NathanKP

I suspect you are greatly overestimating the average organization's ability to run a Git server themselves and keep it secure, while also overestimating how evil GitHub and LLM's providers are.

Starlink residential service is about 10x slower than my home fiber connection at the same price. How is this technology going to compete with data center level infrastructure even in an optimistic scenario?

Where I live in New Zealand the only good internet provider is Starlink, so all my internet is through Starlink. The latency is about 20ms, so while yes you are technically correct that is 10x slower then 2ms, it isn't a major deal breaker.

You are probably thinking about light and fast API calls, where latency is more noticeable. But if you are doing an inference or LLM job that is going to take several seconds of token generation before the full response is available then the difference between 3.002s and 3.020s is negligible.

build some data centers underground in deep fortified bunkers if you want.

This is the defensive build. The US has a tendency to optimize for offense.

Let's roll forward another few decades, and imagine a classic dystopia scenario: pervasive worldwide surveillance systems, armed drones and robots everywhere, etc. Where does the data from those surveillance systems get crunched, and where do the drones and robots get controlled from? Probably not from one central system in one place. That would ironically be too high latency. These systems would most likely end up as generic shells with minimal on board smarts, controlled by AI "brains" up in LEO, 20ms away via radio. The AI observes from overhead via it's surveillance systems, and it acts via it's robot bodies down on Earth.

It sounds like sci-fi, but you have to remember the world is full of megalomaniac nerds. They love this type of stuff, and if they think someone might be able to build it, then they want to be the one to build it.

It's easy to see the benefit in DC's in space if you look at a few ingredients:

1. The recent Iran drone attacks on AWS data centers

2. Growing anti-AI and anti data center sentiment at home, plus Larry Fink (ceo of Blackrock) in a recent interview being terrified of dissident groups using consumer drones to attack data centers.

3. Anthropic, Grok, and other AI vendors becoming more and more integrated into defense and military, plus increasingly reliance on AI for other national surveillance systems

Data centers are and will be targets, both for national military attacks as well as home grown dissident attacks, so they are proposing to move some of the critical workloads to somewhere that the only group that can attack the data center hosting the workload is a nation state with space launch capabilities. That significantly reduces the number of actors that can attack the data center. And if the US wants to they can probably bomb all the other space rocket launch facilities worldwide in less than 24 hours, leaving extremely limited capability to attack a space hosted DC.

Is it insane? Probably, but the US has done insane things with military budget before, and will continue to do so for a long time. If you are Elon, its a great time to milk that US defense budget for some more R&D, and even if the main project doesn't work out, he's still going to be able to keep some innovations within the company and apply them to Starlink and other more realistic endeavors.

Agreed. The "Tesla backed into objects, one into a pole or tree at 1 mph and another into a fixed object at 2 mph" stood out to me in specific. There is no way that any human driver is going to report backing into something at 1 or 2 mph.

While I was living in NYC I saw collisions of that nature all the time. People put a "bumper buddy" on their car because the street parallel parking is so tight and folks "bump" the car behind them while trying to get out.

My guess is that at least 3 of those "collisions" are things that would never be reported with a human driver.

Bingo. I elaborated on this idea more in my comment: https://news.ycombinator.com/item?id=46884246

People are acting like a space data centre would be running a traditional workload. No, it's probably running a military one, some sort of AI powered modern version of Dead Hand (https://en.wikipedia.org/wiki/Dead_Hand). Autonomous warfare could get real dark, real fast.

My dark theory is that the goal is to run an AI overlord in space such that it is difficult to counter it from Earth.

If you assume that these people aren't completely stupid, then there is some reason why they want this workload running at great physical distance from all the people down on Earth. It's probably not to protect people on Earth. After all they'll happily deorbit satellites and other junk from orbit and let it rain down on us. And they will happily destroy the environment with all those rocket launches too. Therefore it must be to protect the workload from us.

What is a workload that is something that people would probably want to destroy, and which would also provide enough value to offset the expense to launch and run in space? The only thing that might make sense is a military AI platform. Think something that observes Earth, launches missiles, and controls terrestrial drone armies remotely, with relatively low latency.

It gets built and launched thanks to endless military budget, and once it is up there, running such an AI from space means that effectively the only people who can take it out are nation state level foes who can launch rockets into low earth orbit. And this thing is a satellite, probably part of a network that is watching the Earth all the time. Start building something that looks like a rocket launch site, and the AI will see, then you get hit by a missile or taken out by a drone first before you get a chance to attack the platform.

