__MatrixMan__

Not sure why you're being downvoted. This is a very reasonable proposal. Once you're beyond a certain level of rich, you should be constrained by whether the people around you consent to your behavior, not how much money you have.

What eats CEO's?

Markets exist to meet the needs of the people who use them. Subverting them such that it's a negative sum game where profits improve marginally while material conditions deteriorate rapidly is in fact antisocial behavior.

It's probably not time to build a guillotine for the memory company CEO's and top shareholders but their behavior does resemble other more problematic moves. Consider United Healthcare denying claims in contradiction with their own policies just because they've realized they can get away with it. It has the same shape relative to the adjacent incentives.

My case for why Elixir is especially suited for use with AI: https://news.ycombinator.com/item?id=49835781 (summary: it's mostly about the BEAM, which is more of a k8s killer than a rust killer).

I don't think this really applies to any of the benefits I've described there:

> AI can replicate a lot of its features in Rust for one-tenth of the cost and 10 times the speed

You can declaratively manage your phone with something like github.com/devindudeman/nix-android so you'd point at a decoy config file and rebuild before crossing the border and then point at the up-to-no-good config file once across.

Of course you do have to be able to reference those files somehow, so they have to be online, but if there are no references to them on the device then the border folks would have to somehow know about them and be able to access them in order to link you and your device to the declared configs. There are a lot of ways to hide such a thing.

Presumably the most sensitive thing here is your social graph and messages. I haven't verified if nix-android handles them in a way that would make sense here, but it's at least plausible that it might.

I dream of a future where the disk image in my laptop both functions as an operating system, but can also be decrypted into secret files. We'll keep the extra entropy in a pile of AI slop next to the other piles of AI slop. That way, under scrutiny, it just looks like an OS full of AI slop.

Maybe what we need is an unauthorized client, like nitter. Then we can just remove the feed.

Like termites, moving on to the next tree. One day we'll figure out how to be rid of these parasites, then we'll have healthier trees.

I'm sure jobs are an important aspect here, but I kind of doubt that we can fix this without also finding a way to refocus our efforts on endeavors that most people approve of. As it is, letting the US succeed is bad and letting it fail is worse. That's unlikely to remain stable forever.

Reminds me of LIGO, which injected fake results until the humans got good at handling them, then later announced that one of the results wasn't fake. Seems at first like an upside-down way to discover gravitational waves, but I think it's a pattern worth exploring in more cases.

To paraphrase Doctorow:

> This is why the TSA are some of the most water-bottle-findingest **ers, but regularly overlook firearms when a red team takes one through. Humans do not stay vigilant to signals which almost never happen.

In some more enlightened timeline there's a spirited competition between institutional and amateur astronomers towards predicting these, and some amateur is sitting on a $10M this-person-told-us-so-and-we-bet-against-them prize, which gets bigger every year (no prizes for the institutional side, they get a salary).

In that timeline's future we're then competing over who can most reliably prevent them.

The winners of such frivolous competitions over protecting the moon get funding for the more serious job of protecting Earth.

If you're stressed out all week and you pick up an infection then your immune system might not settle into fight mode until the weekend, because you can't afford to be down for the count when what you really need right now is to run from the bear, except it's not a bear, it's just the grind.

If there are more potential infections out there than there were before (also, greater population sizes to communicate them) that's a higher concentration of weekends lost to sickness.

That's gonna have an impact on how often you go to a friend's house... squared, because it's both you and that friend. Like the network effect's evil twin.

It's gonna have to be so good that it creates entire industries that didn't exist before. That would be awesome, but I'm not seeing it so far.

I think figuring out how to make everybody happy is fun. You end up exploring options you never otherwise would.

I'd be interested to see how the drop correlates with pets. We got a dog during COVID, she acclimated to a house which never has guests. Now she can't handle guests. Shame on us, perhaps, but anyway there it is.

You can use other generic dev env managers like mise, or a language-specific solution like uv or npm, or a container or a vm... it's a widely available capability that I'm talking about here. There's nothing to do with alignment, it's just about making sure that if you depend on some bits it's known that you depend on precisely those bits, and its quite helpful for making agents useful while still sandboxed.

If that were the case there would be more of them in jail.

I'm excited for ladybird to reach alpha

HuggingFace's opinions shouldn't be a blocker.

If I run over your mailbox maybe we can come to an agreement where you don't sue me, but if I was driving recklessly I've still committed a crime that you cant absolve me from.

It should be the same re: losing control of your agents.

> Sure, creation is much cheaper, but anyone who has run software in production at scale knows that maintenance, reliability, security, scalability, etc. is the majority of the cost

Costs which largely go away if the software you're using is so custom that it's only applicable to your six person team anyhow. That wasn't feasible before, but it is now.

The previously fashionable one-size-fits-millions approach to software hasn't treated users well enough to expect them not to defect when they're suddenly able to go it alone. For many, the quality issues are worth tolerating, because they're still less painful than something which was designed to be sold rather than to be used.

Feels like survivorship bias to me. CEO's fail up, except those that fail down, and the latter group aren't CEO's anymore, so they're less likely to show up in a sample.

If your project truly depends on those things, they should be declared dependencies. Presumably you have some tool for injecting such things into a shell that the agent can use (I use nix for this). So if you run the agent from that shell, it has what it needs. If the shell doesn't have what it needs, that's a bug which the agent can fix by declaring new dependencies, but you have to relaunch the agent in the updated shell--so there's your opportunity to weigh in on whether the new resources are appropriate.

The benefits of being persnickety about precisely defined dependencies have outweighed the headaches since long before agents came on the scene. Agents have just made it even more important to do so, because if you let them fetch things all willy nilly like you'll have "works on my machine" problems at a much greater rate than was previously possible.

I don't see why it needs wide unattended access. There's no getting around spending some human time on expressing your wishes and constraints, but we have choices about what form that takes. Markdown files and wide access seems to work, but so does custom handcuffs for each job. You just have to shift your guidance out of documentation and into interactive help, error messages, or other facets of the handcuffs (e.g. a custom CLI for this task which is the only way for the agent to act outside of its sandbox).

We're mining https://universalpaperclips.fandom.com/wiki/Yomi now

Nature is not, but scientists are. A theory which humans can map onto experiences at their scale and use to make predictions via intuition is superior (for use by humans) to a theory which makes accurate predictions for inscrutible reasons--supposing their predictive power is equivalent. Beauty, I guess they call it. It leads to improved decision making, experimental design, etc.

I did some more reading and it turns out I've misunderstood flash memory. I thought it was read via tunneling, but it's written via tunneling. I appologize.

This being the forum that it is, I was hoping to find an example of this sort of thing: https://pmc.ncbi.nlm.nih.gov/articles/PMC4066974/ but in electrical engineering instead.

Capsid assembly is a good example if you want to make a macroscopic analog that works over less than geologic time, but the next step is for the assembled capsid to find and ingest a nucleic acid containing a viral genome as it floats around a broth containing all sorts of other nucleic acids.

It's wildly nonintuitive that this manages to happen. Maybe you're right, and it can be explained by the numbers being so tremendously large that they break intuition without the need to import any nonintuitive quantum weirdness.

But if so, isn't such an explanation problematic in its own right? It banishes:

> The quantum world is impossibly nonintuitive, and it's having an effect

in favor of

> The scales here are impossibly nonintuitive, and that's having an effect

Like... we're still shoving the explanation under a rug, it's just now it's a different rug. Either the quantum effects explanation or the large numbers explanation needs further development before we can call this phenomenon adequately explained, and rather than doing that development it seems like the whole field is dedicated to championing their favored side as-is.

I hope the result of cases like this is a law or legal precedent which ensures that you waive any intellectual property rights on a model which was trained on somebody else's intellectual property.

There's a lot of waste associated with preventing distillation. It's a distraction that goes away if we just compel the makers of these trained-on-everything models to publish their weights publicly.

To preserve competition maybe we compromise and give them a three month grace period.

You're right on both points. The article isn't misleading, I just think it's an entry into a space which has overreacted to quantum woo woo in an unhelpful way.

Take flash memory for example. We read the bit via quantum tunneling. We don't have to wait for the electron to actually bounce out of its little box, we can detect the small probability that it might, and then we can act on that detection.

So every time I act on the basis of such a read event, am I not being guided down one of gulleys that were set up by the designer of my device (or, analogously for biology, by evolution)?