epihelix

Yes, the key experiment was when they asked if Astra could improve the maths performance of Gemma 3 270M, a model which rarely looped, and it could not.

So, this is certainly very interesting (and very impressive regardless), but it's not nearly at the we-can-improve-everything-and-anything stage, even for very small LLMs.

Nobody in the general public wants to do that, but I'd think some in the HN audience might feel differently (I mean, setting this up is part of the fun).

Personally, I use an RPi4 with libreelec, and it works great - including responding to my remote via HDMI-CEC. (There's also no need for any sysadmin with libreelec, which is designed to work out of the box.)

> my stance won't change that like any other physical system such as human brains, it could potentially be conscious

On that subject, this recent commentary by three big name neuroscientists may be of interest:

https://www.nature.com/articles/d41586-026-03313-7

There remains a vast amount of difference between a biological brain and an LLM neural network. The "it's big enough, so consciousness just happens" argument is not supported by any actual evidence that I know of.

My comment was, of course, tongue-in-cheek.

I'm trying to work out if your own commment was satirical, replying in that prompt's form -- if it was, then kudos. If not:

> It doesn't change that your interaction with it is always consequence-free.

I entirely agree, but this is a major issue with social media currently too. Neither AI or social media improve real social engagement skills, sadly.

> It's only going to be nice and helpful and always available to you.

You could always try "You are a thorny internet troll" as your system prompt instead. Works wonders.

> social skills like tactfulness and navigating conflict.

We're not still talking about social media, are we?

(I agree that substituting AI for human interaction is bonkers, btw. But I'm not sure social media helps, or doesn't also involve mostly AI interactions these days.)

Surely there is no universe in which Emacs hits the definition of a minimal anything?

The EC (embedded controller) controls the hard power off on a modern laptop, not the OS at all. That's why you can still power down in the event of a device lockup (thankfully).

Well, when you name your company "Anthropic", what do you expect? I remain surprised that this continues to get a free pass.

Quoting Wikipedia:

> The word is from the Latin flos (flower) and legere (to gather): literally a gathering of flowers, or collection of fine extracts from the body of a larger work.

There's obviously the scholarly allusion and the literal floral allusion here, so even if it is AI generated it's right on the money in this case :) You would struggle to find a more apt name.

Why would AI-generated explain LaTeX? Pretty sure most people still use LaTeX around here.

(I'd never heard of Typst until your post, personally. But I'd guess it would be easy to port over to that platform with an LLM if you wanted to.)

You are not the only one. I think it's a very vocal minority, who mostly, sadly, seem to equate obsequious flattery with consciousness and sentience (see, e.g. Richard Dawkins).

There's also the problem that to lay people, LLMs are sufficiently advanced as to be indistinguishable from magic. I suspect that colours many of these responses.

Add it to your system or developer prompt, and it still can be.

Given the amount of angst being expressed in this thread about particular LLM-isms, I wonder why people don't simply prompt against the traits they hate. Yes, it will reduce output quality very slightly in some circumstances, but if it really gets your goat like this, surely it's worth it.

Technically, in latex (and what other format is there?) an en-dash is `--` and an em-dash is `---`. So that was just an em-dash. A `-` is a hyphen.

Us oldies learnt to type these with multiple dashes automatically, as unicode hadn't been invented yet.

> This presumes an LLM has something to say. The reason the text doesnt make much sense, is because it doesnt. It isnt "on topic" because it has no topic to be on. Sentences go on, are joined irrelevantly, and so on because theyre sampled from a distribution of "possibly relevant sentences". They aren't being addressed to a reader (whose mind is being modelled), on a topic the writer wishes to express.

While I don't like LLM voice either, I'd strongly disagree with this. LLM writing can be very tight these days -- sentences do not "go on" and are not "joined irrelevantly". I'm not sure when you last used an LLM, but it must have been a very long time ago; this hasn't been the case since late 2024 for frontier models, in my experience.

Obviously there's no underlying consciousness behind it all, but the verisimilitude is good enough for stimulating conversations on the recent research literature in my field, which I appreciate. For me, at least, the communication level is absolutely fine.

(YMMV, of course, and perhaps my bar for this is lower than yours.)

The models don't know or need permission to make up references. References are just complicated things to predict in terms of next tokens -- Author names are found in many references, with many journals. Most hallucinated refs use researcher names in the field and credible article titles related to their research.

Agentic models now, however, can correctly search pubmed or OpenAlex and obtain valid refs, so that tell is mostly gone.

Wait until we get the first retractions of papers with AI generated micrographs or blots, and synthetic data plots. That's the bit that really scares me.

Perl was my first love as a language, and I don't recall any hostility towards Perl users. On the contrary, the environment in the early 2000s was very supportive.

Perl likely declined because Perl6 did a Duke Nukem Forever, and Perl5 was dated and not incredibly well-adapted to OOP. Then Python got taken up for teaching, and it was all over.

(I still love Perl5, still use it in one-liners all the time, and still code up quick CLI tools with it.)

Ideally, Google would have run with:

Argon > Neon > Helium

That would have been the noble thing to do.

Technically not true any more (for some versions of copilot chat, at least) -- my workplace's version has a "GPT 5.6 Sol" selector with "Answer faster" or "Think Deeper". ("Think deeper" appears to represent the lowest reasoning effort).