It sounds like sci-fi, but in the future, if we let it happen, there could absolutely be nearly invulnerable autonomous AI platforms in space overseeing everything, and making decisions, and issuing commands. Of course there could still be a massive solar flare event, or a Kessler syndrome event that releases us all from AI enforced servitude. Anyway, it's a not so fun thought experiment, and let's hope this stays sci-fi, so we can just enjoy a fun Hollywood film about this rather than experiencing it firsthand.

So far in this thread the majority of Hacker News users have decided they don't like the tone of your comment, therefore they have downvoted your comment.

If you don't see the irony, let me point it out: this is a real time demonstration that your following statement is not an inevitable thing:

eventually converges on the same thing in the end: a place to dump out ugly things

When a community has the tools for self governance, then it can resist influences that it does not appreciate. In the case of Bluesky in specific, you may not be familiar with how easy it is to subscribe to labelers and blocklists. This decentralized self governance model allows anyone to curate their Bluesky experience, and it allows sub communities to collectively govern in ways that filter the cesspool and remove the ugly things.

In short, the fact that you are complaining about downvotes while simultaneously saying that it is inevitable for communities to devolve into places to dump ugly things is highly ironic. One thing that you are complaining about, is the solution to the other thing that you are complaining about.

You are getting downvoted for tone. By tone I mean the polarizing terms like: "cesspool of stupid opinions" and "low-thinking idiots", etc.

When social communities allow the type of tone you are using, that is precisely how they end up worse over time.

Hacker News downvoting comments like yours is how it has maintained quite high quality over the last 16 years I've been on this website.

No.

This is a real issue. If a robot is fully AI powered and doing what it does fully autonomously, then it has a very different risk profile compared to a teleoperated robot.

For example, you can be fairly certain that given the current state of AI tech, an AI powered robot has no innate desire to creep on your kids, while a teleoperated robot could very well be operated remotely by a pedophile who is watching your kids through the robot cameras, or attempting to interact with them in some way using the robot itself.

If you are allowing this robot device to exist in your home, around your valuables, and around the people you care for, then whether these robots operate fully autonomously, or whether a human operator is connecting via the robot is an extremely significant difference, that has very large safety consequences.

"the disconnect between a solid economy and an anxious public"

Maybe... the economy isn't actually as solid as he thinks?

In case anyone else is wondering about practical ways to reproduce these effects, I did some quick searching:

Most chocolate / cocoa products are processed in a way that destroys 80%-90% of the flavanols. You either have to buy specialized high flavanol cocoa powder (what the study used), or you would have to be consuming multiple cups of matcha tea, or squares of dark chocolate ever day. You'd likely also want to add high flavanol foods like blueberries, blackberries, and cherries to your daily diet.

As someone who spends a lot of time sitting, and also has a family history of heart issues, it sounds promising. I'm planning to give it a try.

This is coming from Swiss Re, which is the world's largest insurance company for insurance companies.

Basically you pay an insurance company premium so that if you have a health emergency the insurance company will take on the cost of your emergency.

Insurance companies pay Swiss Re, so that if the insurance company faces a financial squeeze from unforeseen mass disaster, then Swiss Re takes on the cost.

Swiss Re is basically warning their clients (insurance companies) that Swiss Re is seeing an ongoing trend of excess deaths post Covid, though they expect it to trail off by 2033. They highly recommend their clients factor that in when they calculate what premium they need to charge to be profitable.

I'm more fascinated by the question of whether it scales up... imagine much smaller and more efficient electric engines for cruise liners and cargo ships.

You'd be surprised how technical farming can be. Us software engineers often have a deep desire to make efficient systems, that function well, in a mostly automated fashion, so that we can observe these systems in action and optimize these systems over time.

A farm is just such a system that you can spend a lifetime working on and optimizing. The life you are supporting is "automated", but the process of farming involves an incredible amount of system level thinking. I get tremendous amounts of satisfaction from the technical process of composting, and improving the soil, and optimizing plant layouts and lifecycles to make the perfect syntropic farming setup. That's not even getting into the scientific aspects of balancing soil mixtures and moisture, and acidity, and nutrient levels, and cross pollinating, and seed collecting to find stronger variants with improved yields, etc. Of course the physical labor sucks, but I need the exercise. It's better than sitting at a desk all day long.

