😂🎉

What did you use 😆

@@CrazyLuke11 I used openai’s gpt api, with the gpt 3.5 model. The timestamps with their titles might be inaccurate

Thanks men!!

Dope stuff!

This literally became an industry. Making chatbots and creating "posters" using apis

Anybody knows this Theo dude background? Reddit slams this guy all the time

Not flat earthing, just void timing I'm a void timer

Yeah I like 2023. Can January 1st 2024 just be December 32nd 2023?

that's wild

I love it when someone gives an opinion without any explanation. very nice! 😂

I disagree

That's a fair take, but it doesn't have to be all or nothing. There is probably room in all of these frameworks to have the endpoints send either something like HTML or JSON contextually(hell, rails was doing that eons ago, and if rails can do it, there's nothing stopping us in 2023 from doing one thing for our webapp, and one thing for more general access to an API)

I think the whole idea is to not try and think of "what if I want to use this end point for other languages/services" esp. when it comes to communication between a client (browser/renderer) and a server. For communication between services JSON still makes sense but between server and client/renderer it makes less sense.

Also why would server needs to be aware of what client is doing. Isn't being stateless something we aimed for. Now we have state in client, state in server in this backend for frontend and then other services. Why? I really haven't encountered a problem if designed correctly that json via rest doesn't solve

@@MrEsyphelon Yeah I think my major concern is that sending something like JavaScript becomes the norm... I don't want to have to implement a JS engine in all the languages I use, just because it's now popular to embed JS in HTTP responses...

@@MrEsyphelon What do you mean nothing is stopping you in 2023. Can you build anything in less than 2 days? Can you????!!!!!!

This whole pattern has been done multiple times -- and developers keep rejecting it: Flask + Jinja; Play Framework + Twirl; etc HATEOAS also. This was nasty to setup and most devs I worked with hated it. It's has value if you are implementing an Open API, but for lots of projects where the backend is consumed by the same team it's not so useful

well said

" then again we're increasing complexity." versus what? shuffling increasing complexity to the frontend and making everyone's devices work it out. users LOVE when a website/app loads MB's of UI frame and THEN starts to load the actual content they care about THEN start finally rendering it, optionally freezing UI the whole time. lulololol "The request is going to be made by Javascript and the language already has native support to JSON parsing," You know what javascript has REALLY good native support for? DOM manipulation it's literal reason for existing. JSON parsing support is poor justification for forcing server like behavior into the browser at every chance.

@@kitsunetantei it's like no one know what a middeware/middlelayer is.

​@@TheNewton Javascript does have support for DOM Manipulation, the Backend doesn't neither does Mobile Apps. Don't compare CSR with SSR, you can have a CSR with a HTML-based backend which will also take an imense time to load things, your argument is the same of Theo and it's almost like a beginner's argument. I disagree on the take that we should merge Backend and Frontend code into a Fullstack framework, although I prefer it for small projects, I feel more organized when it's separated with a shared package between the two. JSON is a language that has amazing support in Java, Kotlin, C, C++, Rust, Go, Flutter, React Native, etc. HTML does not have that support as it is not an object but a markup language. WIth that, let's compare things: HTMX: - Adding web code on the APIs. - Adding web code on the web client. - Adding JSON > UI code on mobile. - Adding a conditional to handle differently mobile and web. Traditional: - Adding web code on the web client. - Adding JSON > UI code on mobile. - Adding JSON > UI code on web. We do have SSR which includes client code in the server runtime, but that is different from adding client code to APIs. Again, are you guys sane? You're proposing adding tailwind to the API code. This JSON to UI schema happens in every traditional client, you guys want to make the browser the black sheep. With that you're increasing complexity as you're creating edge cases. - Make it all the same.

What was the general conclusion of the meeting? I think web development (the broader business where most developers at least took part at some point) will become increasingly easier thanks to AI copilots, and in turn, the remaining effort for developers will be spent in the new requirements, like managing custom LLMs for private data or services or simply implementing big LLMs like GPT4 apis into online services. It looks like programmers will have to evolve into some kind of "software architect". Programming will still be the most important requirement but less so, because it will be far from the only thing that you have to do now.

Nice!! Keep going man and you will break into the industry, good luck!!!

Hi I’m a college student, where do I find talks like this?

​@@vishnuvelayuthan7716maybe there's a google developers student club in your uni, they do talks in mine

@@tasmto thanks so much, I appreciate it!!

Tech recycles old styles as much as fashion! But the new stuff is always a little bit different, for better or for worse 🎉

Just because it's not considered trendy by tech youtubers, Angular is still the second most used framework in the market after React especially with enterprise stuff. If you actually want to get a job you either learn React, Angular or Vue not the newer stuff like Svelte or Solid etc. (sadly 🥲)

Angular is especially popular in banks or financial companies here in Europe, with either Spring or .Net (this is my anecdotal experience)

WASM's gonna be lowkey until either a killer app/workflow makes a splash, or it gets DOM access.

