Hacker Newsnew | past | comments | ask | show | jobs | submit | altcognito's commentslogin

Entrepeneurship is not a career path.

https://personal.utdallas.edu/~otoole/CGS2301_S09/7_split_br...

See page 53. While it is absolutely more prevelant in LLMs, human brains can also want a story for why their brains do things they are't plugged into.


Part of this is that memory companies recognize that nobody is going to enforce antitrust law for the forseeable future, so collusion to raise prices is the norm now.


With body cameras this is a lawsuit.


But is it a winning one?

Qualified immunity tends to chime in.


That doesn't mean what you think it means.


No? Jessop v. City of Fresno is worth a peek.

https://law.justia.com/cases/federal/appellate-courts/ca9/17...

> The panel held that at the time of the incident, there was no clearly established law holding that officers violate the Fourth or Fourteenth Amendment when they steal property seized pursuant to a warrant. For that reason, the City Officers were entitled to qualified immunity.


And handed down in only one circuit, so the other 80% of cops in the country can say "well, in my circuit there was no established case law that said stealing the property was a constitutional violation."


That's not exactly consistent with the given scenario. Use of force issues tend to have much better case law at both the federal and state levels than property related issues.


https://www.generalservices.state.nm.us/wp-content/uploads/9...

> Corbitt v. Vickers, 929 F.3d 1304 (11th Cir. 2019): Qualified immunity granted for officer who, hunting a fugitive, ended up at the wrong house and forced six children, including two children under the age of three, to lie on the ground at gunpoint. The officer tried to shoot the family dog, but missed and shot a 10-year-old child that was lying face down, 18 inches away from the officer. The court held that there was no prior case where an officer accidentally shot a child laying on the ground while the officer was aiming at a dog.

> Young v. Borders, 850 F.3d 1274 (11th Cir. 2017): Qualified immunity granted to officers who, without a warrant, started banging on an innocent man’s door without announcing themselves in the middle of the night. When the man opened the door holding his lawfully-owned handgun, officers opened fire, killing. One dissenting judge wrote that if these actions are permitted, “then the Second and Fourth Amendments are having a very bad day in this circuit.”

> Estate of Smart v. City of Wichita, 951 F.3d 1161 (10th Cir. 2020): Qualified immunity granted for officer who heard gunshots and fired into a crowd of hundreds of people in downtown Wichita, shooting bystanders and killing an unarmed man who was trying to flee the area. The court held that the shooting was unconstitutional but there was no clearly established law that police officers could not “open fire on a fleeing person they (perhaps unreasonably) believed was armed in what they believed to be an active shooter situation.”

(And a bunch of others.)

And a matching case has to be very specific:

> Baxter v. Bracey, 751 F. App’x 869 (6th Cir. 2018): Qualified immunity granted for officers who sent a police dog to attack a man who had already surrendered and was sitting on the ground with his hands in the air. The court held that a prior case holding it unconstitutional to send a police dog after a person who surrendered by laying on the ground was not sufficiently similar to this case, involving a person who surrendered by sitting on the ground with his hands up.

"No clearly established law", my ass.


https://www.casemine.com/judgement/us/5d778b8b342cca3e584ef6...

The prior opinion in this case, found at Jessop v. City of Fresno , 918 F.3d 1031 (9th Cir. 2019), is hereby withdrawn. A superseding opinion will be filed concurrently with this order. Plaintiffs-Appellants’ petition for rehearing en banc remains pending.

I picked the second one to start. So I don't think that's a great source.


What was the outcome of the lawsuits against the agencies? You don't have to win a suit against an individual. Most of the big payouts have to come from the cities.

Here are a bunch going the other way. https://policefundingdatabase.org/explore-the-database/settl...

I never said that qualified immunity wasn't an issue, just that there tends to be more protections when use of force is involved than with property.


> Most of the big payouts have to come from the cities.

In other words, from the victimized populace.

I think a cop who steals seized evidence should be personally liable to the person they stole from.

(…and I'd note "v. City of Wichita" is clearly responsive to your question.)


I would probably say that both the city and the cop should, independently, be liable. Given the position of authority the city provides, it is ultimately responsible to hire and properly train people who will use that authority well, while the individual is also responsible for their own actions.


If the cop is following procedure, the city and others who set the procedure should be liable. If the cop is breaking procedure, then they should be liable. If there is no clear procedure, then they should both be liable.


Both is good with me, yes.


"In other words, from the victimized populace."

Sadly, yes. They're also the populace that voted for that leadership. There are many leaders of major cities that continually push policies that are highly probably to result in legal action due to their conflict with existing law and case law. I don't like it, but its true.


The city can still be liable, it's not as if there's no redress.


Yeah, the "qualified" part is relatively misleading, it makes it sound like there are clear limits to police immunity.


ask a high end LLM to do it


They who and what?


They’ll give you a small handful of examples, of which a number occurred in the UK (famously not a member of the EU), most of which were actually arrests for incitement, and of the remainder the majority were thrown out before ever going to trial, or subsequently on appeal.

Very few of the cases they present will have involved citizens being murdered in the streets by the government for exercising their absolute right to free speech.


The UK has more arrests for social media posts than any other country in the world, including authoritarian countries like Russia, Belarus, etc. Germany is the third highest. Both have thousands, not "a small handful".