Anyway, maybe the farmers and shepherds also want to become software engineers. I just know I'm already well on the way to becoming a farmer (with a homelab setup as an added nerdy SWE bonus).

I'm right behind you on the escape to the mountains idea. I've actually already moved from the US to New Zealand, and the next step is a farm with some goats lol.

That said... I don't necessarily hate what AI is doing to us. If anything, AI is the ultimate expression of humanity.

Throughout history humans have continually searched for another intelligence. We study the apes and other animals, we pray to Gods, we look to the stars and listen to them to see if there are any radio signals from aliens, etc. We keep trying to find something else that understands what it is to be alive.

I would propose that maybe humans innately crave to be known by something other than ourselves. The search for that "other" is so fundamentally human, that building AI and interacting with it is just a natural progression of a quest we've already been on for thousands of years.

Yep, it's just a question of whether on average the "new thing" is more good than bad. Pretty much every "new thing" has some kind of bad side effect for some people, while being good for other people.

I would argue that both Tesla self driving (on the highway only), and ChatGPT (for professional use by healthy people) has been more good than bad.

It is estimated that more than one in five U.S. adults live with a mental illness (59.3 million in 2022; 23.1% of the U.S. adult population).

https://www.nimh.nih.gov/health/statistics/mental-illness

Most people don't understand just how mentally unwell the US population is. Of course there are one million talking to ChatGPT about suicide weekly. This is not a surprising stat at all. It's just a question of what to do about it.

At least OpenAI is trying to do something about it.

Running stuff on an underpowered Raspberry Pi is a good way to sniff test whether an infrastructure or software setup is sane. Powerful computers can hide horrible decisions for a long time, while less powerful devices make it immediately obvious if you need to switch a more efficient configuration.

I think this is one of the main reasons why Raspberry Pi has such a strong representation in homelabs, including my own.

Objects that need new hard drives every 3-5 years to store, replenished on a constant cycle

The replenishment of these hard drives is baked into the cost of S3. If there is a major disruption of hard drive supply then S3 prices will definitely rise, and enterprises that currently store lots of garbage that they don't need, will be priced out of storing this data on hard drives, into Glacier or at worst full deletion of old junk data. That's not necessarily a bad thing, in my opinion.

There is lots of junk data in S3 that should probably be in cold storage rather than spinning metal, if merely for environmental reasons.

ChatGPT 5 still says "My knowledge cutoff is June 2024"

There is a reason these models are still operating on old knowledge cutoff dates

I just ended an eight year stretch of working for AWS. I quit in order to move out of the United States to New Zealand, so indirectly I quit over RTO. I wanted to work outside of a US hub city, even if it would have required relocating to AWS New Zealand and taking a resulting pay decrease that would have saved the company significant money to get the same amount of work from me.

Acquisition and retention of good talent is absolutely a major issue for AWS. Don't get me wrong, I still like AWS a lot, even all it's frequently chaotic mess, but I'll probably wait until Amazon starts its Satya era before I'd consider reapplying to work there.

The graph showing the difference between home price index, and consumer price index does not consider that many of the items in the consumer price index are heavily subsidized by the government.

For example notice the item categories in the index: https://www.bls.gov/news.release/cpi.nr0.htm

And then compare them to the companies in the subsidy tracker: https://subsidytracker.goodjobsfirst.org/parent-totals

So another story we could be seeing here is that heavy government subsidies are barely managing to keep consumer prices down in many categories, except for housing.

My general opinion: we don't have just a housing emergency. We have a general emergency across many, many categories. If the government stops all these subsidies we'd see prices for everything else skyrocket across the board to match housing prices. Then wages would be forced to rise too, and you'd see the true underlying crisis: hyperinflation. Houses are worth so much because they are one of the best hedges against hyperinflation. If the US dollar gets inflated my house suddenly gets very easy to pay off and its now my primary form of wealth. So no wonder housing is so expensive.

I've seen this regularly on fringe articles that are clearly being manipulated. I don't have direct links right now, but things I have seen in the past:

* A sketchy online university that was clearly manipulating their Wikipedia page with lots of positive information about themselves to suppress info about their active lawsuits and controversies

* On medical topics: non scientific, baseless claims about the efficacy of various herbal treatments, vitamin supplements, or other snake oil treatments.

* On various fringe politicians. Someone clearly rewrites the article or adds additional things to the article with claims about what the politician has done or not done or wants to do, but these claims are arguably not fact based.

