In that case I think you can have a refund subagent that is responsible for checking if the user really asked for refund before doing these dangerous things. But it only minimize errors, LLMs are non-determinitic by nature.
Interesting, I have built https://github.com/michaellee8/voice-agent-devkit-mcp exactly for this, launch a chromium instance with virtual devices powered by Pulsewire and then hook it up with tts and stt so that playwright can finally have mouth and ears. Any chance we can talk?
Just sent an connection invitation on Linkedin. This is actually designed for allow e2e automation using playwright-mcp for a previous startup i worked in that does voice-based job interview agents. The http endpoints is provided by a daemom sitting on the background, listening all input to the virtual mic and transcribing and storing it. The agent can hit /speak and /transcript through an mcp. We have built Livekit Agents specific solutions by injecting text responses but felt that is not enough since we want to be able to test the whole thing end to end so I hacked a way to do virtual mic/speaker. It was designed for closing the dev-test-debug loop so that Claude Code can develop on its own rather than relying on human to test it.
I only run software from Chinese companies inside a sandbox, either on my Android/iOS phone or inside a VM for desktop apps and only enable necessary permissions. Unfortunately Mainland tech giants have no sense of user privacy and would like to maximize their profit by collecting every single bit of your data because they don't profit on selling you the software, they profit on selling your data.
I’m not in a position, nor do I have the skills, to fully validate exactly what I’m agreeing to. Let us assume that what I’m sharing is merely my app usage data: what I listen to, my likes, follows, comments, usage patterns, etc.
They share this data with 954 “partners” - what exactly does this mean? What other data do those organisations have? Who do they share it with?
I don’t think the average user has any chance of fully understanding what they’re agreeing to.
There is a difference when you simply lazy, or don’t care enough to understand the information in front of you, or when they don’t provide those information. You’re right, most people don’t care enough, but this is a huge difference. And west is magnitudes better with this.
Also I’m living in the EU. If I want I can get all of the information which you asked for.
But on the other hand, companies purposefully make those information as obscure as possible. Also, I’m not sure that people would care even if it had been clear. People love free stuffs.
I'm not sure why "954 partners" is surprising: log10(954) is between 2 and 3 so, if you assume Soundcloud uses at least 10 SaaS products to manage data (AWS, Snowflake, Datadog, etc. this number is definitely a low estimate). And then you assume each of those entities process the data through 10 partners of various kinds, it only takes 3 steps out to get 1,000.
You really have to put everything in a box nowadays. Companies are indiscriminate. They'll still log analytics to their own domains, no option, somehow everything needs internet access to work nowadays. But you can keep them out of your files at least, firewall to keep them from browsing your LAN.
>You really have to put everything in a box nowadays.
What if that was always a good idea.
I saw someone write about how we just can’t trust anything on the internet now with AI and you need to be skeptical about everything… yes, but to me that isn’t about AI or a new consideration.
I quite like Shelter [1]. Shelter apps are installed in a separate work profile, which essentially sandboxes it from the rest of your data. It also has a neat feature to automatically disable (freeze) specific apps and seamlessly re-enable them when you launch them through Shelter.
This is what I do too. If i need to use or test something i don't trust then I use an old phone. All of the phones use crDroid(1) and I have scripts to quickly wipe and reinstall the OS whenever I need a full nuke.
The context is somebody asking "Mainland US or Mainland China?" The comment you're responding to brought up Taiwan because that's the natural "not-mainland" when you're talking about China.
Almost. Both China and USA have threatened military action in Taiwan and Greenland respectively, but legally the USA and Greenland are not one; Greenland is a territory of Denmark despite having an independent government. Taiwan and Mainland China also have independent governments, but legally both consider themselves China, so it would be like North and South Korea if they had never agreed that they are separate countries now. Recently Taiwan has begun changing their identity as an independent country, and began the legal updates, however this is not internationally recognized because mainland china has resisted it, and frankly few countries want to go against china and risk sanctions or other political action from china. Even the USA doesn't recognize taiwan as separate, officially, although actions speak louder than words, and it is clear that most respect Taiwan's desire for independence and treat them as sovereign.
