sho

> everything tagged as 'uncensored' or 'abliterated' will get wiped from huggingface

... you do know you can download from huggingface and make your own copy, right? And even if you couldn't, you could run heretic (or whatever approach they used) yourself on the official image. It's all just a well-understood process. A decent abliteration run on DS 4.1 flash will probably cost you about $25k.

They're not small but they're hardly out of reach. I have backups on my home NAS.

> very few people have the legit on-prem hardware resources/budget to run something like an abliterated Q8 of deepseek v4.1 flash 1017

It's something like $300-400/day for the raw hardware on RunPod, a little more on AWS. Not for home users, sure, but for a business wanting to eg check its security against the best of what attackers might bring, that's nothing. And it will go down.

What are you trying to say here? That no-one can ever be accountable for using tools wrong? Always blame the tool, or it's "no true scotsman"?

I'm not sure you've quite thought this through. "Holding it wrong" is very, very often the case, especially with technology.

Pretty clickbait. No fan of Google, but it sounds like they just jumped the gun slightly on the clearing and it would have been/will be perfectly fine once the permit comes in, which I'm sure it will unless their planning process is totally broken. If you sell Google a great big slice of land to build a DC, I don't think it's very surprising they cut down the trees to build the thing you sold them the land to do.

Finland has something like 20 billion trees and plant tens of millions a year for forestry. The bloody country is nothing but trees. 250k isn't nothing but it's hardly the end of the world.

I don't even know what you're trying to say. They do exist, and they're "out" for the people with access. Knowing that this is the case is crucial for decision making, whether you're a company wondering what's coming or a government thinking about how to manage this power imbalance. You can pretend otherwise of course, but I don't see what it buys you.

Don't be snarky. This is very well known, including by public disclosures by the labs themselves, along with countless reliable testimonies by people who would know. It's actually a huge governance problem we're yet to solve.

Most airplane crashes are straightforward pilot error - I don't see why AI couldn't improve on it with a bit of effort. I fully expect AI to be flying commercial planes inside 20 years, indeed I expect human piloting to be looked back on as dangerous and error-prone. The FAA will bow to the numbers, like they did with fly-by-wire and all other technological progress.

Everyone is aware of these schoolboy facts about communism, which is why no-one suggested implementing it. They are instead talking about tweaks to the existing balance, which every developed country has implemented for hundreds of years, and which you as an English speaker certainly come from. So please stop tiring everyone with this teenage straw-maning - you already live somewhere that at least pays lips service to "collective, distributive, balanced control of resources"; we could just do a better job of it.

> Google sez

I see the problem.

Prior to LLMs, Crystal was my language of choice when I needed a distributable binary. It works great for this, and for the kind of packages I was writing - mostly agent-daemons lurking in servers - the networking worked great. The only real alternative was golang, which I've never really liked and has (had?) a more complicated linking story too. The typing worked well too, if used with discipline - if it compiled, I could be pretty confident it would work. I still have systems deployed running Crystal agents, in production, without issue or memory leaks, for years.

LLMs though - I confess they've changed the calculus. Now I no longer have to look at the language, and obey its stupid ideas about error handling, I just use golang for its better library support. But I still have a soft spot for Crystal!

Many modern fighter jets literally cannot be flown by a human without automation, they're too unstable. It's simply not possible for a human to process the inputs and control the outputs in real-time sufficient to keep the fighter in the air. And that's old technology compared to the flight control systems in modern quadcopters - again, literally impossible for a human to pilot.

Suffice to say that what's in current civil jetliners is very far from the state of the art. I fully expect properly implemented and integrated AI to be far superior to human pilots in the future - it probably would be already, if we put the effort in.

> It's been tried many times. Doesn't work.

It's hard to take anyone seriously when they make these kind of ludicrous, absolutist claims. Yes, it's been tried, and continues to be tried, by literally every country on earth, in various formulations. Maybe we haven't quite hit on the winning formula yet, but that's no reason to stop trying.

If you'd read any history, you'd know we'd tried pure unchecked capitalism without redistribution before, too. Late 17th century Britain comes to mind. I don't want to spoil the story too much, but it turns out that "doesn't work" either.

America will 100% not do that, unless they want matching craters in all of their fabs, or worse, in Taiwan’s.

I really dislike attitudes like the author's. I think it's great and important that people and companies spend considerable amounts of their time and money exploring the limits of design and what's possible, even if the concepts produced have little direct, practical use today. That kind of speculative R&D can lead to great things and who knows, maybe in 20 years it will lead to something amazing that will benefit a great deal of people.

Trying to pooh-pooh the ideas of tomorrow because they don't immediately solve all the problems of today is just terrible, short-sighted thinking.

And on a more pedantic note, obviously someone asked for it, else it wouldn't have been made at all. What the author really means is "I didn't ask for a crab chair".

