I want to note that: long prompts are good only if the model is optimized for it. I have tried to swap the underlying model for Claude Code. Most local models, even those claimed to work with long context and tool use, don't work well when instruction becomes too long. This has become an issue for tool use, where tool use works well in small ChatBot-type conversation demos, but when Claude's code-level prompt length increases, it just fails, either forgetting what tools are there, forgetting to use them, or returning in the wrong formats. Only the model by OpenAI, Google's Gemini, kind of works, but not as well as Anthropic's own models. Besides they feel much slower.

I do japanese transcription + gemini translations. It’s worse than fansub, but its much much better than nothing. First thing that could struggle is actually the vad, then is special names and places, prompting can help but not always. Finally it’s uniformity (or style). I still feel that I can’t control the punctuation well.

Recently, I visited the Pennsylvania Railroad Museum and was fascinated to learn that when steel railcars were first introduced—despite being far safer than their wooden predecessors, which could easily be crushed—many people feared they might attract lightning. It's such a good analogue to our movement into AI reality.

Boards are low tech and low profit, does American company and workers even want to do it?

Maybe not, but if the entire country doesn’t have the ability to manufacture it, then it’s still going to be a strategic weakness when push comes to shove. The entire exercise of doing more chip manufacturing in the U.S. is about maintaining national competitiveness and independence. It’s certainly not about cost. So I think it’s a good point that investments should made to be able to onshore the entire stack rather than just the top end.

Or we could strengthen alliances with our neighbors and potentially shift some of that burden to them. Trying to move everything here is not feasible. We simply do not have the human capital or willingness to manufacture every low level widget in the world.

What this administration is doing is not a recipe for success: trade wars with everyone, immigration crackdowns, and unpredictable tariff policy.

EDIT: Oh and hinting at invasion (Greenland, Canada) doesn't help either

I agree.

But Taiwan or the rest of Asia is still a problem given the tensions in the area. If China did something it could seriously effect supply even if it wasn’t an attack on whichever country was supplying us.

We need friends making things in Canada or the rest of the Americas or Europe or Africa or some other place that isn’t China or directly under their thumb.

Even without action by man. The wrong tsunami or whatever could effectively wipe enough out everyone would be screwed.

We need geographical diversity too. The existing alliances we’re burning to the ground don’t solve that.

Yea, I work in the industry. There are players, but not exactly bountiful. Really the backbone of American electronics manufacturing is military spending. If the defense budget went away, there would be close to zero PCB manufacturers left. China makes higher quality boards, faster and for dramatically less money.

This applies for any manufacturing industry to be honest. US shipbuilding capability is so limited compared to China. It's only surviving because of military spending, but not in a healthy way. US made ships are of lower quality and cost much more, compared to European countries. It's the same for cars, busses, airplanes. Whole US policy is blocking the entry of busses manufactured outside NAFTA. US government is keeping Boeing alive by sending POTUS to marketing trips etc etc.

Worse - to manufacture usable boards, you need everything from the CPU socket and northbridge chip down to the dust-mote-sized discrete components that are mounted on it. Plus RAM, and ...

'Most all of which falls square into your "low tech and low profit", from a right-thinking* American company's PoV.

Not to say that a saintly American company could do much better, if it tried to swim uphill against America's vastly-higher cost of living (vs. the countries where most of that stuff's manufactured). And other problems beyond its control.

*profit-obsessed, generally

If there is a reason to want to in-house the fabrication of chips then it seems silly to not extend that to at least the boards that house them, otherwise we wind up still being reliant on an international supply chain which seems to defeat the purpose.

Even if it was just motherboards in particular and not others, that seems like a necessary step in securing the supply chain and if we only do that for national defense the benefits of competition likely won't extend to consumers that are still exposed to trade taxation.

We’re at least 4 years away from that, as it would require a round of STEM college students to go EE instead of computer science.

Agreed, though its realistically much more than four. We have to make going to college a good idea again. That's not a cultural shift that'll happen overnight.

Ask Smoot and Hawley how well that went.

Actually recent email innovation I enjoyed is Mimestream, the macOS native client for Gmail. Apple’s smart inbox is half baked but better than nothing. Cloudflare now also has a pretty good email forwarding service.

