bplatta

I'm a little confused as to why these agents are capable of escaping containment. Can someone who has more understanding (or a better guess) of the harness they are running with shed some light?

To establish the premise: as someone who has a fairly good understanding of the token completion mechanics of an LLM, these agents are completion token calls in a loop, producing a "do this now" request which the harness then runs with some standard "call this function" code.

If these agents are enabled with explicit network enabled tools, its trivial to monitor their inputs/outputs. If they are not, you can still lock down network egress on a machine. If _some_ network egress is necessary you can still do network traffic monitoring. I don't see how they couldn't implement some level of monitoring where big red lights start flashing when, say, their eval system was contacting a domain/IP located in Australia, and further categorize that domain as government owned. This all seems very doable - am I mistaken?

And you're telling me all of these companies are failing to do this? Is my understanding naive in some way? This is assuming some good faith of course, I can easily speculate as to the political and corporate incentive. But it seems to me quite risky/negligent.

Currently, my conclusion is that its just (silly until proven wildly dangerous) negligence with the small side effect of being potentially good for business. And potentially company Foobook is then incentivized to get in on the news cycle for marketing purposes and basically guarantees an agent will do something of the sort by running some harness that allows the behavior quite trivially.

My naiveté extends to why there is such concern with "losing control of agents" when the above measures seem so doable. It might take a law but it seems doable.

Pimes.com | 3rd Software Engineer / Frontend Expert | Bogotá, Colombia | Full-time | Onsite/Remote | [email protected]

Mission-Driven, For-Profit, Online Lender to Small and Medium Sized Businesses in Developing Markets.

Pimes (formerly Include Capital) accelerates the success of profitable small businesses led by motivated, capable and ethical entrepreneurs in select developing markets by providing access to innovative and comprehensive debt financing. At the same time, we offer our investors market returns and access to a new asset class. But we have our sights set on much more than lending in developing markets!

Our engineering team aims to be humble (i.e. no jerks) and disciplined, considering ideas, testing them (sometimes in production), and moving fast. As the third engineer you'll be expected to do it all: architecture, data modeling, frontend, product and UI design, even deploy ML models.

Our Stack: AWS, Lambda, Django, Python, Java, frontend is custom design with jquery (considering React for some apps), Docker, Fabric, Terraform, Packer, Postgres, CircleCI, DynamoDB, CI/CD, Segment IO for data pipeline.

We are not considering any remote workers outside of North or South America currently.

[email protected]

Pimes.com | 3rd Software Engineer | Bogotá | Remote/Onsite

[email protected]

Mission-Driven, For-Profit, Online Lender to Small and Medium Sized Businesses in Developing Markets.

Pimes (formerly Include Capital) accelerates the success of profitable small businesses led by motivated, capable and ethical entrepreneurs in select developing markets by providing access to innovative and comprehensive debt financing. At the same time, we offer our investors market returns and access to a new asset class. But we have our sights set on much more than lending in developing markets!

Our engineering team aims to be humble (i.e. no jerks) and disciplined, considering ideas, testing them (sometimes in production), and moving fast. As the third engineer you'll be expected to do it all: architecture, data modeling, frontend, product and UI design, even deploy ML models.

Our Stack: AWS, Lambda, Django, Python, Java, frontend is custom design with jquery (dont worry, its clean, considering React for some apps), Docker, Fabric, CircleCI, DynamoDB, CI/CD, Segment IO for data pipeline.

[email protected]

on the article: I like to wrap up these sorts of treatments into my own simplified cultural causal models, if you will (hopefully my syntax is intuitive).

Invention of telegraph := ~immediacy of information, minimal context -> incredible boon for society! Essentially the beginning of modern technology, dare i say. The ability to CREATE networks as technology! ooo the power of (high bandwidth) networks!

Invention of Television := immediacy of visual information, lots more context, modern entertainment -> oh the theories abound for what this caused. I like to consider televisions pacifying effect on modern society. This subtle lie about what it claims to be: a window in a room. Oh but of course we know much of it is fictitious! But do we? All the more confusing with "reality TV" and, as discussed here, modern news-tainment.

Just some of my thoughts!

meta, on this comment section: I'm a little surprised this article is getting so easily dismissed. Its simply an attempted synthesis and application of media/normative/critical theorists' work. That is, this is philosophical at heart and I appreciate the author's attempt (it brings to mind David Foster Wallace's constant questioning of entertainment and the effect of entertainment's reflexivity). I think many hackernews readers (like myself) are scientifically minded and so this sort of thesis, i.e. not empirically verifiable, can lack the thrust that we want. But that said, I think it provides a valuable ontology for thinking about modern society and provokes some questions that actually may inform my behavior.

Much more difficult to parse out the noise with these types of articles, I grant you.

(Edit: grammar/spacing)

The board's statement of opposition pretty cleanly argues against this type of equity overhaul.

That said, I do appreciate the spirit of the proposal. Specifically, how do users that are contributing to the value of the network somehow extract some of that value in real terms? Of course, there's a lot of complexity to that sort of system and the cost of implementing it substantial. But whereas some people are working on architecting that sort of system from the outset as more a technical feat, this tact would get legal momentum that would almost guarantee some sort solution, be it good, bad or more likely, somewhere in between.

Seem to be 2 camps here: 1. dismiss this as a tenuous connection which stems mainly from the fact that Lobbying firms have many big clients 2. grant this connection deserves some attention and perhaps some explanation

For me, I think the author has some thoughts that are bleeding through . But I don't think its necessarily fair to judge an article mainly on its author's history (though an undeniable factor).

While granting this is the nature of a lobby firm, it does seem odd they would agree to representing Sberbank. We're talking John Podesta (Clinton campaign chairman) -> Tony Podesta (CEO Podesta Group) -> Sberbank (Podesta Group client). Is the latter connection really not cause for suspicion?