RandomBK

On (2), I think there's some general policy discussions we should have around government spending rescuing people willing and knowingly doing obviously dangerous/stupid things.

Americans fucking around in North Korea or active war zones is another example that comes to mind.

I think that would be a better use of time than spinning up a 100-person government licensing program to regulate the 1000 people playing with submarines.

Instead of targeted regulation specific to submarines, I would look at the laws around legal liability and depth of disclosures. That would scale across many more types of activity rather than any single one.

1. Is what you're doing dangerous?

2. Did you tell your customers about the danger?

3. Do they really understand the danger, including past incidents?

4. Is anyone else being harmed or are you just gambling your own life?

Even then, I would posit that this case probably passes this bar - OceanGate was quite clear in their waiver and there's a level of due diligence that a reasonable person spending 250K should be expected to make.

Is anyone else hitting errors submitting the request?

The aforementioned Sarbanes-Oxley act, plus the principle of "negative inference" in civil suits which dictates that a court can infer that someone refusing to testify (or destroying evidence) has something to hide. There's no 'plead the 5th' in civil suits.

It's worth asking just how small did scientists in the era consider the contamination risk. Was this a case of willful ignorance of a legitimate risk, or a begrudging theater to placate a general public's overactive imagination?

Even based on what was known 50 years ago, what was the hypothesis for how microscopic particles with billions of years of geographic isolation might hurt the biosphere?

I'm reminded of the "LHC might create a black hole and swallow the earth" craze in our more recent memories.

In the case of DNSSEC, the ultimate 'root key' is handled quite carefully and is interesting to read up on.

https://www.iana.org/dnssec/files

https://www.cloudflare.com/dns/dnssec/root-signing-ceremony/

This of course, only works for your own internal use. I feel this disclaimer needs to be called out explicitly.

If you're trying to do something that will be trusted by the global internet, you'll need to use an official CA, ergo the complaints of DNSSEC being too 'centralized'.

Yes, you will be able to train an AI model that is capable of beating Doom using this machine.

I'm still rocking a Brother DCP-7060D from over a decade ago, using an aftermarket toner I bought from my university tech store.

It's not just the ability to correctly answer this, but the consistency.

I asked this exact question to the `oasst-sft-6-llama-30b` model and it was able to consistently get the correct answer. I then tried the smaller `vicuna-7b` model, and while it usually gave the correct answer, there was the occasional miss.

Interestingly, `oasst-sft-6-llama-30b`'s ability to answer correctly seems to be fairly stable across multiple configurations. I tried various temperature settings from 0.2 up to 1.2, different topP configs, and they all answered correctly.

This is very useful for running containers in lightweight untrusted environments.

I personally use this setup for my GitLab CI runners, using a mix of podman and buildah to build containers. I can't 100% trust the container build script, so avoiding having to expose a Docker socket to that untrusted code is a critical safety layer.

Unfortunately [2] is just a placeholder for now, but it does look like the intent is to publish the weights.

Plus there's bound to be false starts, reverts, crashes, etc that bump up the actual reproduction cost. Most training cost estimations take an extremely rosy best-case view assuming everything goes smoothly on the first try and no gpu cycles were wasted.

Depends on the backup. Accidentally backing up a couple of `node_modules` folders would be sufficient to hit the 5M limit.

I can't speak to Parent's situation, but temp tables are common practice in OLAP workloads across many database engines (personal experience with Postgres, Presto, Spark, Netezza).

At a certain level of query complexity, no query planner is able to accurately predict the characteristics of sub-queries or CTE, resulting in query plans that are ill-suited to the problem. More often than not, materializing CTEs inside a large query into 2-3 temporary tables results in order-magnitude performance as the database now knows exactly how many rows its dealing with, stats on # of nulls, etc.

Ignoring the policy side of things for now, as I'm sure we'll get plenty of discussion in other comments...

On a technical level how well do these adversarial attacks survive trivial transformations on the original pixel data? I would imagine even a simple JPEG compression pass would ruin the effect.

>There is nothing stopping a 400 lb programmer, to quote Donald Trump, from bootstrapping their own product.

