Hi all-

I built Sandclaw (https://github.com/qix/sandclaw) to test an idea of having a human in the loop on any write path that the agent might do. The core agent has access to read anything, but other than to the LLM provider that information is securely firewalled off.

Over time I can loosen the firewall, and allow safer actions (i.e. adding a task to my todo list is always safe), and perhaps build smarter checks to allow more autonomy. But the `DEFAULT DENY` to sending/deleting emails, browsing untrusted websites, or running shell commands.

Is there any project that actually does this? I've been searching but it looks like all the options focus on letting the agent handle when to ask, which feels... problematic.

Here is the thing: if intelligence gets commoditized because LLMs are competing with each other, each trying to beat each other, and the prices are going down and down, it feels like there is a pricing war going on here. If this intelligence is now available so cheaply, like how Sam Altman said that it will be so cheap that it can be measured like electricity is metered, and then if the body, the harness an that makes an LLM agent like openclaw or hermes, becomes open source and it becomes better than anything that a startup can provide, then data becomes the final moat. At the end of the day every super intelligence or AGI will actually be good at only a subset of things. If that is true then the data that is affecting their training, if the pre-training thing continues and no zero-training LLMs are produced, then what happens is that the data that is affecting these LLMs will become the primary factor in determining which subset of tasks that LLM is good at. Therefore data becomes the final moat. any thoughts on this? (this is my first post on hackernews :))

Now that Wikipedia has banned AI generated articles. - https://en.wikipedia.org/wiki/Wikipedia:Writing_articles_wit...

I wanted to try and see if it was possible to get somewhat decent engine based on signs of AI writing from wikipedia themselves.

It runs entirely in the browser via WASM. Added instructions to further add more ways to figure out as we find out more about LLMs.

Would love feedback!!, especially: - cases where it completely fails - patterns you think are stronger indicators - whether this approach is even worth pursuing vs proper ML models, considering this is a classical classification task, I just don't like the idea of detection of a black box with another black box.

I have issue with making claude code work well to maknig trade.

I have connect to IB, but it always make mistake over and over again. I hope to perfect, but it always make mistake, and I need it working OK so I can recover loss.

Where I looking I see no code for this. Anyones have code for this?

This guy teleported all over the place testing it out. Uses quantum entanglement and natural duplication of physical person. He proved parallel worlds such as heaven and hell. The project discovered we never die just get duplicated in world vibration the individual matches. "Worlds within worlds all at different frequencies of vibrational matter" was the findings.Unites science and religion heavy duty concepts with his documented 20 year experiments.