Now these things usually don't last for a long time. They do get rolled back or removed. But it doesn't have to be on there long for it to be utilized. For example, someone just needs to modify the Wikipedia page long enough to get through their active lawsuits, or the snake oil salesman just needs their info up on Wikipedia for long enough to use it to increase their perceived authenticity to trick some seniors. There is such a constant stream of bad actors trying to put this stuff out there that you'll see it eventually, and it doesn't even have to be up there for long for it to be harmful.

I think the explanation is simple: there is a direct correlation between being too lazy and demotivated to write your own code, and being too lazy and demotivated to actually finish a project and publish your work online.

The same people who are willing to go through all the steps to release an application online are also willing to go through the extra effort of writing their own code. The code is actually the easy part compared to the rest of it... always has been.

To be clear I did not have a 95% acceptance rate. I'm saying that in the final published repo, 95% of the lines of code were written by AI, not by me. I discarded and refactored code along the way many times, but I did that by also using the AI. My end goal was to keep my hands off the code as much as possible and get better at describing exactly what I wanted from the AI.

if you are accepting 95% of what random output is being given to you

I am not, and don't expect to be able to do that for many years yet. The models aren't that good yet.

I would estimate that I accepted perhaps 25% of the initial code output from the LLM. The other 75% of output I wasn't satisfied with I just unapplied and retried with a different prompt, or I refactored or mutated it using a followup prompt.

In the final project 95% of the committed lines of code in the published version were written by AI, however there was probably 4x as much discarded AI generated code along the way that was also written by AI. Often the first take wasn't good enough so I modified it or refactored it, also using AI. Over the course of using the project I got better at providing more precise prompts that generated good code the first time, however, I rarely accepted the first draft of code back from Kiro without making followup prompts.

A lot of people have a misguided thought that using AI means you just accept the first draft that AI returns. That's not the case. You absolutely should be reading the code, and iterating on it using followup prompts.

That is interesting. So far we are just using the task list to keep track of the list of implemented tasks. In the long run I expect there will be an even more rigorous mapping between the actual requirements and the specific lines of code that implement the requirements. So there might be a fourth file one day!

Coding standards / style guide are both part of the "steering" files: https://kiro.dev/docs/steering/index

False. In order to maintain high quality I often rejected the first result and regenerated the code with a more precise prompt, rather than taking the first result. I also regularly used "refactor prompts" to ask Kiro to change the code to match my high expectations.

Just because you use AI does not mean that you need to be careless about quality, nor is AI an excuse to turn off your brain and just hit accept on the first result.

There is still a skill and craft to coding with AI, it's just that you will find yourself discarding, regenerating, and rebuilding things much faster than you did before.

In this project I deliberately avoided manual typing as much as possible, and instead found ways to prompt Kiro to get the results I wanted, and that's why 95% of it has been written by Kiro, rather than by hand. In the process, I got better at prompting, faster at it, and reached a much higher success rate at approving the initial pass. Early on I often regenerated a segment of code with more precise instructions three or four times, but this was also early in Kiro's development, with a dumber model, and with myself having less prompting skill.

The original submission to HN stated that it was from Amazon / AWS in the title of the submission, however that has since been edited by a moderator to match the title of the blogpost, which does not mention Amazon / AWS.

To be clear, we have no intent to hide that Kiro is from Amazon / AWS, that's why you'll see Matt Garman, for example, posting about Kiro: https://www.linkedin.com/feed/update/urn:li:activity:7350558...

However, the long term goal is for Kiro to have it's own unique identity outside of AWS, backed by Amazon / AWS, but more friendly to folks who aren't all in on AWS. I'll admit that AWS hasn't been known in recent years for having the best new user or best developer experience. Kiro is making a fresh start from an outsider perspective of what's possible, not just what's the AWS tradition. So, for example, you can use Kiro without ever having an AWS account. That makes it somewhat unique, and we aim to keep it that way for now.

I have a personal deployment of the game, but it costs money to run the LLM so I'm not sharing that with all of Hacker News haha. I've got an appsec ticket open to host an "official AWS" version where AWS pays the LLM bill, but that might take a while longer to get approved. For now the best way to experiment is playing with it locally.

I'm also thinking of creating a fork of the project that is designed to run entirely locally using your GPU. I believe with current quantized models, and a decent GPU, you can have an adequate enough fully local experience with this game, even the dynamic image generation part.