Ah, the Joe Rogan school of geopolitics finally rears its HGH malformed head

{1}https://www.independent.co.uk/news/uk/home-news/wales-englan... {2}https://pa.media/blogs/fact-check/fact-check-international-d... {3}https://www.standard.co.uk/news/tommy-robinson-uk-speech-cla...

Most of the erroneous conclusions come from a cursory interpretation of a Times article from last year:

{4} https://www.thetimes.com/uk/crime/article/police-make-30-arr...

In 2023, UK police forces made around 12,000 arrests under the Communications Act 2003 and the Malicious Communications Act 1988. These laws cover sending messages that are "grossly offensive, threatening, indecent, or menacing over communications networks" (which includes social media). Prosecutions resulting tend to come from a small subset of serious crimes - stalking, incitement to hatred, endangering minors etc...

This was gleefully misinterpreted by Musk, Steven Forbes and the rest of the right-wing braintrust as "12,000 people were arrested for saying politically incorrect things."

Germany at third highest is equally in the realm of complete fantasy. The Tagesschau debunked it and concluded that the German numbers make no sense. There is no statistic in Germany for the number of arrests, but the number of people investigated is lower for the period claimed and not all led to arrests so the number is simply a fabrication.

https://www.tagesschau.de/faktenfinder/grafik-festnahmen-onl...

Finally, the notion that China or Russia would self-report less cases than the UK and expect the figure to be believed is farcical. There isn't even something comparable to the anti-activism laws or the HK47 in the UK.


The question isn't whether or not we have vaccines, it is whether or not we have the most effective vaccines.


It’s a good thing the specific criticism of this trial is that they didn’t use the most effective vaccine for 65+ people, since you’re concerned about having the most effective vaccines.


How do you know if you don’t do a study?

You can call anything a criticism, but it doesn’t make it true.


So, the majority of us, people under 65 are completely unaffected. And yes, the vaccine can be approved for 65- while not approved for 65+.


The fall 2025 approval was limited to 65+/preexisting conditions.

If this vaccine wasn’t being tested for 65+, it might not be approved at all based on that.


I don't think people really understand how big of a wall outside of GPU/TPUs the CPU, ram and even flash market has hit. We're paying as much if not more for the same stuff we were buying 4-5 years ago.

I do think we're more efficient and matrix operations are better (again, GPU/TPUs), but by and large, the computer hardware world has stopped exponential growth.

The period from 1990 to 2005 was amazing. Both in transistor counts and clock speed, it seemed like we were nearly *doubling* performance with every new generation. Memory and disk space had similar gains.


Context here matters, red finds its way into Chinese forbidden or warning signs quite often.


Explain this though. The code is deterministic, even if it relies on pseudo random number generation. It doesn't just happen, someone has to make a conscious decision to force a different code path (or model) if the system is loaded.


Its not deterministic. Any individual floating point mul/add is deterministic, but in a GPU these are all happening in parallel and the accumulation is in the order they happen to complete.

When you add A then B then C, you get a different answer than C then A then B, because floating point, approximation error, subnormals etc.


It can be made deterministic. It's not trivial and can slow it down a bit (not much) but there are environment variables you can set to make your GPU computations bitwise reproducible. I have done this in training models with Pytorch.


There are settings to make it reproducible but they incur a non-negligible drop in performance.

Unsurprising given they amount to explicit synchronization to make the order of operations deterministic.



For all practical purposes any code reliant on the output of a PRNG is non-deterministic in all but the most pedantic senses... And if the LLM temperature isn't set to 0 LLMs are sampling from a distribution.

If you're going to call a PRNG deterministic then the outcome of a complicated concurrent system with no guaranteed ordering is going to be deterministic too!


No, this isn't right. There are totally legitimate use cases for PRNGs as sources of random number sequences following a certain probability distribution where freezing the seed and getting reproducibility is actually required.


And for a complicated concurrent system you can also replay the exact timings and orderings as well!


That's completely different from PRNGs. I don't understand why you think those things belong together.


How is this related to overloading? The nondeterminism should not be a function of overloading. It should just time out or reply slower. It will only be dumber if it gets rerouted to a dumber, faster model eg quantized.


Temperature can't be literally zero, or it creates a divide by zero error.

When people say zero, it is shorthand for “as deterministic as this system allows”, but it's still not completely deterministic.


Zero temp just uses argmax, which is what softmax approaches if you take the limit of T to zero anyway. So it could very well be deterministic.


Floating point math isn't associative for operations that are associative in normal math.


That would just add up to statistical noise instead of 10% degradation over a week.


Catastrophic error accumulation can produce more profound effects than noise.


Just to make sure I got this right. They serve millions of requests a day & somehow catastrophic error accumulation is what is causing the 10% degradation & no one at Anthropic is noticing it. Is that the theory?


FYI something in that region happened last august/September. Some inference bug triggered worse performance on TPUs vs GPU.


There's a million algorithms to make LLM inference more efficient as a tradeoff for performance, like using a smaller model, using quantized models, using speculative decoding with a more permissive rejection threshold, etc etc


It takes a different code path for efficiency.

e.g

if (batch_size > 1024): kernel_x else: kernel_y


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: