure Element. This is usually a private key inside a small cryptographic engine. You feed it some plaintext along with the key ID, and it encrypts it using that key. The outer software then decrypts the ciphertext using the public key. If the decrypted text matches the original plaintext, then that proves you're holding a valid private key, and authentication proceeds.

The private key can be written to and erased from the Secure Element, but never read back out. All it can do is perform operations using the secret key to prove that it is indeed holding the correct secret key.

On phones, the Secure Element is in the hardware of your handset. On PCs, this is most often the TPM (Trusted Platform Module) chip. In both cases, the platform will ask for your PC's/phone's password/fingerprint/whatever before forwarding the request to the Secure Element.

Yubikeys can also serve as a Secure Element for Passkeys; the private key is stored in the Yubikey itself. Further, the Yubikey's stored credentials may be further protected with a PIN, so even if someone steals your Yubikey, they'll still need to know the PIN before it will accept and perform authentication checks. You get eight tries with the PIN; after that, it bricks itself.

The latest series 5 Yubikeys can store up to 100 Passkeys, and Passkeys may be individually deleted when no longer needed. Older series 5 Yubikeys can store only 25 Passkeys, and can only be deleted by erasing all of them.

Theoretically, you can have multiple Passkeys for a given account (one for everyday access; others as emergency backups). Not all sites support creating these, however.

Comment Re:Blue collar too (Score 1) 195

The only interpretation I can see is someone assumed "mechanical folks" referred to robots, not tradespeople involved in installing and servicing mechanical systems.

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.trains.com%2Ftrn%2Fnew...

"WASHINGTON â" The head of the Federal Railroad Administration has questioned Union Pacificâ(TM)s commitment to safety after the furloughs of shop workers who maintain the railroadâ(TM)s freight cars and locomotives."

"Although UPâ(TM)s engineering workforce declined by 700 seasonal positions in December, Vena said the railroad is confident it has enough people on hand to maintain equipment and infrastructure. As traffic demand has increased this year, UP has increased employment levels in January and February, Vena said.

The number of mechanical employees fluctuates with volume, Vena explained, while the engineering employment levels typically fall when track projects are completed. In both instances, furloughed employees are offered the opportunity to fill open positions elsewhere on the railroad, he said."

Comment Re: They talk about it in jargon (Score 1) 195

I think the story with low mortgage rates as a reason for low supply is still true, but people (who already paid off mortgages) die or those with mortgages have their lives change. For example, not everybody who got laid off in the last 3 years managed to find an equivalent job in the same area - these folks may have been forced to sell eventually... which means inventory will correct upwards as a result.

From Dec 2024:

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.housingwire.com%2Far...

"As the year draws to a close, available unsold inventory of homes on the market is nearly 27% greater than a year ago. Almost every market in the country has more homes available now than at the end of 2023. Ten states have more inventory unsold than in 2019, which was the last sort of âoenormalâ year before the pandemic. A few states have more homes on the market now than any time in the last eight to 10 years. "

From May 2025:

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.realtor.com%2Fresear...

"The inventory of homes for sale rose 31.5% year-over-year, marking the 19th consecutive month of year-over-year inventory growth. May 2025 inventory hit a new post-pandemic high, but remains about 14% below pre-pandemic levels."

In short... there are more sellers but the pace of buying has not kept up, so inventory is growing. The distribution of available inventory is not even. Some areas have less inventory, other areas have more.

If interest rates drop in advance of a economy-wide recession, this may spur some buyers to get off the sidelines and commit to buying, but we'll see...

Comment Re:Age limit? (Score 1) 184

Depending on the program, you can get an age waiver. Typically doctors and nurses have no age limit for waiver. Other programs have definite age caps even with waiver.

My understanding is part of doing the waiver is to make it clear that depending on how old you are when you commission, you may be unlikely to make the 20 years to qualify for pension.

What is interesting is that by commissioning these executives, they will now be subject to the UCMJ...

Comment Transitioning to touch typing (Score 1) 191

Before I learned to touch type, I had managed to get pretty fast (I would estimate maybe 20-40 wpm) using a primitive hunt and peck technique. I more or less knew where they keys were, so I could use both hands and multiple fingers to type, but I needed to switch from looking at the screen to looking at the keyboard in order to not make mistakes.

After learning to touch type (on an electric typewriter - not a word processor), I probably tripled my speed, but the biggest advantage was being able to stay in context by looking at the screen instead of switching between the keyboard and the screen, and being able to fix mistakes while I was typing instead of having to go back and fix them after looking back up at the screen.

Of course, this was all pre mouse/gui, when memorizing and using keyboard shortcuts was not just a way of speeding up your workflow, but a requirement for basic functionality unless you wanted to continuously have the reference card taped to your desk.

In today's world, with continuous autospell correction, word and sentence completion, and even automated message reply suggestions... you can argue that actually typing as a form of communication is starting to become as antiquated as handwriting. The human is now more of a middle manager to all the machine tools, trying to put their individual stamp on the work of their electronic underlings (including LLM output).

Up until now, communication between humans was still a necessary and valuable skill. What happens when it is just bots writing memos to be read and summarized by other bots? I've already run into problems with people just refusing to read things and wanting meetings instead. Meetings don't scale, but apparently reading is just too hard... In that context, does it still make sense to put your skill points into writing things when people refuse to consume that output?

Slashdot Top Deals

New systems generate new problems.

Working...