Michal Valášek

Ruby 3.4.0-preview1 is out now, and with it a breaking change: frozen_string_literal is now the default, meaning strings are immutable; therefore, code like this: something = "line 1\n" something << "line 2" ... will no longer work. However, this should still work: something = "line =1\n" something += "line 2" This is because the second creates a new string, whereas the first modifies the string in place. This is a small annoyance for the programmer, but this change creates big wins for performance, so on the balance is a good change. ... and the fact that I run a programming firm that does a ton of maintenance on Ruby apps and makes money off of every breaking change in Ruby has nothing to do with that opinion, I promise. 😄

1

MVPs shouldn't go to the bin, unless the business pivots to something else. Way too often I've seen agencies & devs building MVPs in a 'poor' way (and that's a nice way of putting it), because 'who cares, it will go to the bin anyway when you get PMF and fundraise'. I disagree. There's absolutely NO reason that your MVP should go to the bin, unless your business model or product change dramatically. And even then, a lot of boilerplate (like signup, subscriptions management, emails, etc) - can and should stay the same. If done right, that should be a given, with 0 extra effort on the dev side. If someone's quoting you to do a poor job, don't work with them. You can stay agile & lean (as a startup should) without compromising on quality, and forcing yourself to rewrite the next version of your app. When you're investing in something, especially at an early stage when money is tight, it's important to NOT buy garbage.

10

2 Comments

I just made a minor release https://lnkd.in/eGwVtnw9 (under version 0.9.2, and I am waiting for community feedback before making it 1.0). It has 3 major features: 1. Allows the creation of multi-type trees. 2. Provides relative access to the trees. 3. Trees take shape with the creation of the visitor. This means a structure can belong to different trees; you just need a different visitor. It's a crate designed to make it easy to transform all structures in Rust into a tree. Every property of a structure can be thought of as a branch, making it a multi-type tree (different from a self-recurring type). Its need arose from a linguistic requirement for my universal binding generator, and it allows me to access different parts (with different types) of an AST with a relative path. I hope you find it useful.

4

We had our best year last year, and are on track to have an even better year in 2024. Writing good software is hard. That's why we love it so much. It's hard because it's highly complex. AI tools ADD complexity to the development process, which actually raises the bar on the type of person capable of producing efficient quality software. This doesn't price out US developers, who already have a very high bar to entry in a global marketplace due to high cost of living. Instead, it drives out the junk coders (overseas and elsewhere) charging $50 or less per hour. Before AI, these sub-juniors could still be marginally valuable in very specific cases. Now, GPT and other tools blow them out of the water in the hands of people that know how to manage the additional complexity (and productivity) they provide.

7

2 Comments

We’re getting close to typing some C#! We’ve described our components in plain language and listed their dependencies in the Architecture documents. At the very end of the stream, I talked a little bit about Contracts in software development, and why they’re important. We’ll talk about that some more this afternoon. The stream is archived in the usual place, I’ll link it in the comments. See y’all at 1:00 pm eastern over on Twitch! #thatDeveloperDad #liveCoding #softwareArchitecture #cSharp #dotNet 

1

1 Comment

Had a chat with Eric Meyer and Brian Kardell about CSS masonry, does it belong in the grid spec or not? With a bit of a discussion about how and why we ask for developer feedback on CSS. https://lnkd.in/d3C-_dwq #css

33

React 19 was originally anticipated to be released back in May during 2024 React Conf but we're now heading in July and there is still no release date in sight. The initial feedback on React 19's new features was rather positive, new hooks, no longer having to worry about memoization, etc. As developers started using the release candidate more they encountered massive performance issues when using suspense boundaries. If sibling nodes under a suspense boundary previously triggered asynchronous data fetches, those operations would execute in parallel. Under React 19 that has changed, they are now executing sequentially. The React team's response can be summed up as "you're doing it wrong, this new way is better", which if you rewrite your code it could be but this is a rather large change that will negatively impact applications that utilize suspense and don't rewrite. If you want to go in depth into the drama behind React 19's release there's an excellent blog that you can check out https://lnkd.in/gvZHbQXd

5

2 Comments

Another update finished. Got the github thing working better now. Still trying to understand fully what each api call is actually pulling, pretty sure I don't have the commits label right but hey at least I am displaying data at this point. I even added some flashing text, so hop over to https://lnkd.in/ebsvVzaZ and find the blinking text to take a look at what I have for my git project... I will even supply the link to the repo if you want to try and follow along https://lnkd.in/ed9YkNCi

1

New blog post! Here's how to build a question-answering UI with citations using our function-calling LLM framework -- BAML. https://lnkd.in/gaUUvWXx The example streams LLM responses from NextJS in a structured manner. Check out the interactive examples at https://lnkd.in/gAU3p58A

15

I'm blown away. Now you can create fully functional video games and web apps in just 25 seconds—no coding needed! Anthropic dropped Claude 3.5 Sonnet just 2 days ago, and it has already dethroned ChatGPT 4o and Opus. And became the MOST POWERFUL LLM on the market. It's so impressive that we've switched to the Claude 3.5 Sonnet API for our content automation tool, GhostContent—10x better than ChatGPT 4o. Here's what people are doing with it: •Ammaar created an AI sound effects generator app  •Allie K. Miller coded a fully functional Mancala web app in 25 seconds •Purav used it to teach him backpropagation via animations Highlights of the new model: • Free access to public(paid versions available) • Outperforms previous models and competitors. • Artifacts feature enhances collaboration and editing. • 2x as fast as Claude 3 Opus. No wonder people are canceling ChatGPT plus subscriptions and going wild for this. PS: What's your take on the new update? Follow me for more insights on AI and startups. #claude #aiupdate #technews #artificialintelligence