In the end WASM is going to be great for a whole lot of things except frontend development

Wasm as a replacement for containers Few pieces of the puzzle needed yet, but it’s getting there

I agree. I also will bet that more and more platforms will become a norm, where you jump in and can program business logic in an app and that is all you need to concern yourself with. Everything else that is needed is ready and waiting (like the many services AWS offers). Serverless simply can't do that.

Monitoring has actually improved a lot for servelress. Things like SST actually make k8s look complicated and serverless look simple. Serverless cost: I am basically running 3 full stack apps on aws with sst for sub 5 dollars a month. That is crazy. Storage cost more than running the actual apps and heating them 24h/24h on multiple regions. Serverless is the life for me in 2024 and has been for the last 6 months.

​@@kobibr9362 When you have no users... yeah, looks cheap. I've run the numbers on an app I host. Serverless is 100x the cost of running on a VM with the VM at 100% load. For 100x cost, you can over provision a lot before you're paying more. Problem is, switching costs can be significant. So you have to do the planning up front. Where do you want to be if your vision for the app comes to reality? How do the base costs play into your product cost structure?

Could you tell me why fetching data in useEffect is a bad idea?

It’s not, just having too many useEffects is bad

@@shubhamdhingra6089 what if I just use it to fetch data?

Tbh not using use query at this point is simply stupid.

@@thefilteredcoder could you tell something more?

I think you're misunderstanding what vertical slices are. Vertical slices are teams specifically built to cross the traditional borders present in a business with a siloed structure. Instead of having a design team, a frontend team, and a backend team, you have a feature team that has a mix of designers, frontend devs, and backend devs working specifically on that feature. That way the communication barriers are minimized because everyone is aware of the context (ie the feature) when talking about things. Smart companies apply a second layer of communication, where backend devs in different vertical teams have a channel where they can talk to each other and ask for input when they encounter a problem or are looking for best practices. Spotify is known for having this structure.

@@Ian-sm9uv Silos have trade offs. Any "grouping" you put on a company is effectively a silo, it's just a name placed on this grouping. It's required at particular sizes because you can't run infinite communication paths and get anything done. The primary ways to make silos in engineering orgs is functional vs product. A functional silo would be putting all backend devs on a team, all front end devs on a team. A silo based on the work they do. With product silos you have teams responsible for an end product, say an internal authentication service which includes backend and frontend both on the same team. The classic drawbacks are, product silos/teams create duplicate functions for many things. Your various product silos may re-invent the wheel separately. Silos built by function will have less duplication, but they tend to make worse end products. Connecting silos is a time honored tradition to try to get the best of both worlds, but I haven't ever seen it be massively successful. The gravity pull of the primary siloing tends to be too strong and it's hard to strike the communication balance well in a large organizations. Theo's take doesn't appear to contain serious experience in some of these aspects of siloing. Small teams building smaller more specialized tools are generally better/quicker/faster at getting to the end state. I tend to favor this myself, there's a clear value play that is appealing. But software tends to be a winner-takes-all market, this is a historical norm going back decades. The web has disrupted that some with the flattest market access in history, but it's still true in so many ways. I don't like his tool preferences. Building your stack atop a half dozen fly-by-night VC fueled vendors has enormous platform risk. However nice they might be to use. Staring into the economic uncertainty of poor national ledger, poor bank ledgers, rocky political environment, foreign wars, and VC capital backstopping an incredible amount of companies, including zombie companies. Platform risk should scare you. Be wary, be very wary.

Phase just gets better and better, and Theo works on his setup a lot! Plus, they are both good at taking good natured criticism. ❤

" it is more performant than JS " - ha ha ha

@@RaZziaN1 my statement is based on my own testing done against Vanilla JS. I tested CPU computations, HTML rendering, and 2d canvas rendering. I used Firefox, Chrome, Safari, Edge browsers on i7, i9, and Apple M1 CPUs. In most cases Blazor WASM was 2-5 times faster. In some cases they performed basically the same (0.1% difference). Without AOT compilation they were either the same or JS was a bit faster.

Ahah local LLM on a local box, just to think at one time it was said no one would ever need more thant 640kb.... Hhahaha :) Crazy thinking a friend that works on SQL is building a new machine with two 4090's to do pretty much to do this... Man how times have changed.

@@dand4485 Well, currently CPU inference is gaining traction, as well as optimization of models to have them working on more modest hardware. As example, Q_4 7b model might take 'only' 8GB of RAM, and speed of content generation is also not that bad - on mobile R5 4800H I'm getting 4-5 tokens/second from such model on CPU only :)

I agree with that. OpenAI models are too large and expensive for intensive usage and quality of open-sourced models is not comparable yet for many use cases. Looking forward to see how open-source world will catch the gap :)

Copilot doesn't seem super useful to me. Rarely has accurate suggestions for anything but the most extremely basic code.

It was using a GPT-3 fine tuned model up ‘till recently, which was not very good I agree. For example it made up a lot of non-existent methods conflicting with actual typescript suggestions. However I found GPT-4 capable of much better suggestions. Still on small scales but that’s what junior devs might struggle with. Or for example figuring out weird error messages.

kzfaq.info/get/bejne/Zp6Dqs6hsLyvqoE.htmlsi=kUbE-Jo8F92OBqFG