baconner · HackerTrans

What the other poster here said for testing against a reference, but also as an easier to get started with base for my own coding sandbox with coding agents. Took me quite a while to build one on my own that I was semi-happy with but I'd imagine one solid enough to run cowork on safely might have some deeper thinking and review behind it.

baconner

4 mesi fadiscuss

Ok I'd seen some sample sandbox scripts for this from anthropic before but not a full reference container. nice, thank you for sharing.

baconner

4 mesi fadiscuss

FWIW I think many of us would actually very much love to have an official (or semi official) Claude sandboxing container image base / vm base. I wonder if you all have considered making something like the cowork vm available for that?

baconner

4 mesi fadiscuss

I was trying to make no particular call on the actual reason aside from pointing at how obviously not the real story and false the statements made so far are. What a knot you have to tie yourself into to seek out an explanation where OpenAI has not made an ethical compromise to stay in the game here. I can stretch and think of some ways but they are far from the simplest explanation.

Lots of responses below give the likely real reasons most of which are probably true in part, but my opinion is it's the primary reason all who is in and who is out decisions are made by the trump administration - fealty. Skills, value brought, qualifications, etc. none of that matter above passing frequent loyalty tests, appealing to ego, bribes (sorry, i mean donations). Imagine thinking "hey, we'll work towards fully autonomous killbots because our adversaries will get them too but the tech isn't strong enough to allow them loose yet" or "yes you can use our ai for your panopticon surveillance, but just not on our own citizens because that is illegal" are lefty woke stances but here we are. Dario failed the loyalty test, as anyone rational would.

baconner

4 mesi fadiscuss

"We do not think Anthropic should be designated as a supply chain risk"

...but we're not willing to reject a contract to back that up, and so our words will not change anything for Anthropic, or help the collective AI model industry (even ourselves) hold a firm line on ethical use of models in the future.

The fact is if one of the top tier foundation models allows for these uses there's no protection against it for any of them - the only way this works if they hold a line together which unfortunately they're just not going to do. I don't just see OpenAI at fault here, Anthropic is clearly ok with other highly questionable use cases if these are their only red lines. We don't think the technology is ready for fully autonomous killbots, but will work on getting it there is not exactly the ethical stand folks are making their position today out to be.

I found this interview with Dario last night to be particularly revealing - it's good they are drawing a line and they're clearly navigating a very difficult and chaotic high pressure relationship (as is everyone dealing with this admin) but he's pretty open to autonomous weapons, and other "lawful" uses whatever they may be https://www.youtube.com/watch?v=MPTNHrq_4LU

baconner

4 mesi fadiscuss

Respectfully, it's very hard to see how anyone could look at what just happened and come to the conclusion that one company ends up classed a "supply chain risk" while another agrees the the same terms that led to that. Either the terms are looser, they're not going to be enforced, or there's another reason for the loud attempt to blacklist Anthropic. It's very difficult to see how you could take this at face value in any case. If it is loose terms or a wink agreement to not check in on enforcement you're never going to be told that. We can imagine other scenerios where the terms stated were not the real reason for the blacklisting, but it's a real struggle (at least for me) to find an explanation for this deal that doesn't paint OpenAI in a very ethically questionable light.

baconner

5 mesi fadiscuss

They can't catch everything but they can make your product you're building on top of it non viable when it gets popular enough to look for, like they did with opencode.

baconner

5 mesi fadiscuss

For sure, yes. They already added attempts to block opencode, etc.

baconner

7 mesi fadiscuss

NVIDIA makes money no matter if the model is open weights or not. I don't think open is a concern for them and they'd very much like to be servicing China and their batch of open models I think. what's concerning them more likely is

A. The inevitable breakdown of their massive head start with CUDA and data center hardware. A serious competitor at real scale.

B. Anything that'll cool off the massive data center buildouts that are fueling them.

Seems clear that locking up a major potential competitor especially the minds behind it solves for A. And their ongoing machinations with circular funding of companies funding data centers is all about B - keeping the momentum before it fizzles.

baconner

7 mesi fadiscuss

There are a couple of decent approaches to having a planning/reviewer model set (eg. claude, codex, gemini) and an execution model (eg. glm 4.6, flash models, etc) workflow that I've tried. All three of these will let you live in a single coding cli but swap in different models for different tasks easily.

- claude code router - basically allows you to swap in other models using the real claude code cli and set up some triggers for when to use which one (eg. plan mode use real claude, non plan or with keywords use glm)

- opencode - this is what im mostly using now. similar to ccr but i find it a lot more reliable against alt models. thinking tasks go to claude, gemini, codex and lesser execution tasks go to glm 4.6 (on ceberas).

- sub-agent mcp - Another cool way is to use an mcp (or a skill or custom /command) that runs another agent cli for certain tasks. The mcp approach is neat because then your thinker agent like claude can decide when to call the execution agents, when to call in another smart model for a review of it's own thinking, etc instead of it being explicit choice from you. So you end up with the mcp + an AGENTS.md that instructs it to aggressively use the sub-agent mcp when it's a basic execution task, review, ...

I also find that with this setup just being able to tap in an alt model when one is stuck, or get review from an alt model can help keep things unstuck and moving.

baconner

7 mesi fadiscuss

That's a nice idea, so nice in fact that it already existed as 18F until they closed it under the guise of efficiency earlier this year and are now starting over.

baconner

8 mesi fadiscuss

Its not hard to distugush individual pictures that contain trackable attributes like a license plate number from building a large scale database of them for sale. Or making such a database not legal to sell access to without removing that information, etc. It doesn't need to center on the contents of a single photo.

comments