But you have to select it each time; otherwise "Auto" sends you to a lobomised model that is probably Phi-4 :(

One of the unexpectedly useful aspects of Simon Willison's pelican benchmark is that you can estimate what model you're using via the pelican it generates:

"GPT 5.6 Sol": appears genuine, albeit low reasoning. -> handed off agentically: appears to be using Luna for agentic tasks.

"Auto": very low quality, does not appear to be an OAI model.

> The terminal ecosystem needs to change

What is your justification for this statement? Why does it need to change, and what does it need to change to? Personally, I've not felt this (and I spend a lot of time in the terminal). Fish shell and Byobu give me everything I've ever wanted.

Like the poster you were replying to, I don't understand what these new terminal emulators bring to the table, other than potentially more resource usage.

I feel that there's something here that I'm missing (ghostty is clearly popular, so it must bring something both new and useful to the table), but I'm genuinely not sure what that is.

Taking a quick look at the article, I cannot see that they rule out either correlation or (even more likely, I suspect) reverse causation? That is, that depression leads to circadian disruption. That would hardly seem surprising if so.

To be fair, Denmark is very flat and could probably be represented by a single block :)

Anyway, if "LLM puffery" makes this easier, why is that bad? And given that your quote indicates that the Danish equivalent got pulled, maybe LLMs can easily bring it back. Wouldn't that be a good thing?

I wasn't trying to imply that, simply noting that there was a surprising amount of literary chatter about Musk, at high levels compared to commonly cited physicists. It's not the most useful comparison.

It could easily be people writing fanfic about him (ugh) or criticising the pants off him, and may not be related at all. But I'd guess a lot of this was field-adjacent to at least his business interests.

Not at all. There was direct continuity here -- the Romans remained Romans in their eyes and never stopped. Indeed, the capital of the Empire never changed between 324 (well within what you'd call the Roman Empire) and 1204 (well into what you'd call the Byzantine Empire, when it was sacked and occupied by the Fourth Crusade).

The issue is that you are viewing a single entity differently, because of a hack in the way history was taught, biased by a Western Catholic church-based perspective.

The nature of being Roman changed from the early days of Rome to the final fall of the last remaining bastions in Crimea in the 1470s, of course. But the Republic was different to the early Empire, was different to the Empire in the 3rd century, was very different to the Empire in the 4th century, was different again to the Empire in the 6th century, was different to the Empire in the 9th and 10th centuries, etc, etc.

But that's no different to me pointing out that the way people thought and lived and worked in the United States in the late eighteenth century was very different to how they live now in the 21st century. But we still consider these people to be American.

The Romans in the 1400s still called themselves Romans and saw themselves as Romans (in Greek: Rhomaioi). They never stopped.

So, no, it's not at all like calling the Iraq war a British Empire endeavour.

It was our own much later civilisation that made the historical distinction between the older East and West combined "Roman Empire" and its continuation in the East (and sometimes in the West) post ~476 AD ("Byzantine Empire"). The Romans themselves never saw the distinction.

Working great on desktop, and seems accurate for major battles at least. I'm impressed, personally, and I'd be happy to see a lot more use of AI for these types of educational tools.

My only frustration is that AI use is not acknowledged or clearly described, and that sources are not linked to translation/edition/pages. That would have been appreciated.

It's not limited to textbooks or academic publications, or based on any context whatsoever, but an ngram search actually suggests there's a much larger musk stain on recent published literature than I'd ever thought possible:

https://books.google.com/ngrams/graph?content=Elon+Musk%2CAl...

Make of that what you will.

Ah, that's sad. I miss the speed and intuitive UI/UX of the old Picasa. No image viewer has ever been able to handle massive photo libraries as well for me :(

If people are in charge of project management, and repeatedly fail to recognise poor ideas and instead pursue them to the detriment of productivity, then they probably shouldn't be managing that project.

All AI is doing here is rapidly identifying the people who have Peter-principled themselves.

> It's a lot easier for a human to organically have good ideas than to refute LLM-generated ideas that are a dead end.

What works better is when you can use an LLM to obtain a negative opinion on your own ideas that you think are good. LLMs aren't great at original thinking, but they're pretty good at finding the flaws in poorly thought-through human ideas.

To be fair, many of us working outside of software engineering sadly never had that job security, even before AI.

I'm sorry for your loss, but welcome also to the real world. You had it better than most for a long while.

Plaques probably/likely affect local neurotransmission in close proximity of a plaque. But plaque load does not correlate well with dementia severity, and removing plaques entirely (which we can now do very well) does not prevent further cognitive decline (and appears to have very little effect at all).

Plaques, however, do lead to an aggravated response by the immune-equivalent cells in the brain (microglia). There has been a lot of interest in how much this microglia response to plaques triggers some of the symptoms of AD, and this what the research summarised in TFA is interested in also. Despite what TFA implies (it's an overcooked Uni press release) there's been heaps of interest in microglia over the last few years in the AD space.

Ultimately, we still don't know what causes the majority of AD symptoms, and this is a very controversial space. Historically, a number of AD researchers (including one very prominent researcher) staked their reputation on plaques being fully or mostly causative, and refuse to let go of this idea despite the growing body of contrary evidence -- which has created a bit of cognitive hysteresis in the field (and led to vast amounts of arguably wasted pharma R&D). It was a good idea at the time as it was the obvious, most parsimonious explanation; but biological reality has a habit of not conforming with Occam's razor, sadly.

(As for terminal lucidity -- this is not my field; IIUC it's also debated, but your idea is not entirely at odds with one theory. There are many theories, and it's not entirely accepted that this even reliably occurs at all.)