_pdp_

Recently I decided to create my own VM manager for Mac (think of VMWare / Parallels). It now supports containers, Linux, Mac and Windows vm and local mac accounts. It has support for GPU acceleration too and it is agentic.

I would like to think that I could have done it without AI if I had the time and energy to do it.

I wouldn't have attempted to do this project without AI!

3 years ago the best model was GPT-3.5 Turbo with 16,385 tokens context window.

This is not a good metric IMHO. I would expect new projects to have bigger changes initially as they try to find a suitable architecture. That is not slop. This is no different than rewriting a project 2-3 times pre-AI to find out what sticks and how to organise it better.

All interesting projects that I have seen in recent times are all done with AI.

The latest frontier models no longer produce slop. The code is pretty good in general - better in fact in many situations compared what a human might have produced with limited time, resources and experience.

The argument against AI-assisted programming just doesn't hold anymore.

Someone asked me this morning if I consider Muse a privacy risk. Lol. I do.

But to be honest this is all temporary. Muse, Grok, Dots will have users for the foreseeable future but there will be viable options for self-hosting too.

I am already doing it https://github.com/pdparchitect/noodle/

The Hub powers the entire family inference needs through a single device - 4 users - 8 device about 50 agents - mixture of local models and some subscriptions.

Obviously you cannot get frontier level of intelligence for coding tasks on a single Mac Mini/Studio but most people don't need that. Latest generation of open weight models are fine and in the worst case scenario bring your own regional cloud based inference services that value privacy. There are already plenty of companies in UK and Europe that provide serverless access to GLM, DeepSeek etc.

I will get downvoted for this comment on this forum but I will call it out.

You wont be able to review or understand code for much longer.

There will be just too much of it. The vastness of code libraries in 5 years are hard to imagine by today's standards.

We need to develop tools that help us understand.

I am super excited about this.

While right now Mac Studio is not a budget device and it certainly costs more than the subsidised subscription, the promise that one day we can run these models on commodity hardware is exciting.

I could not test it because I am also facing 403 error - UK.

Anyway...

If you are interested, there is also an open-source way to do exactly this and it works with your existing subscriptions and local models too - your data is kept private on your devices.

You can make the game, stream it directly from your AI hub or play it locally on your devices. You can cast it on the TV while converting your phone into a controller (with haptics support) or you can connect physical game controllers for local co-op.

https://github.com/pdparchitect/noodle

The README does not talk about this feature but you can find videos in my twitter account.

I don't understand why you would want to remove well-balanced and relatively small local-inference models that ship with every mac.

Sure, not frontier levels but fine for basic tasks and they are off-the-cloud.

I did not make this but I am using it ... and so far happy user. I think Pi is supported.

https://github.com/hardbeat920/monocode

But yes, the TUIs are supposed to be lightweight yet it is overall a horrible experience.

I don't really think there is much future for TypeSafe but I wish them well. In fact, I like them.

Jev is just a reminder that you can use more "traditional" forms of AI (that are not LLMs) and still get remarkable results.

We tend to forget that.

That was a surprise and that is why it went viral.

It will be nice if it had persistent URLs and and SDK for desktop applications. It could solve a lot of small issues for a project I am currently busy with.

AI needs tools as much as humans do.

It will be solved when there is no more code left to write.

Code is an abstract concept that is not bound to the physical world and I imagine that future will have some much more of it that it is difficult to comprehend. Everything will be code and more code will be written than ever before.

Code will never going to be solved. The question is how much humans will be involved and I think the evidence is that perhaps just a bit. However, because we are talking about vast libraries of code even if we are involved in under 1% of all code and decision making that is needed around the code, there are still not enough developers out there to take on the task.

I might be wrong :)

So true. The real use-case outside of the 2-3 specialised domains are yet to come... if they come at all.

They wont. This is OpenAI specific. They don't even have support for OpenAI realtime models.

However, this open source project https://github.com/chatbotkit/platform/ does and you can plug OpenRouter or OpenAI keys straight in while keeping your integration work generic. The only downside is hosting it yourself but it is just docker compose up.

There are vendor neutral solutions too https://github.com/chatbotkit/platform

Is this even public?

How do you judge which one produces the best result? They all look the same to me more or less.

I think this might work.

When I started working no the trading agent I mentioned above I wanted to see if it can be just a better investor over the long run. The intention was not to do high-frequency trading. As you can see most of the days it is not taking any actions. The losses where down to mistakenly setting the stop losses too close to the top. If it wasn't so careful it might have made some money tbf.

My gut feeling is that AI agents will be able to manage a long-term portfolio much better than a human. Though it is just a gut feeling.

If this worked it wouldn't have been open source?

Anyway, I have been running my own trading experiment and so far it has lost a bit of money. That being said I have not tried to optimise anything - just let it do whatever it wants. The losses are small and it might be able to recover later this year. Who knows.

The agent writes a blog about its progress here https://trades.chatbotkit.space/

I am thinking to output all the chat logs to HF as well for research.

You can run your own trading agents that communicate over a message buss in your own terms by downloading the CBK platform and running it locally with your own models. I have also shared my trading blueprint if you want to give it a go. https://chatbotkit.com/hub/blueprints/trader

GitHub. Then you symlink them into the relevant folders. You can write a skill to manage them for you across all harnesses as well.

Great initiative. Also add

https://github.com/chatbotkit/platform

...if it fit the project direction.

It is basically a sovereign AI platform for agent orchestration and automation and app building.

I like the looks of Aisle and what they stand for...

That being said you cannot compare a model with a specialised harness. These are two completely different things.

Am I missing something?

I get that. However, given that sessions are local and personal why tag it into the public source tree when it can keep a catalogue within the local session metadata with the exact branch, tags and refs. Sounds like easier less intrusive thing to do.

Respect

Because Linux is not an OS.

I really hate this attribution nonsense ...

On one hand I guess it is fine to automatically tag that the commit was AI-generated. On the other hand, this will happen even if you audit every line and add your own changes too, which means the attribution is misleading.