jwblackwell

It's starting to feel like we'll soon be able to run open source models on our own hardware and use them for serious coding projects. Even if some tasks still need to be handed off to larger closed source models, that's a huge improvement over where we are today.

The trend also seems pretty clear. These models will keep getting better. Coding may already be close to a "solved" problem for LLMs. Yes ofc there will always be frontier stuff that you need gigantic cutting edge models for but let's be honest, most software is not that.

Nice! works well - I couldn't get huggingface to work either

Yea as somebody who played extensively with Lego as a kid, almost exclusively just the "base components", I never really got the appeal of those set with all sorts of custom, single-use items

I absolutely love the idea of Replit and I think it's an awesome platform and idea.

I do wonder how sustainable it is as a business though. I expect Replit is sending the majority of that money to the big AI labs through API costs

As soon as anything becomes serious you're going to try and take it off Replit and use something like Claude Code and AWS etc

With all due respect somebody could launch a version of Discord that's 10x faster tomorrow and nobody would know about it

It's very difficult to unseat those incumbents, especially those with strong network effects.

Plus the people that work in those larger companies are not at the edge of AI coding at all and not motivated to rock the boat

A lot of people in the UK seem to think the trains in Europe are so much better, but that's not the case, at least in my experience. We took an interrail trip a couple of years ago and spent half the time sat on the floor and waiting for delayed trains. If anything, the overall experience was much worse than trains in the UK, especially if you factored in the difficulty in procuring the correct tickets.

It works much better with AI though, which is really all that matters at this point

Yeah I am currently enjoying giving the LLM relatively small chunks of code to write and then asking it to write accompanying tests. While I focus on testing the product myself. I then don't even bother to read the code it's written most of the time

Yeah I do feel the pressure to run multiple instances of Claude Code now. Haven't really managed to find a good workflow, I find I just get too distracted swapping between tasks and then probably end up working slower than if I had just stayed in one IDE instance

This is just crying out for AI to help you get started.

One slightly unexpected side effect of using AI to do most of my coding now is that I find myself a lot less tired and can focus for longer periods. It's enabled me to get work done while faced with other distractions. Essentially, offload some mental capacity towards AI frees up capacity elsewhere.

> the current PHP+MySQL combo was not fit for purpose anymore

He lost me here. Sounds like he tried to change from a boring stack he understood, to Go and Clickhouse because it's cooler.

You're asking for trouble. LLMs are great and getting better, but you can't expect them to handle something like this right now

Your site would look way better with mx-auto max-w-7xl :)

The content is far too wide on big screens

Initially, I just posted my latest product, Quizgecko on HN and then manually DMd a few hundred people on Linkedin etc. That was enough for 100+ users.

I think if you have a relatively novel idea, that is quick to try out, then people will naturally share.

After that, I submitted to various AI tool directories. They drove a bit of traffic and helped Google find the product. Then some landing pages I made started to rank and it has snowballed from there.

“ In some countries, more than 90% of the population lives on less than $5 per day. If agentic AI code generation becomes the most effective way to write high-quality code, this will create a massive barrier to entry”

The author is essentially arguing that fewer people will be able to build software in the future.

That's the opposite of what's happened over the past year or two. Now many more non-technical people can (and are) building software.

Yeah this is one fundamental reason I don't see MCP taking off. The only real use cases there are will just be built in natively to the tools.

The Guardian used to be considered a serious paper, then a few years ago the quality declined dramatically. Now it's not much better than most tabloids.

The cost of robotics is coming down, check out Unitree. A couple of robot arms would cost about the same as a minimum wageworker for 1 year right now. But of course they can go virtually 24/7 so likely 1/3rd the cost

The upshot of this is that anyone will be able to order a couple of robot arms from China and then set them up in a garage, programming them with just text, like we do with LLMs now.

Time to think bigger.

Larger context windows are helping solve this, though.

I use ALpineJS which is not as well known as React etc, but I just added a bunch of examples and instructions to the new cursor project rules, and it's now close to perfect.

Gemini models have up to 2M context windows, meaning you can probably fit your whole codebase and a ton of examples in a single request.

Furthermore, the agenetic way Cursor is now behaving, automatically building up context before taking action, seems to be another way around this problem

Above & Beyond's label has a huge following. I have been to many events around the world. They do a big gathering each year e.g. https://www.youtube.com/watch?v=JEqySVgNkIA

It feels like AI has made this redundant.

I honestly cannot imagine hand typing out some JSON now, or most code for that matter.

I just write in natural language what I want and the AI will perfectly output valid JSON.

I am using Laravel Blade components which are effectively just fully server rendered templates in component style, so you get the benefit of that mental model.

Then you can just add Alpine as necessary.

Totally agree. React and Vue are overkill and bloated, and actually lead to a worse experience a lot of the time e.g. users starting at spinning loaders with broken browser functionality (like back buton)

A hybrid approach of server-side views with a simple lib like AlpineJS or HTMX for dynamic bits is far better in most cases. The initial page load is blazing fast because the server does the work. Then, for things like filtering or updating lists, you use the lightweight library for smooth, efficient updates. You get the best of both worlds—a super-fast initial experience and the ability to add dynamic features without the React baggage (and often-neglected backend).

Does anyone else struggle to wrap their head around a lot of this new cloud stuff?

I have 15+ years experience of building for the web, using Laravel / Postgres / Redis stack and I read posts like this and just think, "not for me".

This is a sham study funded by UpWork whose business is likely being hit by AI, since many low-skilled office tasks can now be automated.

The speed things are moving, it feels like we'll get a GPT-4 level "small" model really soon.

A 2% conversion rate is actually quite high. The mistake here is that they assume the reason people don't subscribe is because of the cost.

basically no open source fine tunes are censored, you can get an idea of popular models people are using here: https://openrouter.ai/models?o=top-weekly

teknium/openhermes-2.5-mistral-7b is a good one

You don't need all 14 files, just pick one that is recommended with a slight loss of quality - hover over the little (i)icon to find out