With the cost of training such models from scratch dropping every year, this type of legal shenanigans will only encourage model development to move into the shadows where there's be even less accountability and AI ethics.

>the moment you step out of FAANG

Even in FAANG, most data is going to be extremely messy.

* There's usually very little incentive for good documentation, so you have to guess what the table contains based on column names and exploring patterns in the data.

* Enforcing a single pattern/process for data across the entire org is a pipe dream.

* Logging tables are written to by hundreds of engineers across dozens of teams, with no standardized naming scheme.

* The product contains multiple edge cases/special overrides for very specific circumstances, so there's very few simple queries that don't come with 100 footnotes attached.

FAANG is not immune to large-organization problems, and data quality is no exception.

Apple (and to a lesser extent Android) have already gone this route, and Windows 11 originally wanted to push _hard_ on enforcing the presence of more platform security. DRM schemes already perform remote attestation before you can stream 4K Netflix.

Linux and Open Source may be one of the last remaining barriers to this becoming widespread. It's one of the main reasons I whole-heartedly support Valve and their Steam Deck ambitions, and encourage everyone to do the same. Money trumps everything else, so as long as the money to be made from supporting SteamOS > money thought to be saved from piracy, I believe we can still thwart this.

Every time I hear about the XBox's platform security, I think back to this talk [1] about how Microsoft protected the XBox One. A lot of it boils down to clear requirements and good engineering, and many of these technologies are now showing up via Project Pluton.

[1] https://www.platformsecuritysummit.com/2019/speaker/chen/

There's no official announcement yet, but some people are given an option to "Upgrade Plan".

Mods: Please feel free to change the URL once something official comes out.

It depends on whether they actually lowered the price by $1500, rather than simply bumped the price by $1500 across the board then 'graciously' canceled it out if you subscribe.

Of course, barring legal discovery or a leaked email we'd never know.

Securities and Exchange Act of 1934, Section 10b-5 [1]. You'll commonly see this shortened to just "10b-5" in common parlance.

There's quite a large industry [2] involving lawsuits anytime someone can credibly claim that an executive knew or (the real kicker) should have known they were making a false statement.

The vast majority of cases settle - there's rigorous and standardized models for computing how much to settle for. The special thing in this case is how it's actually gone to trial.

[1] https://www.investopedia.com/terms/r/rule10b5.asp

[2] https://securities.stanford.edu/

Edit: The specific case here is https://securities.stanford.edu/filings-case.html?id=106701 (Docket 18-CV-04865)

BorgBackup has most of what you're looking for, though it doesn't implement CDC and doesn't replicate the files as-is in the backup location (instead using a compressed/deduped/chunked storage format)

Because there's crypto, and then there's "crypto".

While FTX & co. don't match the technical blockchain-based definition of crypto, they are what people think of when they talk about the crypto ecosystem, and are part of the crypto hype chain.

If we filter the crypto market to only the true not-my-keys-not-my-coin crowd, the market wouldn't be anywhere near as large and influential as it is today.

7.4% of profit (2/27) versus 50% of profit (2/4) was the factor of ~7 parent was referring to.

Granted these are comparing annual cost versus quarterly figures, but the overall narrative still stands.

Buying some puts also works.

Tough to say. Do I need a Toyota? no, but I'd happily buy one for $1K.

Demand elasticity makes any report on the volume of demand irrelevant unless it also covers the pricing of that demand.

We did weigh the risks. We did study asbestos. And we as a society made an informed decision that the risk of asbestos outweighed the risk of buildings catching on fire.

I don't understand the point you were trying to make here. Why shouldn't we invest effort to research a technology that might mitigate the 2nd leading cause of death in the developed world?

Having used many different providers (though not all of the ones on your list), be very careful with suspiciously generous or unmetered anything. More often than not, you'll hit a soft limit where your performance will degrade and/or you'll get kicked for ToS violations.

Most apps never hit those limits, but once you get to multi-thousands/mo in cloud bills, there's benefits in pay-as-you-go billing where the provider is incentivized to let you use as much as you want to pay for.