We actually have a big shortage of talent

That review may lead to changes in roles, responsibilities, and continued employment. Individual outcomes will be communicated directly once decisions are finalized.

Knowing how the hiring/onboarding process works in IT — i.e. how fucking hard it is to find, hire, onboard, train and build up a working relationship with an adequate employee — I strongly doubt they have a “shortage of talent” if they’re willing to fire (or even just threaten to fire) people over what amounts to tooling use.

Oh, also, if I got an email threatening my “continued employment” over shit like this I’d be looking for a new job ASAP. And I have been using LLMs for my work for a couple of years now.

Soooo happy that I invested the time and effort into self-hosting my own email server.

Until dovecot and/or postifx enshittify, that is.

It will definitely be profitable for the hardware makers. Companies whose entire value proposition depends on the model weights not leaking in an increasingly sloppy vulnerability-filled world of IT? Not so much.

Contributions from russians make up a sizeable portion of Linux and GNU tooling. A myriad of foundational software, from nginx to strace, were created and are currently maintained by russian citizens.

While there has been a lot of brain drain (hello), russia is still a country of 140 million people and the remains of great soviet theoretical education. In such conditions there will be both a bunch computer science geniuses and a small army of half-competent coders which would allow you to roll your own OS, let alone a distro, if the efforts were organized enough.

Baikal went bankrupt in 2023 because they used TSMC for chip production.

Elbrus is barely hanging on but it’s helplessly far behind, you can’t really run a modern DE and browse modern web on anything but their flagship Elbrus-16C, but that’s made for servers, AFAIK there are no “desktop” form-factor motherboards for it at all, and it’s exceptionally pricy compared to competition.

I think a more likely outcome is that russia starts bulk-purchasing chinese CPUs soon.

I think you’re joking, but in case you’re not… This is just like the paradox of tolerance, everyone has a right to privacy and freedom but if they violate other’s rights, society will in exchange take away theirs. We don’t (generally) call imprisoned criminals “victims of kidnapping” either.

If you just pipe an LLM straight into a feedback loop with a console, it is going to destroy things rather soon, depending on the context window and other factors. All the agentic fluff is basically the workaround for LLMs going off the rails.

Pipe it into root on your laptop for a couple days and see how it fares.

Nah, LLMs are pretty useful for a wide range of tasks. I would grant them the status of “locally intelligent for text manipulation”. They can find patterns, transform, compress, and even generate text pretty well if you formulate the prompt correctly.

The problem with them is two-fold:

1. There is no actual grounded intelligence involved and yet they are used for all sorts of tasks where it is required, like research or decision-making or software architecture or creative writing.
2. They also captured human emotional states really well, which makes them scarily addictive for people looking for human connection (see all the “companion” chatbots) and terrifyingly useful for pushing propaganda, misinfo, and scams. This also multiplies the first problem tenfold because LLMs are trained to come off as confident which means they will “pretend” to be good at things they really are not, and if you don’t know how the underlying tech works you can be lulled into believing its confidently incorrect research results or execute a totally insane decision it makes.

Wars in the Middle East start sounding more and more like statistics that no one really wants to hear about

Has this not been the case for the past 25 years already? No real need for “AI” when you already have an obedient and trigger-happy military with plenty of remote-controlled toys and your populace is brainwashed to believe “brown person = terrorist”.

way less soldiers to choose to disregard unlawful orders

What is the last known case of a US soldier disregarding an unlawful order? If anything, the military is full of assholes who are happy to kill for fun. See: Graham Platner

Will AI make it easier and/or cheaper? Probably, sadly. But it won’t be a qualitative change.

five minutes earlier: “hey dude, marketing here. if you quit and post this dramatic text about the inevitability of AGI on your X account you’ll get a huge golden parachute”.

Seriously, how can competent people look at the current state of LLMs and think that it’s anywhere near AGI material?

