PeterisP

This is a use-case where Stackoverflow fails as well - oh so often I have a problem where Stackoverflow has a question with answers that were true some years ago for an earlier version, but is not true anymore, so the existing answers are worse than useless (as in, they're actively misleading and wasting time) but re-asking the question gets closed as duplicate.

Being unable to clearly mark the bounds of relevancy for answers (in a structured way that affects search) is a major weakspot of SO.

No, that someone would also be able to indeed receive your messages if there is an attacker in the middle who gave you his public key and is then forwarding the message using a channel established with the true recipient's public key.

Message secrecy does rely on being able to authenticate the recipient's public key.

"compliance" alone is a sufficient misbehavior. No matter how people get to a price number, you can't have a contract or any other mechanism for the competitors to try and enforce the coordinated price, that's straight out wrong by itself.

In some fields having it be pre-published pretty much disqualifies that work to be ever published in a way that "counts" as almost all high impact factor journals will only consider unpublished research, so placing it in arXiv means that this research will never be considered 'quality' by the formal criteria and the work is wasted from the perspective of fulfilling the obligations towards the funding source which required Q1 publications.

It's a funding thing - if you don't publish in highly-filtered places, you don't get grants so you can't do science; no one will pay for you, your students and your equipment if you only publish on ArXiv.

I don't want to look for the source of analysis right now, but I recall reading a study demonstrating that a large part, if not most of the word frequency shift was caused by RLHF training done on data predominantly generated by people hired from lower income English-speaking countries which simply have a different dialect of English with a noticeably different frequency of certain phrases and expressions, so e.g. at least some versions of ChatGPT got RLHF-trained to speak more in a Nigerian English dialect.

Since there isn't a single English (English learners generally get informed about the choice of UK vs US English only, but most English is spoken outside of UK and USA in other places and other dialects), but multiple different Englishes, any English speaker will probably find something to be surprised by, and there is an economic incentive to get data from people other than the relatively expensive native speakers of UK or USA English.

There is quite a lot of software (mostly in B2B market) that has a very high marginal cost, where the cost of doing sales and customer-specific customization can easily dominate the fixed cost of making the pre-sale software.

Formal proofs can only prove that the system matches a specification.

Many (most?) non-trivial bugs are actually flaws in the specification, misunderstandings about what exactly you wanted and what real-world consequences arise from what you specified.

> its simply the fact they are a person living in the world.

That's the whole point! If a task requires some time from a human, then you have to include the appropriate fraction of the (huge!) CO2 cost of "being a human" - the heating/cooling of their house, the land that was cleared for their lawn, and the jet fuel they burn to get to their overseas trip, etc, because all of those are unalienable parts of having a human to do some job.

If the same task is done by a machine, then the fraction of the fixed costs of manufacturing the machine and the marginal costs of running (and cooling) it are all there is.

The notion of labor shortage relies on the assumption that productivity is the same, but there is a opposite offsetting effect that's comparable (stronger? weaker? who knows, but IMHO of a similar magnitude) - structural unemployent caused by increased automation, as the electronics and AI systems become both more capable and cheaper. There are already many jobs which could be done better by machines, but it's simply cheaper to pay peanuts to a person instead - and as soon as the machine becomes cheaper and/or wages grow that the person is 1% more expensive than the machine, that profession will shrink in numbers.

It's counterproductive to try to grow a workforce segment that's likely not be needed soon. If there indeed is a labor shortage for some years, the society will handle the increased wage cost (which also imply increased push for automation) much better than it can handle a large segment of angry unemployed people, so while the future is unclear, it's better (at least for the society - the billionaires definitely benefit from cheap desperate potential workforce) to err on the side of doing less to solve this "problem" rather than doing too much.

For practical poker, I'd formally define "optimal strategy" as the strategy that maximizes profit per time (or per game) for a set of opponents, including also the actions needed to "explore" and discover any bias before exploiting it.

Assuming at least one of opponents is not playing Nash equilibrium (which is a very solid assumption), playing the Nash equilibrium becomes suboptimal as it doesn't exploit the exploitable as much.

To run on AA batteries you'd probably have to go to microcontrollers like ESP32 which are way less powerful than that Intel device but also way more powerful than TRS-80. Assuming a power consumption of 0.25W for that SoC and basic 2000mAh AA batteries, a 4 AA battery setup could run the device for 48 hours.

It also would make some sense for various real-time-request-augmented bots to not only use the user agent string of the user's browser, but actually use the user's browser to make the request.

The simple answer to a question posed in the article "On the other hand one could ask why there isn’t a requirement that published research should be reproducible using open source tools (or at least tools that are freely available to the public and other researchers)?" is that the obvious immediate outcome of such a requirement is that papers failing that requirement - like the Quviq QuickCheck papers, which seem to have been useful to the author and others - simply would not get published, and the community would lose out on that gift of information.