17

3 Comments

This is quite interesting I've built many web crawlers, they are very useful for many tasks in vertical searches, (collecting wide domain corpora for building language models, building knowledge graphs, question answering etc ) Web automation is a very challenging task "In this work, we propose a crawler generation task for vertical information web pages. The goal of this task is to automatically generate a series of predefined rules or action sequences to automatically extract target information. This task calls for a paradigm that LLMs generate crawlers. Compared to traditional wrappers, this paradigm can be quickly adjusted according to different websites and task requirements. This flexibility enables crawlers to handle diverse and changing web environments more efficiently. Compared to the generative agent paradigm, it introduces intermediate rules to enhance reusability and reduce the dependency on LLMs when dealing with similar tasks, thereby improving efficiency when handling a large number of web tasks." https://lnkd.in/g9xkbHhH repository: https://lnkd.in/gzZ7www3

8

1 Comment

New issue of the Weekend Read just dropped. We will explore the operational semantics of async await, both from an application-level as well as a platform-level https://lnkd.in/gaYx28KF

14

"But will.. your no code tools doesn't give us the backend source code" This is a concern that a founder had when realizing that the "backend" source code is not available with the no code tool i suggested. I want to address this issue once and for all 💡 ⏰ Lets take a look at AWS here There are 2 types of code we’re talking here 1.⁠ ⁠Application Source Code: Owned by you, it encompasses your unique application logic and functionality 2. Infrastructure/Service Code: Owned by AWS, it includes the infrastructure that supports your application 🫣 If you are worried about intellectual property, you shouldn’t be using google cloud or aws in the first place. You should create your own server and setup your own configuration ☁️ AWS, other clouds & no code tools are both “managed services” meaning they make your development life easier and faster without having to code everything from zero. without AWS you would need to •⁠ ⁠purchase, set up, and configure physical servers or rent virtual private servers (VPS) from a hosting provider. •⁠ ⁠installing and configuring the operating system, web server software (e.g., Apache, Nginx), database servers (e.g., MySQL, PostgreSQL), and other necessary services. •⁠ ⁠⁠secure the server against threats. This includes setting up firewalls, SSL/TLS certificates, intrusion detection systems, and regular security updates. And many more. All these requires code and AWS handle all the above we can focus on building the “Application logic” 🧑‍💻 Since AWS has become like the “standard” of development people don’t realize that if they use AWS or google cloud or some other cloud, the only 𝗜𝗻𝘁𝗲𝗹𝗹𝗲𝗰𝘁𝘂𝗮𝗹 𝗣𝗿𝗼𝗽𝗲𝗿𝘁𝘆 that you have is only for the application logic. Before AWS was introduced in 2006, companies would have both the intellectual property for both application logic and server. Before we end this post, let me ask you this : 📽️ If i give you a task to compete with GRAB and i will give you the best traditional engineer to build GRAB app in 6 months, can you win against GRAB ? The answer is most certainly no, what's the use of focusing on having your source code on a losing battle ? Any decent programmers can build GRAB clone but what makes GRAB valuable is not their tech but their user network. Of course, if you are building a new innovation like openAI's algorithm.. you do need to make sure you own your algorithm hence you will need to own source code But other than that, you don't need to worry about owning your source code. The key is to find no code tool that can let you migrate easily when needed. Cheers 🍀 #Linkedin #Engineering #Sourcecode #nocode

1

Server-Side Rendering (SSR) might sound technical, but it’s a game-changer for your website. Imagine faster load times, improved SEO, and a vastly smaller carbon footprint. Instead of your browser doing all the work, SSR lets the server handle it, speeding everything up. Our amazing team at RY have developed this feature and built it into our Umbraco offering, so all our clients can enjoy these great benefits! I will be covering this and more at Umbraco codegarden next month... #Umbraco #Codegarden #RY Read below!

28

2 Comments

If you're having context problems I feel bad for you son,I got 99 problems but context windows ain't one (soon). Click show more for goodies -> This lil gem dropped 5 days ago: https://lnkd.in/gjfcyUbB TLDR: This more or less converts long text to graphs which facilitates jumping back and forth through text much like a human does. GraphReader with a 4k context window absolutely smashed GPT-4 with a 128k context window on four fairly rigorous benchmarks including 1-hop and N-hop with text lengths between 16,000 and 256,000 words. This mitigates the token recall curve (a lot like autonomous re-prompting) more so than bypasses the context window. Context windows are fine, they're 1M+ now, its' token recall curves that are the bottleneck. You know what else has a 4k context window? 3.5 Sonnet. Hint hint.

6

2 Comments

Ooo nice! Didn't know about this https://devhints.io/ Saw this in a little tooltip on a Tulip Interfaces app I'm developing. There's a bunch of javascript libraries covered and even unix Cron 👀 🤩 #programming #plcprogramming #automation #mes #tulip

2