not2b

I'm reminded of the story of Helen Keller, and how it took a long time for her to realize that the symbols her teacher was signing into her hand had meaning, as she was blind and deaf and only experienced the world via touch and smell. She didn't get it until her teacher spelled the word "water" as water from a pump was flowing over her hand. In other words, a multimodal experience. If the model only sees text, it can appear to be brilliant but is missing a lot. If it's also fed other channels, if it can (maybe just virtually) move around, if it can interact, the way babies do, learning about gravity by dropping things and so forth, it seems that there's lots more possibility to understand the world, not just to predict what someone will type next on the Internet.

A president can revert an executive order from the previous administration. But the Inflation Reduction Act is a law, not an executive order. If Trump doesn't like it, he can get Congress to repeal it. But he isn't the king. The constitution requires that the president "take care that the laws be faithfully executed". It isn't fraud when funds that were allocated by the act were distributed according to the act. If someone cheated then by all means they can bring charges if they have any evidence, which they apparently do not.

It was already disclosed to the bad guys that someone managed to break their encryption, when they didn't get paid and they saw that the customer had somehow managed to recover their data. That probably meant they might go looking for weaknesses, or modify their encryption, even without this note.

Other victims whose data were encrypted by the same malware (before any updates) could benefit from this disclosure to try to recover their data.

I'd say that they are more uncertain than industry jobs at this point, unless you're talking about companies that have recently been taken over by private equity where the new owners intend to gut the place and sell off the parts.

Kapersky, perhaps? Flagging an antiwar message as dangerous?

Even in a non-Euclidean space with positive or negative curvature, the limit of the ratio of the circumference to the diameter of a circle as the diameter goes to zero is pi.

This sounds like a management failure more than your failure. You should have had someone who would have steered you in the right direction, getting you to focus your energy on things that mattered.

I think that the article is useful because it corrects a number of things commonly said about Bell Labs that just weren't correct. They didn't just hire a bunch of geniuses and let them do whatever they wanted.

The National Endowment for Democracy was only one of many sources of funding for Bellingcat, and apparently a minor one. They used Kickstarter to get off the ground, and most of the grants listed on Wikipedia appear to be European. They get a lot of contributions from individuals as well.

See https://en.wikipedia.org/wiki/Bellingcat#Funding_and_support

Not sure who you are quoting here, I never said "embarrassing". I'm sure that those "nerds" made adjustments based on the failures. That's all I was asking for.

They shouldn't just cancel, but with the same kind of failure twice in a row it seems they should require correcting the tipping-over issue before trying a third time.

I was impressed with how tight the known bounds are. For length 30, the upper bound is 1.0000397 times the lower bound. For 40 the ratio is 1.0000164 and it quickly approaches 1.

Yes, code review can catch these things. But code review for more complex issues works better when the submitter can walk the reviewers through the design and explain the details (sometimes the reviewers will catch a flaw in the submitter's reasoning before they spot the issue in the code: it can become clearer that the developer didn't adequately understand the spec or the problem to be solved). If an LLM produced it, a rigorous process will take longer, which reduces the value of using the LLM in the first place.

If the hallucinated code doesn't compile (or in an interpreted language, immediately throws exceptions), then yes, that isn't risky because that code won't be used. I'm more concerned about code that appears to work for some test cases but solves the wrong problem or inadequately solves the problem, and whether we have anyone on the team who can maintain that code long-term or document it well enough so others can.

I'm not a Ruster, I've spent my career writing a ton of C++. But I'm interested in Rust as an alternative because it doesn't need GC.

But in this case, it's not just the memory safety I'm interested in, it's the data races. If we have multiple threads but can guarantee that any object either has only read-only references, or one mutable references and no readers, we don't have data race issues.

Consider the split_at_mut function. It takes one mutable slice, and returns two mutable slices by chopping at a caller-specified point. It's in the standard library.

The operation is completely safe, as after the call the original mutable slice is no longer live, but the borrow checker won't let you write such a function yourself unless you tag it as unsafe, so that's what the implementation must do.

The same thing happens in the implementation of Vec: there is low-level code that is unsafe, used to provide safe abstractions.

There are other ways to exploit formal verification. For example, you may be able to formally prove that the unoptimized version of your program has the same behavior as the optimized version, subject to specified constraints. This can assure that there aren't any bugs in the optimization that affect program correctness. You're eliminating a class of bugs. Of course, it's possible that the unoptimized version is incorrect, but at least you can assure that your aggressive optimization approach didn't cause breakage. Such approaches are commonly used for hardware verification.

That one, though, is because we are running into physical limits: if we want to build things out of atoms, we can't make features that are half an atom thick. Even above that scale, physical effects that used to be ignorable, like quantum tunneling, no longer are.

From the late 70s through about 2005, scaling semiconductor generations was easy. MOSFET scaling followed rules formulated by Dennard, which provided a fairly easy method of scaling semiconductor designs from one generation to the next, keeping power density roughly constant and continually improving performance. The problem is that by around 2005, if you did it that way, your gates were no longer switches, they were dimmers, and leakage power started to dominate, and that meant that chip architectures had to change radically to keep on scaling.

So, we can no longer just scale designs from one generation to the next, we have to come up with completely new approaches. That's much harder.

https://en.wikipedia.org/wiki/Dennard_scaling

Sometimes the tool doesn't have to know he's a tool. The Russians use the term "useful idiot".

They've created the back doors, I'm sure that the Russians and the Chinese are now in the systems the DOGE people broke.

If they'd only been downloading data that would be bad enough, but they've been modifying code as well, code that they don't understand. It's the biggest security disaster in recent history (perhaps Kim Philby giving everything to the Russians was worse) and entirely self-inflicted.

Yes, but a university or other institution will often also require IRB review in additional cases, because maybe they don't want to let researchers accidentally cause brain damage just because a billionaire rather than a government is paying.

"The main thing holding back wider adoption is a lack of system interfaces. File access, networking, etc. But it's just a matter of time before these features get integrated."

But then you've got to figure out and prevent all the security holes that can be introduced by adding file access, networking, etc. That's what killed the Java write-once, run-anywhere promise. Maybe put the whole thing into a container? Oops, looks like the container wasn't replaced after all (though perhaps it could be simplified).

The shrew had to assume you were going to eat it, and it did what it needed to do to escape.

YouTube doesn't even bother with scene cut detection; they'll insert ads mid-sentence. A lot.

It appears that the intent of the ad scheduling is to be so annoying that it motivates people on the fence to pay for premium.

Is the name "O"? That's a common Korean surname, though many Korean Americans spell it "Oh" to avoid the problems computers and bureaucrats have with it.

Statistics I can find via search differ by a lot, but https://www.appbrain.com/stats/number-of-android-apps says there are currently 1.58M available apps. So, it seems they are moving more apps than they are allowing.

It seems that they have changed their minds about surveillance back doors, after some devastating attacks, where Chinese state actors (among others) used back doors created for compliance with warrants to get in.

But that was the pre-Trump NSA and FBI. Now the Chinese and the Russians just need to get some DOGE volunteer to give them whatever they want, since Elon now has root on all the government payment systems and is too undisciplined to do things in a secure way.

I don't think so. I have many nontechnical friends who are furious at having to deal with bad AI, whether it's a stupid chatbot that they have to talk to instead of a real person or Google "AI overviews" that often get things completely wrong.

Yes, same architect (Dave Cutler), and WNT is (V+1)(M+1)(S+1) though they probably didn't do that on purpose.