rthnbgrredf

> Because it's actually fairly slow.

"We were reading data at 635 GiB/s. We broke 15 million 4k random read IOPS."

Source: https://ceph.io/en/news/blog/2024/ceph-a-journey-to-1tibps/

I don't know man, I think 15M random read IOPS is actually quite fast. I've built multi million IOPS clusters in enterprise settings all on nvme in the past.

I think something around those lines:

Write a comprehensive technical blog post targeting small and medium-sized businesses that use Hetzner Cloud and are evaluating lightweight Kubernetes alternatives. Compare and analyze the following solutions: k3s, MicroK8s, Docker Swarm, and Minikube.

Structure:

Start with a title and abstract introducing the comparative scope.

Provide sections such as:

Architectural Requirements, Deployment & Lifecycle, Management, Cost Analysis, Security & Compliance, Developer Experience, Strategic Recommendations and end with a clear conclusion and deployment advice.

Style:

Technical but accessible to DevOps engineers or small startup CTOs.

Use citation-style references for data (e.g., [1], [2], etc.).

Incorporate code examples (Terraform, Helm, CLI snippets).

Use tables for cost comparisons or benchmarks where appropriate.

Mention open-source tools specific to Hetzner (e.g., hetzner-k3s, hcloud Terraform provider).

Tone:

Professional, analytical, and informative. No marketing fluff.

This reminds me of https://www.adit.io/posts/2013-04-17-functors,_applicatives,...

I think over the recent years, there's been a rise in typed languages that support functional programming like TypeScript and Rust. It will be interesting to see if this trend continues in the context of AI assistant programming. My guess is that it will become easier for beginners, and the type systems will help to build more robust programs in cooperation with AI.

I say the author tells us more than the headline or first sentence of the loop. If you have recently scrolled through Sabines posts on Twitter, or her clickbaity thumbnails, facial expressions and headlines on YouTube [1], you would see that she is all-in for clicks. She often takes a popular belief, then negates it, and throws around counter-examples of why X or Y has absolutely failed. It's a repeating pattern to gain popularity, and it seems to work not only on Twitter and YouTube, but even here on Hacker News, given the massive amount of upvotes her post has.

[1] https://youtube.com/@SabineHossenfelder/featured

I instruct sonnet-3.7-thinking in Cursor Agent mode to draw latex tikz diagrams for me and that works iteratively quite well.

Draw a circle, add a label to the circle, draw a box right to it, draw an arrow from the circle to the box, and so on.

I'm still not convinced that this isn't a tokenizer issue.

Were you able to find a substantial number of questions that do not fall into the letter countinh or word shuffling domsin - problems that are clearly unrelated to the fundamental tokenizer issue of modern LLMs? Otherwise, I would argue that your paper simply proves that the issue still exists.

Category theory and set theory are deeply interconnected, as both are foundational areas of mathematics but with different focuses. Set theory studies collections of elements and their relationships, serving as the groundwork for much of mathematics, while category theory abstracts and generalizes these ideas to focus on structures and their relationships.

One key connection is that the category of sets, called "Set," is a fundamental example in category theory. Its objects are sets, and its morphisms are functions between them. This shows how set theory can be seen as a special case of category theory. At the same time, many categories studied in category theory have underlying set structures. For example, groups, rings, or vector spaces are often built on sets with additional structure.

Another connection lies in how category theory generalizes set-theoretic ideas. Concepts like products, coproducts, limits, and colimits in category theory extend the familiar notions of Cartesian products or unions in set theory. The Yoneda Lemma, a cornerstone of category theory, relies heavily on set-theoretic intuition, as it connects abstract categorical concepts to concrete representations in terms of sets.

Topoi, a concept in category theory, also bridge the two fields. A topos generalizes set theory by providing a categorical framework that behaves like the category of sets but with additional logical structure. This allows set-theoretic reasoning to be carried out in a more abstract setting. For instance, the category of sets itself is the prototypical example of a topos.

However, there are also differences. Set theory often grapples with "size" issues, such as distinguishing between sets and proper classes. In category theory, similar concerns arise, especially with large categories, like the category of all sets, which cannot itself be a set. Set-theoretic tools are often used to handle these size issues by defining notions like small and large categories.

