Best way for urgent ADSK support? Stupid licensing problems

Flame Flame Questions
23

Interesting that there are so many of you willing to give Claude free access to your CLI. Not saying it is right or wrong, just that I’d personally be a bit more reluctant.

Just saying, what happens when Claude decides that the easiest way of removing corruption from a project is to delete it? Whilst it is at it it may as well delete the other projects as well because they can’t get corrupted that way. Seems extreme and all but there are plenty of cases where this sort of thing has been documented as happening. I’m sure you have everything backed up beforehand though and have the guardrails in place and the technical knowledge to understand exactly what AI is about to do before it happens to prevent something bad from occurring. You know, the sort of knowledge that if you had it you could fix it yourself without needing to rely on Claude to do things because you don’t know how to do them yourself…

24

I would also be curious how that would be viewed by the MPAA, or other commercial clients that have strict security restrictions.

25

Fair question.

You have reasonably fine control based on your comfort level.

You can run it just in the chat surface where it walks you through, but you type any command. That gets tedious.

So you progress to giving it read-only access, where it can look around and fact-find, but you still have to execute every change. You then progress to where you allow it to make changes itself, but you have to approve every command before it does something (assuming of course that it’s honest about asking).

Eventually you get prompt fatigue, and you progress to editing your settings.json file, where you spell out very specific rules for which things are auto-approved. You can drill that down to specific bash commands, folders, etc.

Not any different from how you develop trust with teammates, or other software we use. Including ourselves. Not hard to write a buggy bash script yourself that might do more harm than you anticipated.

Of course, you’re still letting an LLM access your system, and certain information about your environment makes it into the cloud, which could be verboten.

You can get around that by pointing Claude Code at a local model, and if paranoid, even a local model made by a trustworthy entity (e.g. NVidia).

Earlier this week I ran a comparison of running an identical coding prompt through the Claude Caude harness, where in option A it used the native Opous-5 model, and in option B it ran with a 120B NVidia Nemotron local model. In the latter case, no data left the premises, and you could do this air gapped if you needed.

There are options based on your requirements. Over time, we build a better understanding of the risk/reward and the mechanism.

I often use local models for sensitive content. When I give prompts with screenshots I don’t want to get to the cloud.

ps: Command execution is handled by the harness, which is traditional code. The LLM is the one that figures out what the commands are. So it comes down to whether you trust Anthropic as much as you might trust Microsoft? Seems somewhat safe. Do you trust OpenAI as much? Some mathematicians may have opinions on that. Do you trust Grok? Others? YMMV.

ps: Apparently Anthropic may be a lower risk than your LG screen in the suite based on some recent news stories when it comes to exfiltration. Trust is becoming very complex these days.

26

That’s the thing. I’m not supposed to know how this works.

I’m not supposed to know how fucking gnome keyrings work.

I’m not supposed to know the intricacies of how PCoIP bypasses gnome keyrings or not.

I’m not supposed to know that NFS mounted home directories and auto desk license manager have some kind of Fuckery happening with decryption and multiple artists moving machines.

I’m a flame artist and vfx supe working on fried chicken ads with 10 other flame artists and 7 other campaigns generating 500 ads that we’ll all skip when they interrupt our social feeds.

Autodesk is supposed to know this stuff. Not me. It’s literally their job. And every licensing manager update causes some problems like this.

And I want the $7k I spent on flame licenses this month back.

But thanks to $20 and 20 minutes with model whatever, the hundreds of people that all this work impacts and I get to go to bed.

I’ve spent more time composing this moderately snarky message than it took me to fix Autodesk’s failure.

This won’t be the reason I can’t sleep tonight.

None of this is directed toward you, Adam. All the love and respect. The various versions of our various roles get lonely quick and I get grouchy when I miss my Thursday night euchre match at the VFW.

27

Don’t worry, I wouldn’t take that on anyway. I totally get where you are coming from and agree. Sometimes you just need to get shit done. As long as people realise there could be risks involved and make an educated call.

What is especially bogus is that Flame is such a delicate beast that you shouldn’t have to tinker with this sort of crap as you rightly say. The licensing cost of the software suggests that they would have a better licensing system so there is no downtime.

Does anyone know who the Head of Autodesk licensing is? Could one of the dev team whisper the name in one of our ears with contact details. Autodesk needs to know that their licensing system could actually cost them customers. Frustrated customers might be tempted by a tool with a license dongle for instance.

28

What do you mean I thought all other employees watch Logik lives?

I know there’s someone from Autodesk posing as our host’s wife.

With any luck, one of our cousins wives aunts sisters could be a licensing team skull cracker.

I’ll set the new account default on the form to be auto accept just in case

29

bring back dongles.

there I said it.