I think the batteries just aren't there yet. You need light for a sports car and current gen battery tech (~0.3KW/Kg vs 10+ for petrol) just suck in this use case (and I say that as, generally, an EV supporter). It'll improve but it'll take a while.

Well that's funny, because you were making some pretty blanket statements upstream about how they "never" do this and "always" do that. And yet your experience seems extremely limited and niche, by your own admission.

Maybe you should experiment a little more. I think you will quickly learn that your previous impression is wrong. The days of them being merely some sort of jumped up autocomplete are years gone.

The joke is that macs are famously slow at prompt prefill and you are not getting anything back in 3 seconds, or probably even 30. Once they get generating, it can be acceptable, but the TTFT is horrendous.

There's a ton of well-understood things Apple can and hopefully will do to massively accelerate every stage of this pipeline and hopefully they're hard at work implementing most of them for m7.

So, that gets us to about where nVidia was with Ampere in 2020. Let's hope the M7 catches us up with at least Hopper.

> On September 22, Apple is releasing the M5 Mac Studio with 256 GB of unified memory [..] it will probably [..] enough to write this snippet of code in 3 seconds

The author has obviously never ran an LLM on a mac! In 3 seconds, it will have possibly started to think about maybe scheduling a date to contemplate the planning timeline for processing the second token in your prompt.

I said machine consciousness might be possible. You said it will never happen. Pointing out that intelligence does not entail consciousness doesn’t establish that machines cannot be conscious. Nor does naming “substrate independence” refute its possibility: you need an argument that consciousness requires physical properties these systems cannot possess. You haven’t supplied one.

When I said "we don't know what consciousness is", I didn't mean humans literally don't know what the experience of consciousness is like, or of its existence. I meant that we don't know by what mechanism the phenomenon arises.

I know what is meant by substrate independence. I quoted it back to you because I didn't know why you were mentioning it; I still don't.

And I have read Tsimerman’s paper when it came out. Curious, I went back and checked - it doesn't even mention consciousness. So no idea why you mentioned that, either.

I think you're deliberately trying to waste my time, so I'm ending it here.

I don't know what you're talking about. "Substrate independence"? Huh? "we did not accidentally pick the right one"? Who or what is "we"?

I'm not claiming panpsychism. I'm saying that when you run a rough simulation of something known to have an emergent internal property - ie, the human brain and consciousness - it's not completely outlandish to suggest that the emergent property might arise in the simulation as well. It's nothing to do with sand, it is the nature of the software being run, be it "on the metal" like us, or somewhat abstracted, like the LLMs.

I don't think this quite reasonable proposition leads to the end of humanity or has anything to do with ethics, and nothing you've said really refutes it (and I am always grateful to be shown how I am wrong).

Preposterous? Insane? My friend, we don't even know what consciousness is, or from what it arises in humans. It could be an emergent property of neural-network-shaped activity - in which case LLM software running on that "sand" is certainly eligible, at least in theory.

These outraged denials you're spouting are based on nothing but emotion. You simply don't know - neither do I, neither does anyone. So stop acting like you do.

Care to supply your rigorous, formal definition of what animacy in an object is?

There's another term for the certain-death variety which I prefer - the "call of the void". I think that captures it nicely.

I'm not sure what point you are trying to make, if at all

That is a completely unsupportable claim. For all we know they might already be, for a brief flash at least.

Well, I'm staying exactly where I am. I'll be keeping my company on 1password, and my family too, because I think it's a good product. I don't personally use Omarchy, but I welcome anything that can bring more people to linux on the desktop, so I support that too. I don't agree with everything DHH says, but I absolutely despise online witch hunts so I guess I support him too just on principle.

Just felt the need to post that.

Nothing weird or exotic - just frequent daily work in two reasonably large applications, with a lot of testing and follow-through. I've found that as codebase size goes up, token usage does too - perhaps quadratically, and a complex back end application just costs a lot to work in, probably because it needs to map out so many pathways and interdependencies.

I tend to work in 5-10 or so parallel streams at a time so that also multiplies token use as a function of time. Much less and I'm waiting for it too much, much more and I can't juggle effectively. My new big project is of course to try to take myself out of the equation further and increase the parallel streams dramatically; pretty hard to get that right so still manual for now.

If you find yourself wishing for more Fable at the subscription rates, just buy another 20x account? I'm currently on 3, spaced out on Sunday, Wednesday and Friday. Just as one is running out the next resets. Problem solved!

I feel the value I get far exceeds $600/m. It's a straight expected value calculation for me and I'm happy to pay. I wish they'd make it easier though - just sell me a 100x account for $1k/m and save the messing around.

Sol is supposed to be 5T according to rumour. The imminent Astra is allegedly 10

Well, do you see the value of writing notes for yourself occasionally, even though they might soon become stale in your constantly changing environment? Yes, right?

Same principle. It's a good idea to have a schedule to clean them up periodically - an idea you can also put into a note.

"Open Source" is still not a word. It is a phrase, a term, maybe even a thing - but not a word!