Sure, it is already making civilizational changes for the worse. Not because it’s good, rather because it makes us lazy and dumb, has decimated creative jobs, and is used relentlessly for propaganda and misinformation and scams. And it’s much better at using the scaffolding of existing structure to make chaos, than it is at making new structure, making it a weapon more easily yielded by the evil than by the good. A few more software vulnerabilities discovered along the way will not be such a big deal in comparison.

They’d have to effectively close down the Czech branch of the company.

Yeah, I think that’s basically what they’re going to do.

Sadly they can, there is no law stopping a company from shutting down offices. They’ll have to pay people who are laid off a few months’ salary with no work performed but it might be a better deal to MS compared to paying higher wages forever.

The article reads like an ad, but I’m not sure at whom it is targeted. Feels weird

xkcd.com/391

Yeah, still, I wouldn’t claim “as no human would end up falling into that”, given that it may not be that unlikely to find at least one human who displays similar writing the more humans you involve.

No, it is actually statistically impossible for a human to replicate this on sufficiently long runs of text.

This is not about replicating writing like a model. This is basically about guessing which words to pick from the list of suitable words based on a rule that you don’t know (because the key is secret).

To reduce this to the simplest possible example, imagine you are writing a “text” from just two letters: “a” and “b”. Let’s say for convenience that the text is supposed to be random. So the text would look something like “ababaaabbababbbabababaabbbabaababbaaabbabaabbaaaaabaaabbbaabaabababbabbbbbbbbabbabaabbbbbbbaabbabaab”

(generated with '''.join(random.choice(['a', 'b']) for i in range(0, 50)))

The watermarking works as follows: the model owner holds a key, and then uses that key to influence the random choices between “a” and “b” somehow, in a context-dependent way. The actual algorithm is quite complicated, but for simplicity let’s just say we have a secret pattern which biases the random choice towards it. In order to see the exaggerated results, let’s say the secret key is “aaaabbbb” (of course this is a bad secret key, once again just an example), and that the bias is strong (let’s say 80%). So this would mean that the first four letters in our text are more likely to be “a”, the next four letters are more likely to be “b”, then the next four letters are more likely to be “a”, and so on.

Then the text would look something like “aaaabaabaabaabbbabaaaabbaaaababbbaaaabbbabaabbbbaaabaabbaaaaabbbabababbaaaaabbbbaaabbbbbaaaaababaaba”.

(generated with ''.join(random.choice(['a', 'b'] + ([key[i % len(key)]] * 3)) for i in range(0, 100)))

You can see visually that the secret key has affected the text. Of course in this example even if you didn’t know the secret key you could probably figure it out, in reality the algorithm is way more complicated than that, relying on cryptography, so you wouldn’t be able to know the secret key or see that the string has been biased at all.

If the text is long enough, and you know the secret key, you can guarantee that the text was generated with it. In our examples, the letters in the text match our key 77% of the time. The probability of an actual random algorithm generating a text like that is already very low, despite the base entropy being only 100 bits. If my math is correct, for our example the p-value is 2.7 * 10⁻⁸, or about 0.00000027%. I would bet a hungy that the text was generated by our watermarking algorithm, with odds like these!

Of course we did exaggerate the bias and our base algorithm was random. In reality the bias is smaller, the algorithm for determining the likelihoods of possible next tokens is very complicated (it’s the LLM itself), and the algorithm for determining which token to bias is also way more complicated (involving cryptography and real secret keys). That said, hopefully it should help you understand why, for sufficiently long texts, this fingerprinting is just not possible to be replicated by humans.

Yes, fair.

Computers could make cars a lot more reliable and moddable. It’s proprietary/locked-down software and hardware that’s the problem.

I would hope that the Switch falls under “Tablets”. Because that’s exactly what it is.

Video consoles are either desktop computers with proprietary parts, or tablets with proprietary parts. There is nothing especially difficult to repair in them compared to the rest of their categories.