It boggles my mind that Microsoft can fail to deliver even bare minimum working software so consistently, with all the engineering talent they house. What are the product teams doing?

I feel it is even worse than bare minimum -- it is so bad it infuriates the user. It is hard to get away from that reaction when one uses better alternatives in their personal life.

Am I missing something? This isn't a vulnerability. Your agent can see the files in its virtual environment. SSH keys are also not necessarily confidential. Please don't use AI to write blog posts.

author here, Yeah the System Files are in the ios app, but what it sent me was the entire runtime as well. from the root.

But should you see that if you just use it as as service?

you won't unless you deliberately try to read all that stuff

But should the AI simply obey?

Unless there is a reason not to. Giving the user the contents of their own sandbox is fine if you're supposed to let them run anything in it to complete the requested tasks.

If the only purpose of the sandbox were to isolate the instance of the service the user used then it might be a different expectation.

As a user, yes please.

It's most likely an axolotl.

I tried the phone model:

"My car crashed I need help"

{ "'crashed' implies need for music. 'play_music' with query 'car crashed' from user's words.", ..., "confidence": 1, }

Hey, thanks for the feedback! I think this is a useful part of a demonstration so I added a 911 tool specifically to demonstrate this capability and the fact that you can guard it with triggers that make it so calling emergency is an unambiguous action given the input. This really shows that constructing the right tool set with the right surrounding setup is a priority when deploying needle.

> 'call nine one one' -> call_contact with name 'nine one one', no phone number given so use placeholder '9101' as placeholder.

> 'call 9 1 1' -> call_contact with name '9 1 1', no required params.

> 'call ambulance' -> call_contact with name 'ambulance'.

> 'call 911' -> call_contact with name '911' and no required params.

Only the last one actually used the emergency_sos function (even though the reasoning says it used call_contact). If I were to use needle in my application, how would I improve accuracy?

Thanks to the scheduler, you can do both!

Is there a bottleneck which would hinder putting this architecture in charge of a humanoid? Would it be able to operate continuously, for example in conjunction with an LLM for long-term reasoning? Doom seemingly works extremely well.

defining the workflow such that the operation is a set of relevant questions

I learned to program from block programming, starting out when I was 12. I'm now a full-time developer and funded startup founder. You absolutely can learn complex concepts from simple primitives, and there is a large underground community pushing the limits of languages like Scratch. See for example Linux running on RISC-V running on Scratch:

https://turbowarp.org/1201938491

In my experience, there’s no better way to learn more complex programming techniques or tools than running into a brick wall with the simple tools you’re familiar with, spending a ton of time on an unreasonably irritating solution, only to realize that you’re running into a classic [n] problem and some clever person figured out a simple logical trick to address it. I mean, it can be enraging but that knowledge tends to stick.

Though I doubt that happens much anymore.

Couldn't agree more. To add on I think you can learn so much by doing something incredibly stupid you're really not supposed to in an intentionally constrained environment (in a good way, like the grandparent comment). The sort of absurd things that get posted here occasionally and get a few commenters genuinely annoyed that they're doing that with their time instead of curing cancer. Plus it tends to be easier than it looks.

(I think some people have caught on to this and found out that Claude is pretty good at it.. not sure what the point of that is but whatever)

Please don't submit AI-generated articles.

LPSO Astra.

I'm looking forward to doing code review on the go on a phone like this! Once a GrapheneOS-compatible brand copies it, that is.