What I hear is that similar techniques should/could be used by explicitly modeling it not as an interval (6.9, 7.1) but as a gaussian distribution of 7±0.1, and a computer can do the arithmetic to see what the final distribution is after a set of calculations.

Ironically, I'm thinking about perhaps moving to the electronic warfare industry, as with the continuing war nearby and the major impact that EW and drones have on the lives and deaths of infantry defending me and my family, I feel some sense duty that this could the most valuable practical contribution I can make, bringing more impact than going to the trenches myself (which is also an option, me being in the reserves).

If you had shit on your hands, would you consider them clean after wiping them with a paper towel a few times, or would you want to wash them with water? The same applies for your butt.

As far as I see, GDPR would not applicable here - GDPR is about control of "your data" as in "personal data about you as a private individual"[1], it is not about "your data" as in "content created or owned by you".

[1] GDPR Art 4.1 "‘personal data’ means any information relating to an identified or identifiable natural person (‘data subject’); an identifiable natural person is one who can be identified, directly or indirectly, in particular by reference to an identifier such as a name, an identification number, location data, an online identifier or to one or more factors specific to the physical, physiological, genetic, mental, economic, cultural or social identity of that natural person;"

So many of the companies are global, or at least have offices both in USA and Canada. Why do they hire devs in USA instead of Canada?

IDK, I have multiple times seen significant practical security improvements as a direct consequence of some "93 section questionnaire" because the very first section had a few questions "Are you doing this simple, well-known best practice thing?", which they were not, because it took some time, effort and/or money and they just didn't care.

But once the questionnaire mattered, they started doing it just so they could legally answer "yes" to that question. Things like finally changing the default admin passwords on that service they installed a year ago, and testing backup recovery to find out that it actually can't be done due to a bug in the backup script skipping some key data.

But Ivan the Terrible simply wasn't a good king - taking Russia from being a grand-principality centred on Moscow to a continent spanning empire is a major change in the world that we have to note in history, but it doesn't make someone a good king, neither his contemporary nobles nor his contemporary peasants nor the conquered peoples really benefited from that.

There's a difference between good and great, having a major impact doesn't imply that the impact is good. Alexander the Great is another example of someone who had an outstanding impact, but was not a good king.

No, he's not a rector because he has a high H index - he's rector because he was the only candidate who applied, and if there was a competition, that factor is barely relevant, the candidate would get chosen by a faculty subjective vote, not according to score on some metric.

Nah, that's often how it works - sure, becoming a rector would mean a pay increase, but it would require you to become an administrator and stop doing what you love, most likely permanently (as the gap of being away from your field of research for multiple years makes it hard to become productive again - returning to active science is definitely possible, but rarely happens), switching your career for an entirely different one as far as day-to-day tasks go. And if you're already somewhat comfortable as a professor, why would you do that to yourself?

I've literally seen faculty 'drawing straws' as of who'll agree to be the sole candidate for a particular elected administrative position, because it's clear that one of them has to do it for the university to function, but no one wanted to.

Okay, but it does seem that this hack is in the entirely opposite direction; a pure transformer is more towards "learning to learn" than any special preprocessing to explicitly encode a different representation of numbers.

We probably do have to move beyond transformers, but not in the direction of such hacks, but rather towards even more general representations that could encode the whole class of all such alternate representations and then learn from data which of them work best.

While board meetings normally are relatively sparse, most of board work happens between these meetings (where you effectively perform any voting rituals required to formally approve the decisions the board members made beforehand) and it is very common practice for executives to continuously keep board in the loop - arguably doing this well is one of the primary duties of a CEO.

For arbitrary documents and queries, how do we reliably segment the text between those two different languages? And if we can do that, why can't the model do it implicitly?

ASCII digits do not always imply base-10 numbers, they can also be identifiers (e.g. phone numbers), parts of words (IPv6, Log4j), and used in various 'written slang' such as g2g, 4ever, m8 for mate, etc, etc.

And, crucially, I'd argue that for in "chatbot" tasks those other uses are more common than arithmetic, so arbitrary focus to specifically optimize arithmetic doesn't really make sense - the bitter lesson is that we don't want to bias our architecture according to our understanding of a specific problem space but rather enable the models to learn the problem space directly from data.

That's not cataloguing, that's recording, and as far as I understand this is long ago done - cataloguing is those "all other details" which require expertise and time; all the things like figuring out that this coin is a roman coin from 1st century, and that other coin from the same find is from another location.

That's kind of even worse, other people have no business at all looking at my private pictures.

A key part is that in US you take less risk because investors are taking more risk - as described in the article, in UK the conditions you'd get for early funding push more risk towards the founder as most UK investors simply won't accept as much risk as US investors routinely do.