And there’s a philosophical aspect. Set theory forms the traditional foundation of mathematics with its ZFC axioms, focusing on elements and collections. Category theory offers an alternative foundation by emphasizing relationships and structures, often referred to as "structural mathematics." These perspectives aren’t in opposition but rather complementary, with each offering tools to understand mathematics from different angles.

In essence, category theory builds on and extends set theory while also relying on it for foundational concepts. The two fields are tightly intertwined, with set theory providing a concrete framework and category theory offering an abstract, flexible way to study and generalize mathematical structures.

Correct me if I'm wrong, but wasn’t there a similar reactor on the other side of the globe, in Antarctica, around the same time?

Ansible is better than Bash if your goals include:

* Automating repetitive tasks across many servers.

* Ensuring idempotent configurations (e.g., setting up web servers, installing packages consistently).

* Managing infrastructure as code for better version control and collaboration.

* Orchestrating complex workflows that involve multiple steps or dependencies.

However, Ansible is not a container orchestrator.

Kubernetes (K8s) provides capabilities that Ansible or Docker-Compose cannot match. While Docker-Compose only supports a basic subset, Kubernetes offers:

* Advanced orchestration features, such as rolling updates, health checks, scaling, and self-healing.

* Automatic maintenance of the desired state for running workloads.

* Restarting failed containers, rescheduling pods, and replacing unhealthy nodes.

* Horizontal pod auto-scaling based on metrics (e.g., CPU, memory, or custom metrics).

* Continuous monitoring and reconciliation of the actual state with the desired state.

* Immediate application of changes to bring resources to the desired configuration.

* Service discovery via DNS and automatic load balancing across pods.

* Native support for Persistent Volumes (PVs) and Persistent Volume Claims (PVCs) for storage management.

* Abstraction of storage providers, supporting local, cloud, and network storage.

If you need these features but are concerned about the complexity of Kubernetes, consider using a managed Kubernetes service like GKE or EKS to simplify deployment and management. Alternatively, and this is my prefered option, combining Terraform with a Container-as-a-Service (CaaS) platform allows the provider to handle most of the operational complexity for you.

I think we need to distinguish between two cases:

For a hobby project, using Docker Compose or Podman combined with systemd and some shell scripts is perfectly fine. You’re the only one responsible, and you have the freedom to choose whatever works best for you.

However, in a company setting, things are quite different. Your boss may assign you new tasks that could require writing a lot of custom scripts. This can become a problem for other team members and contractors, as such scripts are often undocumented and don’t follow industry standards.

In this case, I would recommend using Kubernetes (k8s), but only if the company has a dedicated Kubernetes team with an established on-call rotation. Alternatively, I suggest leveraging a managed cloud service like ECS Fargate to handle container orchestration.

There’s also strong competition in the "Container as a Service" (CaaS) space, with smaller and more cost-effective options available if you prefer to avoid the major cloud providers. Overall, these CaaS solutions require far less maintenance compared to managing your own cluster.

If docker compose is not enough I would suggest to look into some of the lightweight k8s distributions like k3s, with low ressource consumption and able to operate on a single node.

I doubt that you can do it cheaper. To permantenly archive the whole internet is an ongoing task that basically requires a small company, thats why Internet Archive (169 employee) exist (which costs more than 2.5 million dollar per year). It is not done with buying a huge bunch of disk. Setting up a permanent stream to S3 would be the only solution I can think of a single human could handle.

Well buying 220pb of storage space is really not the problem nowadays, at least from a cost perspective. But you need to maintain all that stuff. What happens when a disk goes broke, what if a network switch goes broke, how do you update your software at scale and so on.

I think it would be best to put it on AWS S3 Glacier Deep Archive for about 2.5 million dollar per year.

Sure, but how would such a proof look like? How could one prove to be the real Nakamoto?

Could you provide some instances of how hard evidence might look like?

Even better is to just focus on the here and now. The past cannot be changed and the future is just a mental construct, both can cause anexity and endless thinking.

Will Terraform and OpenTofu diverge in the future to the point where they are no longer compatible or interchangeable?

No reasonable person should trust the security of an iPhone or Android when it comes to state actors. These devices likely offer sufficient protection against common theft, but they are vulnerable to sophisticated attacks from government entities.

