Hacker Newsnew | past | comments | ask | show | jobs | submit | nticompass's commentslogin

I hate to admit it, but I'm pretty sure that's the same for me, too. Dial-up, NetZero I think. Waiting for images to download...

> There is no single control that solves indirect prompt injection

There is, actually. It's called removing the AI agent. Done.


This is the methodology I use.

No determinism, no separation of data and instructions, centrally controlled.

What couldn’t go wrong?


[flagged]


So it can write code to prevent the problem described?

[flagged]


What kind of "standard inbuilt anti injection code" are you referring to? Mysql_real_escape_string()?

Look up "prepared statements", it's pretty well documented.

How does this prevent prompt injection described in the article?

How does it prevent DDOSing and/or exposing the database from an injected prompt?


The user asks for details of the last transaction, the user gets back the amount, the source, and the description in a safely quoted format with the LLM never reading it.

You can't inject the LLM if it doesn't see the data.

An architecture like this won't work in many situations, but it can work for a lot of simple questions.

And if you want the LLM to summarize things, you run an isolated instance that makes a summary and you never show that summary to the LLM that's following the user's instructions.


You can do this, it is useful, but it's just not the same as where the goalposts are now which is: the AI is a person in a box and can do everything a person can.

If we actually limit them to "only accepts tiny ultra well defined problems and ultra well defined outputs" then theycease being a $10T/year idea and become a merely $10B/year idea.

Thus, it is not exactly popular at the moment.


> The user asks for details of the last transaction, the user gets back the amount, the source, and the description in a safely quoted format

What's "safely quoted format" when prompt injection is already safe in the description?

> You can't inject the LLM if it doesn't see the data.

How doesn't it see the data when you literally say "The user asks for details of the last transaction, the user gets back the amount, the source, and the description"?

> And if you want the LLM to summarize things, you run an isolated instance that makes a summary

And it will make a summary exactly how?


> How doesn't it see the data when you literally say "The user asks for details of the last transaction, the user gets back the amount, the source, and the description"?

The above post said how. The LLM writes code to do it. The code has a function to send text to the user. The LLM is not allowed to see the text, only the user is.

> And it will make a summary exactly how?

The second summarizing-only LLM is fed the raw data and allowed to output summary text. This is then sent directly to the user and put in a box with some hazard lines on it. The main LLM is not allowed to see the summary, only the user is.


Ah. Now I see you point. This might actually work for a number of situations.

I respect the amount of work that goes into projects like this; I can't wait to be able to play it.

I learned about this site from the TV show "Going Deep with David Rees" (episode 2)! I recommend that show if you haven't seen it.

I found an HP-12c at Savers for like $2 about 2 years ago! It's a "financial calculator" and I don't really know how to use it :-)

Maybe it's just me, but I hate the word "agentic" so much.


Sorry Spotify, but if it's AI-generated then by [my] definition it's "slop." We - or at least I - don't want AI music, we don't care about your AI. I'm just so tired of hearing about AI and new AI stuff.

Also, one of my friends uses Suno to make "music" and put it on YouTube, and I just don't have the heart to tell him it's "slop" and I've never/won't listen to it.


I have a small collection of programming/technology books, some from O'Reilly, others from Manning and No Starch Press. I've read them and they have been very helpful, but I will admit that I been buying less books. They take up space, and my shelves are full of these and other books like novels and manga.

Also, now when I am trying to learn a language, I just work on a project and search in the docs (or just on Kagi) to find what I am trying to do. Maybe a physical book would be a quicker reference, and maybe I should buy some more of them.

P.S. I enjoy finding old computer books (like DOS/Win95 era) at the thrift store.


This has nothing to do with the article posted or anything, I was just curious... who gets to pick the animal on the book cover? Do you (the author) get to pick, or does the publisher (O'Reilly) pick?


Do you mean the animal in general, or what specific image? For Go specifically, it would be very surprising if it was not a gopher! Ok, Python is obviously even more closely associated with the snake, but a gopher has been the Go mascot from the very beginning (the original design being drawn by the wife of one of the language's co-creators)...


I meant in general.

For whatever reason, I forgot that Go uses a gopher as its mascot. But like, their PHP books don't use an elephant.

I dunno, I was just curious if the author could pick an animal to be on the cover, or if it wasn't their choice.


For Go it had to be a Gopher (as it is the mascot for the language). My editor showed me the artwork and I approved it.


Are we still talking about ads?


Of course. It seems like an illustration of who's getting F-ed


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: