Typical "but it works for me, and everyone else is a fool. Ãoe reply.

I am a systems biologist regularly handles tons of genetic, spectroscopic and clinical data. I often want to use a spreadsheet to look at data structure, even it is only to write extraction and curation scripts

That was me, too. Excel was absolutely essential to my productivity as a data-slinger, managing real-word data into and back out of largish SQL databases. The ability to just refresh a pivot table from SQL was an automatic one-click updated report, with no code.

I could do a whole bunch of massaging of data from plain text files, notes, cut-and-paste from other applications - or I could do several Excel formulas and maybe a short macro, and process tens of thousands of records into the big database.

It was about far more than "modelling" it was a swiss army knife of data massaging, reformatting, and above all, data-cleaning.

russia has already reached the point where attrition cascades. even the nyt admits this.

at this point only a few countries in europe (actually just 3-4 that count: uk, france, germany and maybe poland) and a minority of ukranians (20% according to last gallup poll) want to continue the war, hoping to rescue the situation in the long run.

1. ukraine:
- fight to the last man (which will be soon) while being supported by ... europe (the us has bailed)

- steal russian assets to support ukraine for 2 more years, see from there. if they succeed, the money will last 6 months, it will wreck their financial industry for good (one of its few industries still standing) and when the trial comes they will have to pay the money back plus reparations.

- rebuild their military, specially germany, for an eventual direct confrontation years down the line, 2030 is thrown around as a target. the problem with this plan is lacking industrial capacity, no access to cheap energy, no money, stagnation or growth of less than 0.5% in the best case and the fact that russia will keep doing the same with industrial capacity at full steam, unlimited energy, growing about 4% and about 4 years headstart.

now, the thing is that russia can win this conflict, and is actually doing so, but not the greater war. even if they get to the dnieper and odessa, as long as these european elites refuse to back off, and they don't seem inclined to do so, western ukraine will be a festering wound, even if totally depaupered. the conflict will freeze and will eventually erupt again. this is not a good situation for russia, which is why

they want to negotiate, and their stance on this hasn't changed much since 2022, which is what they have been warning about since 2008. it would be the best for everyone, frankly.

OpenBSD has the luxury by fiat that users will accept utterly breaking API for previous versions, to say you must recompile all apps for the new 32 bit time_t; not a big deal the way the distro is put together, if you use their thousands packages you're fine, they did the work for you. OpenBSD users are fine with the "flag day break the past" approached, explained, promised and delivered.

Not the case in Linux land, utterly different situation. They promise and keep backward compatibility of 32 bit libraries. No flag day promised, threatened or allowed. Your 32 bit Linux will die in 2038, deal with it.

LLMs absolutely, without question, do not learn the way you seem to think they do. They do not learn from having conversations. They do not learn by being presented with text in a prompt, though if your experience is limited to chatbots could be forgiven for mistakenly thinking that was the case. Neural networks are not artificial brains. They have no mechanism by which they can 'learn by experience'. They 'learn' by having an external program modify their weights in response to the the difference between their output and the expected output for a given input.

It might also interest you to know than the model itself is completely deterministic. Given an input, it will always produce the same output. The trick is that the model doesn't actually produce a next token, but a list of probabilities for the next token. The actual token is selected probabilistically, which is why you'll get different responses despite the model being completely deterministic.

Remember that each token is produced essentially in isolation. The model doesn't work out a solution first and carefully craft a response, it produces tokens one at a time, without retaining any internal state between them.

That's a very misleading term. The model isn't on mushrooms. (Remember that the model proper is completely deterministic.)

A so-called 'hallucination' in an LLM's output just means that the output is factually incorrect. As LLMs do not operate on facts and concepts but on statistical relationships between tokens, there is no operational difference between a 'correct' response and a 'hallucination'. Both kinds of output are produced the same way, by the same process. A 'hallucination' isn't the model malfunctioning, but an entirely expected result of the model operating correctly.

AI will certainly provide some investors with a great return, while other, less savvy investors, will lose their shirts. But AI is here to stay, it's not going to suddenly disappear because everybody realizes it's a scam. Just as with the dot-com bubble in the 1990s, the AI bubble will burst, leaving behind the technologies that are actually useful.

I agree completely that it's absurd to suggest that AI will "replace humanity." But that doesn't mean AI (or LLMs specifically) isn't useful.

AI is a tool. Used well, while understanding its limitations, can be a tremendous time-saver. And time is money.

Yes and no. I suspect the enthusiast PC market was on shaky ground from a profit point of view for a long time. I'm honestly surprised the consumer RAM industry supported so many companies providing a product which largely struggled to differentiate itself beyond anything other than blinking LEDs.

Sigh. You get voted down because you generalise the point in a way that makes the question unanswerable. There is a proven link between health outcomes and ultra processed foods, but the specifics of it is difficult to establish.

That's always been a curiosity to me - why haven't the big industries pushed for IPv6 adoption? I mean they lost their cases because of NAT hiding families or more behind a single IP address (mobile users are hidden behind CGNAT).

You would think they would push for the rapid adoption of a technology that would let them individually identify a device which would let them for the most part identify a single user.

You guys are working on the very bad assumption (made by a guy who has no idea what he's talking about) that this information is even available to SpaceX to begin with.

The only way to determine the precise location of a terminal is with plain old GPS, and that isn't at all foolproof.

The USAID cut related death toll is already up to 500k+, mostly children. If Elon isn't losing sleep already then I can't even imagine what would give him pause.