codebolt

> People aren't prompting LLMs to write good, maintainable code though.

Then they're not using the tools correctly. LLMs are capable of producing good clean code, but they need to be carefully instructed as to how.

I recently used Gemini to build my first Android app, and I have zero experience with Kotlin or most of the libraries (but I have done many years of enterprise Java in my career). When I started I first had a long discussion with the AI about how we should set up dependency injection, Material3 UI components, model-view architecture, Firebase, logging, etc and made a big Markdown file with a detailed architecture description. Then I let the agent mode implement the plan over several steps and with a lot of tweaking along the way. I've been quite happy with the result, the app works like a charm and the code is neatly structured and easy to jump into whenever I need to make changes. Finishing a project like this in a couple of dozen hours (especially being a complete newbie to the stack) simply would not have been possible 2-3 years ago.

MS Copilot for Android has some annoying UI bugs which they seemingly refuse to fix. The biggest is that the chat area sometimes gets randomly resized so that you can hardly read anything. There is also no way to search past chats. For all the billions they are spending on AI, their chat interface seems inexplicably half-assed.

Going to need a citation on this one.

I guess you didn't catch this:

> ChatGPT started coaching Sam on how to take drugs, recover from them and plan further binges. It gave him specific doses of illegal substances, and in one chat, it wrote, “Hell yes—let’s go full trippy mode,” before recommending Sam take twice as much cough syrup so he would have stronger hallucinations. The AI tool even recommended playlists to match his drug use.

Giving the LLM tools with an OData query interface has worked well for me. In C# it's pretty trivial to set up an MCP server with OData querying for an arbitrary data model. At work we have an Excel sheet with 40k rows which the LLM was able to quickly and reliably analyse using this method.

I've noticed Gemini tends to struggle with current news.

Another anecdote: I built my first Android app in less than a dozen hours over the holiday, tailored for a specific need I have. I do have many years of experience with Java, C# and JS (Angular), but have never coded anything for mobile. Gemini helped me figure out how to set up a Kotlin app with a reasonable architecture (Hilt for dependency injection, etc). It also helped me find Material3 components and set up the UI in a way that looks not too bad, especially considering my lack of design skills. The whole project was a real joy to do, and I have a couple of more ideas that I'm going to implement over the coming months.

As a father of three with a busy life, this would've simply been impossible a couple of years ago.

I've been a heavy user of the MS Copilot chat app on my Android phone, which I've been pretty happy with as a free basic AI chat option except for some annoying GUI bugs that they will apparently never get around to fixing. But I've yet to see a good use outside of the chat interface.

Atlassian cloud is also having issues. Closing in on the 3 hour mark.

Atlassian cloud is having problems as well.

For many, Lua is primarily known as the Roblox language. Pretty impressive that it's the language of use in a game (/set of games) with 380 million monthly active players - currently the most popular in the world.

In certain situations I'd argue it's a calculated risk worth taking. Let's say I'm tasked with adding a much-needed feature to an older legacy application that is built using frameworks and libraries I'm not familiar with, possibly also with with weak docs. If Claude is able to produce a self-contained piece of code that looks correct when you read through it and behaves correctly during (thorough) testing, then it sure has saved me (and my company) a lot of time and effort, even though we are deploying code we may not fully understand.

Just subscribing to OpenAI, Anthropic and Google on YouTube is pretty helpful. They post demos of all major new feature releases that are good to skim through to get a sense of where the frontier is moving (just take all claims about capabilities with a grain of salt).

I've also got some gems from Microsofts Build talks, specifically whenever Scott Hanselman and Mark Russinovich get together, e.g.: https://www.youtube.com/watch?v=KIFDVOXMNDc

You are simply deluding yourself if you still refuse to accept that LLMs are capable of some level of reasoning.

The crucial point is that we simply do not know (yet) if there is an inherent limitation in the reasoning capabilities of LLMs, and if so whether we are currently near to pushing up against them. It seems clear that American firms are still going to increase the amount of compute by a lot more (with projects like the Stargate factory), so time will tell if that is the only bottleneck to further progress. There might also still be methodological innovations that can push capabilities further.

The crucial point is that we simply do not know yet if there is an inherent limitation in the reasoning capabilities of LLMs, and if so whether we are currently near to pushing up against them. It seems clear that American firms are still going to increase the amount of compute by a lot more (with projects like the Stargate factory), so time will tell if that is the only bottleneck to further progress. There might also still be methodological innovations that can push capabilities further.

Whatever the true stats for mistakes or blunders are now, remember that this is the worst its ever going to be. And there is no clear ceiling in sight that would prevent it from quickly getting better and better, especially given the current levels of investment.

Oh yes, all those parameters are absolutely a pain to work with. But it can still be good to have an understanding of what options are abstracted away by fopen etc. Trying to write an app only using <Windows.h> can be a good learning exercise if you want to understand the fundamentals of the OS.

> You'll also find that the Win32 API has a lot of replacements for what's in the C standard library. If you really want to try and get the executable size down, see if you can write your app using only <Windows.h> and no cstdlib. Instead of memset() you've got ZeroMemory(), instead of memcpy() you've got CopyMemory().

I see he's also using fopen/fread/fclose rather than CreateFile/ReadFile/WriteFile/etc.

Maybe I've been chatting with them so much I'm starting to sound like one myself.

This model also seems to do a decent job with Angular. When I was using ChatGPT it was mostly stuck in pre-16 land, and struggled with signals etc, but this model seems to correctly suggest use of the latest features by default.

On a side note, I'm excited to see more an more ambitious side projects like these as LLMs empower hobbyists to do more in less time than was ever possible before.

If you can get them to approve GitHub Copilot Business then Gemini Pro 2.5 and many others are available there. They have guarantees that they don't share/store prompts or code and the parent company is Microsoft. If you can argue that they will save money (on saved developer time), what would be their argument against?

I've found they do a decent job searching for bugs now as well. Just yesterday I had a bug report on a component/page I wasn't familiar with in our Angular app. I simply described the issue as well as I could to Claude and asked politely for help figuring out the cause. It found the exact issue correctly on the first try and came up with a few different suggestions for how to fix it. The solutions weren't quite what I needed but it still saved me a bunch of time just figuring out the error.

And with this increase in productivity, your boss needs fewer of you to do the work that would previously require a larger team.

Tether collapse is still the big one.

Currently starting my first project integrating with Azure OpenAI using the new MS C# AI framework. I'm guessing that having experience actually building systems that integrate with LLMs could be a good career move over the next decade.

I remember trying to teach myself coding as a young teenager in the early 00's. I started with Java but it was too complex for a complete newbie working only with a text editor and command line, and I couldn't quite grasp the necessary concepts. It wasn't until someone recommended BASIC that I really got going. Really an ideal language to teach basic control flow concepts and other basics to a young newbie. Also great for playing around with graphics. Graduated from that to C and C++ after a few months.

.NET Framework is the most egregious of the last few years.

Any ideas on how the concepts here will mesh with the recently released Microsoft.Extensions.AI library released by MS for .NET, that is also supposed to make it easy to work with different models in a standardized way?