Sort of, except not really, except yes really. It's complicated.
The China that was a founding member of the United Nations was the Republic of China (ROC), and it controlled both mainland China and what we call Taiwan. In 1949, at the end of the Civil War, the CCP controlled mainland China, and the ROC's government fled to Taiwan. Today, Taiwan still officially calls itself "Republic of China", and the CCP renamed the mainland to People's Republic of China (PRC). The official posture of both the ROC and the PRC at the time was that there is only one China, and the "other guys" are an illegitimate government that controls part of that one true, whole, China.
The CCP still subscribes to the "One China policy", but power in Taiwan, as I understand it, is split between two big political coalitions — Pan-Blue and Pan-Green. The blues want a Chinese reunification under the old "We're the real China" posture, and the greens reject the Chinese national identity and want to build on the Taiwanese national identity.
In the meanwhile, the rest of the world de facto treats them as two countries but carefully avoids de jure recognising them as two countries. Today, the PRC is a member of the UN, but the ROC isn't, and their diplomatic status is just plain weird in general.
There are two governments that contain the substring of "China" and their constitutions claim a single unified Chinese country that includes mainland and Taiwan island, most of the world, seems ok with that.
Sounds like 5D chess, since Taiwan applied to be the "sole legal government of China" in the UN back in the 50s. (which was rejected) then they rejected the 70s resolution of "two Chinas". So it comes through as ambitious. But I will let the Taiwanese correct me on that.
Yes, the situation was different in the 50s and 70s. But for the last few decades it has been explicit chinese policy that any change of the status quo would lead to an invasion.
Somewhat similar to HongKong where China apologists always bring up that HK never had any democratic autonomy while conveniently not mentioning that China explicitly stated that such would instantly result in an invasion.
Putting a gun to someones head forcing him to say something and then using that against him.
> Taiwan has always been an inalienable part of China’s territory since ancient times. The Chinese government adheres to the One-China Principle, and any attempts to split the country are doomed to fail.
> Unfortunately Mainland tech giants have no sense of user privacy and would like to maximize their profit by collecting every single bit of your data because they don't profit on selling you the software, they profit on selling your data
In that case we should have some sort of UI test backends I guess? This mcp was more for generic use cases which will allow any TUI framework in any language to work.
i think cs students should force themselves to learn the real thing and write the code themselves, at least for their assignments. i have seen that a lot of recent cs grads that has gpt in most of their cs life basically cannot write proper code, with or without ai.
They can't. Universities will eventually catch up to the demand of companies, just like how the one I attended switched from C/C++ to only managed languages.
With that the students were more directly a match for the in-demand roles, but reality is that other roles will see a reduction of supply.
The question here is: Will there be a need in the future for people who can actually code?
I think so. I also believe the field is evolving and that the pendulum always swings to extremes. Right now we are just beginning to see the consequences of the impact of AI on stability & maintainability of software. And we have not seen the impact of when it catastrophically goes wrong.
If you, together with your AI buddy, cannot solve the problem on this giant AI codebase, pulling in a colleague probably isn't going to help anymore.
The amount of code that is now being generated with AI (and accepted because it looks good enough) is causing long-term stability to suffer. What we are seeing is that AI is very eager to make the fixes without any regard towards past behavior or future behavior.
Of course, this is partially prevented by having better prompts, and human reviews. But this is not the future companies want us to go. They want us to prompt and move on.
AI will very eagerly create 10,000 pipes from a lake to 10,000 houses in need of water. And branch off of them. And again.
Until one day you realize the pipes have lead in them and you need to replace them.
Today this is already hard. With AI it's even harder because there is no unified implementation somewhere. It's all copy pasted for the sake of speed and shipping.
I have yet to see a Software Engineer who stands behind every line of code produced to be faster on net-new development using AI. In fact, most of the time they're slower because the AI doesn't know. And even when they use AI the outcome is worse because there is less learning. The kind of learning that eventually pushes the boundaries in 'how can we make things better'.
reply