[nisten]

The problem with AI ethics, safety and to a smaller extent, privacy groups is that the priority of the work/message is not placed on the practicality of solving the problem, or i.e. calculating improving affordability of food/housing, but is placed,as evident on this article too, on the lack of "governance structures".

In other words the priority of the work is to get these types of people into positions where they don't do any work.

At least with privacy groups you do get here and there some practical advice on using ublock origin or more rarely on how to install a blocklist from https://someonewhocares.org/hosts/, but with AI ethics & safety orgs... well lets put it this way.

I have yet to meet a single AI safety person that knows how to rename a file in linux.

God forbid we have a rogue AI-worm shutting down all servers & BGP routers while these types of people were in charge of safety, they'll be in the way of anyone even fixing it. They can't even get a simple safety benchmark working on lm_eval-harness. They're great at lecturing you why they shouldn't need that.

And this is the key issue with AI Ethics. It's the refusal to work at the problem constructively, and get the most skilled people possible to actually make the damn benchmarks to work, to rank models on the understanding of human rights, to list every current violation and abuse of humans in every single country without exception and to make practical plans on what to do when systems go rogue. Even if they're not technical they could be making the dataset in a csv in excel for that and making it public domain accessible.

Instead we get the most depressed, leechy-office-worker types complaining about how it's all over.

Now back to work, move it.

[dwohnitmok]

> I have yet to meet a single AI safety person that knows how to rename a file in linux.

You are meeting very few AI safety people then. A significant fraction of the AI safety people I've run into can build and train an LLM from scratch (without AI assistance FWIW), let alone have a grasp of basic command-line operations.

[datadrivenangel]

Where are you running into these unicorn AI safety people?

[dwohnitmok]

People who have gone through MATS (ML Alignment & Theory Scholars), SPAR (Supervised Program for Alignment Research), ARENA (Alignment Research Engineer Accelerator), or any of those other similar programs.

[dwohnitmok]

More generally a lot of the denizens on https://www.alignmentforum.org/

[number6]

CEN and CENELEC

[pessimizer]

> I have yet to meet a single AI safety person that knows how to rename a file in linux.

I don't know if instead of saying "safety" here you meant to say ethics, or if you're using "safety" in this sentence just to generally refer to "AI ethics, safety, and to a smaller extent privacy."

If either of those are true, that's weird because the only person in AI ethics most people know is Timnit Gebru, because she got fired and it made the papers. She has a BA and MA in electrical engineering, and her father was also an electrical engineer. After that, she went on to a PhD in computer vision with Fei-Fei Li (Imagenet) as her advisor.

https://en.wikipedia.org/wiki/Timnit_Gebru#Early_life_and_ed...

I guarantee you she knows how to rename a file in Linux.

If, instead, you were referring to "safety" specifically, I'd like to understand how you're making the distinction.

edit:

> Gebru joined Apple as an intern while at Stanford, working in their hardware division making circuitry for audio components, and was offered a full-time position the following year. Of her work as an audio engineer, her manager told Wired she was "fearless", and well-liked by her colleagues. During her tenure at Apple, Gebru became more interested in building software, namely computer vision that could detect human figures. She went on to develop signal processing algorithms for the first iPad. At the time, she said she did not consider the potential use for surveillance, saying "I just found it technically interesting."

[qcnguy]

Gebru didn't do any AI safety work at Google. She wrote stuff about doing social good, wrote the silly stochastic parrots paper that argued AI research was a waste of time, and whose title was proven false immediately (LLMs aren't parrots). The closest she got to "safety" was complaining that AI researchers weren't concerned enough that LLMs might say things Gebru didn't personally like.

Serious safety researchers are doing stuff like understanding neural circuits. Very different.

She surely knows how to use Linux. But she isn't really a safety researcher.

[ben_w]

> Serious safety researchers are doing stuff like understanding neural circuits. Very different.

There's a very broad range of unsolved safety issues in AI, and what she does is perfectly validly in that category even though she herself will deny the value of anything in the general category of "TESCREAL".

It's like how computer science safety can be anything and everything from "Should platforms like Meta be required to have a minimum age requirement?" and "Should section 230 provides immunity for online computer services with respect to third-party content generated by its users?", but also "How can we formalise system testing requirements to prove that things like Therac-25 never happens again?" and "Is the Boeing 737 MAX flight control software actually safe?"

[qcnguy]

I wouldn't describe the first two as computer science safety arguments. Those are purely social and merely the latest iteration of exactly the same debate that occurred around VHS tapes, newspapers, etc. The latter two are very specific to computing.

[ben_w]

230 is something that couldn't have existed with newspapers, because for "letters to the editor" it's obvious both who the editor is and that this is the opinion of some named public person whose opinion they were willing to reproduce. There may have been such an argument about telephones back in the day, but those are so old now…

Anyway, point is that AI does something new besides "is software" in the same way that computers do something new besides "is electrical" — social implications of ${new_tool} is absolutely relevant to all discussions about ${safety|new_tool}. Don't need to invent motor vehicle licenses or motor vehicles safety requirements or traffic lights before the invention of cars.

Edit: or to put it another way, "yes, and?" — they're still about safety of ${thing}, just a question of what level to improve that safety.

[overfeed]

> Gebru didn't do any AI safety work at Google

Wouldn't you find it strange for a co-lead of an Ethical AI team to not to do any AI safety work? I realize the AI Doomer vs AI Zoomer is a culture-war with a veneer or technical jargon, but I hope we can at least agree on the basic tenets of our shared reality even as we draw wildly different conclusions.

[qcnguy]

Not really. Ethics asks, "is this good?". Safety asks, "how do we make this safe?" - very different questions.

[matusp]

There are tons of highly technical papers being published about this topic. Yeah, there are some people without tech knowledge working on this as well, but I would expect them to be a minority, although they are probably very visible.

[tempodox]

Par for the course. This whole “AI” wave is one massive hype fest chock full of “creative marketing”, where you're hard pressed to find any reliable facts, how would “AI Ethics” even be a thing other than a massively hallucinatory artifact? If there were any ethics to be seen around “AI”, the first order of business would be to stop wasting those ridiculous amounts of energy for some cute parlor tricks.

[nradov]

If you're concerned about worms (and other malware) then "AI" is a total red herring. If BGP implementations have some kind of security vulnerability then eventually someone will find and exploit it, with or without AI.

[pjc50]

> to list every current violation and abuse of humans in every single country without exception

Couldn't you just ask the model that?

Joking of course, but this is in and of itself an intractable problem. Do you mean "restate the principles of human rights", which is a pretty small subset of law which is in turn a small subset of ethics, or do you mean actually get out there and enumerate and name every single person having their human rights violated? Not only is that an absurd amount of work it's politically impossible.

[nisten]

Sure yes that too, all of it. Why not. Labs process datasets with trillions of tokens.

The whole point was that these people just complain and make smartass excuses instead of getting any actual work done.

[pyuser583]

> I have yet to meet a single AI safety person that knows how to rename a file in linux.

It's funny, because I can tell when I'm dealing with a non-technical, policy person, because they always use language in a way no programmer would. I'd prefer not share the specific tells, but there are some things technical people just don't say, but policy people say all the time.

[GauntletWizard]

Amen.