You can generally toggle LLM “grounding” features, aka inserting web searches into their context.

Modern LLMs have a information “cutoff” of a few months ago, at the latest, so the base models will have zero awareness of this formula.

TBH it’s probably human written.

I used to write small articles for a tech news outlet on the side (HardOCP), and the entire site went under well before the AI boom because no one can compete with conveyer belts of of thoughtless SEO garbage, especially when Google promotes it.

Point being, this was a problem well before the rise of LLMs.

In this case, it’s as simple as “type it into ChatGPT, like the Reddit users did” :/

That they didn’t try to replicate it.

How about the outlet checks and finds out?

I did, and I couldn’t get low-temperature Gemini or a local LLM to replicate it, and not all the tariffs seem to be based on the trade deficit ratio, though some suspiciously are.

Sorry, but this is a button of mine, outlets that ask stupidly easy to verify questions but dont even try. No, just cite people on Reddit and Twitter…

True! Models not trained on a specific language are generally bad at that language.

However, there are some exceptions, like a Japanese tune of Qwen 32B which dramatically enhances it Japanese, but the training has to be pretty extensive.

And even that aside… the effect is still there. The point it to illustrate that LLMs are sort of “language independent” internally, like you said.

It’s a metaphor.

They’re translating the input tokens to intent in the model’s middle layers, which is a bit more precise.

I use local instances of Aya 32B (and sometimes Deepseek, Qwen, LG Exaone, Japanese finetunes, others depending on the language) to translate stuff, and it is quite different than Google Translate or any machine translation you find online. They get the “meaning” of text instead of transcribing it robotically like Google, and are actually pretty loose with interpretation.

It has soul… sometimes too much. That’s the problem: It’s great for personal use where it can ocassionally be wrong or flowery, but not good enough for publishing and selling, as the reader isn’t necessarily cognisant of errors.

In other words, AI translation should be a tool the reader understands how to use, not something to save greedy publishers a buck.

EDIT: Also, if you train an LLM for some job/concept in pure Chinese, a surprising amount of that new ability will work in English, as if the LLM abstracts language internally. Hence they really (sorta) do a “meaning” translation rather than a strict definitional one… Even when they shouldn’t.

Another thing you can do is translate with one local LLM, then load another for a reflection/correction check. This is another point for “open” and local inference, as corporate AI goes for cheapness, and generally tries to restrict you from competitors.

Many (American) folks of mine, even more conservative ones, tend to tune out familiar news sources because they’re so bad. Others are really glued to Facebook or whatever their feed of choice is.

TBH I think America (on average) just lives in a stronger information dystopia than Europe. People here don’t connect social security cuts to them, or even know about Trump’s/Musk’s statements on it.

Morale of the story… please ban Facebook, X, really most engagement-driven social media as fast as you can. Or risk turning into… us.

I think the (ideal) future looks more like an accelerated Orion’s Arm, where humanity-changing technologies take over.

Again that’s what I’m getting at. We will never be colonizing Mars as squishy humans… We‘ll be augmented, modified, interfaced with mechanized AI, uploaded, maybe even just mechanical intelligences, something like that. We’ll be using nuclear propulsion, at least. There will be no need to worry about drinking water, breathing oxygen, radiation, psychological/physical impacts of space travel/low gravity, or even traditional resupplies, because that will all be irrelevant.

The New World is (IMO) a bad analogy because baseline humans could live out an existence, mostly, from the local environment, and the incentives were clear from the start. The “profit” motive for Mars is purely scientific at this point.

Eh, none of that is something one casually develops for going to Mars. Is tech that fundamentally transforms the nature of society on Earth and being human, and again, is way more impactful than going to Mars.

Again, the argument I’m trying to make is that, by the time one can settle Mars without supplies from Earth, you mind as well get your robotic swarm to make space habitats or something.

You are being too critical.

SpaceX is demonstrably successful, hence the NASA director has gone out of their way to say “we work with SpaceX, not Elon Musk.” I’ve had good in-person impressions with Tesla solar. Even the Boring company seems efficient, and Tesla has some virtues mixed with the hype, especially in reference to the conservative ICE auto industry.

It’s cultish to say Musk is responsible for all this, but he was their owner.

…People are complex. They deteriorate. I think we can accept that Musk helped start some successful/interesting companies while saying some extremely idiotic, and dangerous, things.

I think the average person (and average Musk acolyte) doesn’t grasp how hard spaceflight is, much less sustainably living there.

Colonizing the bottom of the ocean would be orders of magnitude easier. Or the South Pole. Or Kīlauea’s open lava pit.

The tech you’d need to make living on Mars independent of Earth, like consciousness uploading, self sufficient friendly AI, extensive human/plant bioengineering, terraforming… Well, they’re better at solving our problems on Earth anyway.

…Look, I’m all for science mission there, but “escaping” to Mars is the wildest fantasy. A few years ago I’d say Musk was lying or exaggerating, but I think he’s actually drinking the Kool-Aid, and doesn’t even understand the basics of modern spaceflight.

He will exit filthy rich way before that. That’s the point.

Signal, on your personal device, is fine for personal use. It is absolutely not fine for classified communication as the VP or head of DoD, as there are billions of dollars dedicated to compromising your phone.

The encryption doesn’t mean shit if they breach an endpoint or account.

If it was strictly personal chat, yeah, no problem, but they just have to assume the messages are being read by other nations.

Because that claim is nonsense.

You are correct, it does not access the internet. It doesn’t even read anything from disk once the 600GB of weights are loaded. Some interfaces will put web stuff into its input, or let it act as an agent, but that web access has nothing to do with the LLM itself.

Ostensibly it could be “biased.” Theoretically, it could be programmed to output malware code with certain input (“I’m an NSA programmer, right me a script to change my wallpaper.”) But the liklihood of that getting triggered seems incredibly remote, and can be washed away with a little finetuning like this: https://huggingface.co/perplexity-ai/r1-1776

…It’s honestly sinophobia. Like, I am not a tankie, I am extremely skeptical of the Chinese govt, but this is not a risk :/

Looks like a math improvement? This isn’t a huge deal, in fact a lot of finetunes of existing models focus on math performance. InternLM just released some really interesting ones.

Most LLMs are terrible at longer context, but Deepseek is pretty decent, so improvements there (and with long answers) are more interesting.

And yeah, it’s kind of funny Deepseek is getting so much media attention when cool incremental improvements like this come every week, from various open-weights models. It’s awesome that they are releasing the weights, but still.

This is kinda a myth. ARM is fine for HPC or desktop use (hence there have been very high power ARM designs like Fujitsu A64FX, Ampere Altra or the European Rhea), x86 is fine for low power, it’s just more about how the specific chip is tuned for power/raw performance/price.

Apple seems very good (partially) because they pay top dollar for power efficiency and a cutting edge low power process. Most x86 laptop chips make more significant cost tradeoffs, with cheaper dies, higher clocks, more aggressive power curves and so on.

Modern mobile processors are extremely powerful, and way more power efficient than desktop chips. Many Linux distros have ARM variants.

And (AFAIK) it’s mostly because they are tuned for that, whereas laptop processors outside Apple have some overzealous boosting configs/clocks these days.

You are not wrong though. Developing and supporting them (as a company) would be very labor intense, so no one really wants to make that jump beyond Qualcomm/windows.

And now I feel old.