For this purpose you would likely not buy ordinary consumer disks but rather bullet proof enterprise HDDs. Otherwise a signifcant amount of the 5556 disks would not survive the first year, assuming the are under constant load.

Such services already exists. I don't want to promote any in particular, but if you do a research on pay-as-you-go inference of e.g. mixtral or llama3 you will find offerings that offer an API and charge just cents for XY amount of tokens, exactly as OpenAI does.

I'm a strong advocate for writing tests first, maintaining a robust QA process, and ensuring all merge requests undergo peer review. Since I started using GPT-4 for coding, the quality of my merge requests has remained the same. The difference is that I can now produce about twice as many merge requests, as GPT-4 handles the boilerplate writing, Stack Overflow searches and helps to start get going in case of a mental blockade or missing idea.

Does someone know how long an unmanned unclear powered aircraft could fly?

I'm burnt out of calling every new piece of technology that starts to get wide adoption a hype or bubble. Yes, there will be the usual phase of over excitement before it all settles, that we all know is part of human nature, so what? In the Gartner hype cycle modell I'm already in the plateau of productivity with my use of GPTs for code generation, technical troubleshooting and everyday life questions.

I think it is not so much surprising that large language models are particularly good at languages, including programming languages.

Thanks to LLMs, we are quite close to achieving this. I can write code in Python (sometimes even plain english), and GPT can convert it to Go or even Haskell if I like. The conversion is accurate 95% of the time on the first attempt in my use cases, and I expect this to improve further with more powerful models in the near future.

> I think that a well-architected large codebase

I agree with you, but the problem is that such codebases rarely exist in practice. This is partly due to laziness and time pressure. When high standards are not maintained or time becomes a constraint, developers start to make compromises in large codebases, such as placing or connecting functionality in inappropriate parts of the codebase. This is much harder to do when access is restricted by an API (contract) between teams, as it forces communication.

Additionally, if parts of the codebase are completely messed up, refactoring in large codebases can be extremely difficult due to complex interconnections with numerous other components. In contrast, with a microservices architecture, if you have 50 microservices and five are poorly designed, you can rewrite them from scratch while ensuring the new versions maintain the same API. This ensures that the other 45 services adhere to the API contract and cannot interfere in any other way.

The best codebase is the one you fully understand. I prefer codebases that are small enough to understand within a week. This is why I like Microservices. Large codebases can be overwhelming and even senior developers working a decade in the company of might not fully understand them. Instead, I prefer maintaining a few Microservices that our team fully comprehends, where the entire codebase fits into a clear mental model. We then interact with other codebases, that have active mental models in other teams, via APIs.

There was a russian scientist that did research on domestication of fox's. http://en.wikipedia.org/wiki/Dmitri_K._Belyaev

He found that by breeding and selecting for the more human friendly animals they became tamer and tamer. Now they are to the point of domestication. True domestication as I have heard from some means that the animal cannot thrive in the wild as its wild counter part can. This is a technical labeling issue as in some dogs have reversed domestication on some islands etc.

The fox story in russia is interesting since they found that there was a reduction in adrenaline in the fox's as they became tamer and tamer. The reduction caused them to adopt traits similar to domesticated dogs: wagging tails, floppy ears, coat color change. They did embronic transplants and found wild mothers with tame embryo's raised tame offspring and the opposite was also true. So it was a genetic change that resulted in this taming. It only took 50 generations to result in the change.

It would not be that far off to think that humans gave food to cooperative animals and killed the aggressive ones.

Under labor conditions I would say at least 50 generations. But without these perfect conditions it is easy to imagine that it takes longer.

A bubble forms when a large number of investors pour money into investments that lack real inherent value. One example is Bitcoin, or in the year 2000, companies like Pets.com that had no substantial business but were valued like the most valuable companies in the world. This scenario does not apply to companies like Nvidia, Microsoft, and Amazon, which have huge, successful businesses. It is very likely that these companies will still exist in 10 years and remain highly profitable. However, their valuations could experience sharp ups and downs, similar to Tesla. Therefore, I would prepare for a volatile ride, but I would not sell an S&P 500 ETF out of fear of impending hype.

I believe that even Hacker News faces this issue. While some bots might be detected, delving deeper reveals that it's impossible to identify all of them. This challenge is exacerbated by the sophisticated techniques used by bot developers to mimic human behavior, making detection increasingly difficult.