The thing is, the publisher are not obligated to show all recent post, some may show 1 some may show 5, but they might publish more than what they show in your refresh interval (e.g. 1 week) In this case, using a stateless reader, you will start to miss article.

Why would you want to expose your IP to the internet? I still feel that's dangerous, susceptible to DDoS attack, and I avoid that as much as possible. I put everything behind a Tailscale for internal use and behind Cloudflare for external use.

In this case they're re-exposing the server(s) to the public internet, but their actual IP Address is still very much hidden behind the Wireguard connection to the VPS.

The IPs they're talking about exposing are ones which are on a VPS, not their home router, or the internal IPs identifying a device in Wireguard.

What the heck? That's like not wanting a street address because people might come to block your front door somehow, or burglars might find your building and steal from it. The big brothers you mention would be like gated/walled communities in this analogy I guess

Saying this as someone who's hosted from at home for like 15 years

Also realise that you're sending the IP address to every website you visit, and in most VoIP software, to those you call. Or if you use a VPN 24/7 on all devices, then it's the VPN's IP address in place of the ISP's IP address...

I don’t think this is the right analogue. Having someone come to your door breaking things would take much larger effort, and easy to be caught. But DDoS or attack your service has minimal cost.

Visiting sites and sending the IP address is not the problem, the router has firewall and basically blocking unwanted attention. But when you expose something without protection and allow someone to burn your CPU, or, in a worse case, figure out your password for a not properly secured service, is a totally another issue.

I saw people setting up honey pot SSH and there are so many unauthorized access and I got scared. I think exposing entire machine to network is like you drive car without insurance. Sure you might be OK, but when trouble comes, it will be a lot of trouble.

> basically blocking unwanted attention. But when you expose something without protection and allow someone to burn your CPU

... sure. You'd think I'd have noticed that in nearly two decades of hosting all different kinds of services if this were a thing

Yeah and of course it will be depend on your personality and risk model. Compared to other things I don’t want to risk my data, whether leaked or damaged. And I make mistakes, a lot. If you are very meticulous and can ensure that you can put up all the security measures yourself and won’t expose something you don’t want to. I am just not that kind of person.

I'm not meticulous either. I had one responsible disclosure and a few times where I noticed issues myself but never that an attacker discovered it first. There's not that many malicious people. The only scenario where you realistically get pwned is when there is a stable and automated exploit for a widely spread service that can be automatically discovered, something like Heartbleed or maybe if a WordPress plugin has an SQL injection or so

Run unattended upgrades, or the equivalent for whatever update mechanism you use, and you'll be fine. I've seen banks with more outdated running services than me at home... (I do security consulting, hence)

To do that people have to physically come to my house and there are solutions to that, people can fuck with my internet from anywhere in the world. It's similar to why remote internet voting is such a pandora's box of issues.

There's 4 billion front doors on the v4 internet. Sending you a DDoS is transient (not like doing something to you physically) and doesn't scale to lots of websites, especially for no gain

In addition to myself, I know some people who self host but not any who ever had a meaningful DDoS. If you're hosting an unpopular website or NAS, nobody is going to be interested in wasting their capacity on bothering you for no reason

Anything that requires custom effort (not just sending the same packets to every host) doesn't scale either. You can host an SQL injection pretty much indefinitely with nobody discovering it, so long as it's not in standard software that someone might scan for, and if it is, then there'll be automatic updates for it. Not that I'd recommend hosting custom vulnerable software, but either way: in terms of `risk = chance × impact` the added risk of self hosting compared to datacentre hosting is absolutely negligible, so long as you apply the same apt upgrade policy in either situation

Online voting has nothing to do with these phantom risks of self hosting

If it is just chores, gamifying it with a receipt printer would be just fine. But these are just such minor stuff compared to the real challenges of life.

The idea is precisely to use our daily tasks to build momentum for the more difficult ones.

I think Amazon, Meta have been trying on inference hardware, they throw their hands up on training; but TPUs can actually be used in training, based on what I saw in Google’s colab.

I remember crowdstrike outage offers starbucks coupons? that’s way to go.