By way of background: I studied the philosophy of ethics at university, I co-authored a book chapter on AI/ML ethics, my wife and I quit our jobs and worked for free for years entirely focused on trying to help society adapt to and benefit from AI, I wrote the actual article we're commenting on, and the article is about LLM fine-tuning -- a field I to some extent created by developing the ULMFiT algorithm.
The person in question is, IIRC, at Governance AI, a group whose work I spent months studying in depth -- work which I believe is more likely to cause harm to society than to benefit it, as I explained here:
My point is that there was no reason to dismiss her good faith question when a simple no would have sufficed. Women are dismissed enough already. I don’t have any reason to believe he treated her this way because she was a woman, but I think in a field where women are constantly dismissed, and in a situation where there was no bad faith on her part, it was inappropriate for him to dismiss her. Women are in fact at a disadvantage in the field, and the best thing men can do is treat them with respect when they are operating in good faith. Men deserve that too! But it sticks out even more when it is a man dismissing a woman.
Easiest way to avoid this is to treat everyone with respect.
You are talking out of both sides of your mouth. In another comment on this same thread, you say this:
> [...]women in the field are more readily dismissed, and I think they shouldn’t be. It’s a moment to check our internalized biases and make sure we’re operating in good faith.
In your original comment you explicitly accuse the OP of operating in bad faith, presumably as a result of "internalized biases" as described above. How does this not add up to an assumption that OP treated the researcher differently because she's a woman? It is exactly what you are implying.
The scaling issue with potential runaway AI can be excluded. The potential for virus writing / security exploitation perhaps but such risks are already present with existing models so this point too can be excluded. I'm not sure there's any inherent risk here compared to what's easily available with a considerably reduced amount of resource requirements. The write up here seems concerned with allowing independent and democratised research which is a greater benefit than concentrated efforts.
Stepping back, this is the kind of discourse that Twitter can sometimes reinforce: short responses that, lacking context, can be interpreted in polarizing ways. After a bit of reading on the participants (because both of them are working in interesting areas), my belief is that the "pencil" response is actually shorthand for a whole set of beliefs that provide a lot of context and nuance to the discussion, but if you don't know that, it sounds like saying "AI is as dangerous as a new brand of chewing gum".
In addition, without defining what risks we're talking about, it's really hard to know the scope the answer is addressing. E.g., societal risk due to AI in general? vs. say, the risks of affecting the quality an existing model with fine-tuning?
So, I am chalking this up to a misunderstanding due to the limitations of the platform as well as the short-hand used in the response. And I could be completely wrong :)
> especially towards a professor studying societal impacts of AI
I can't help but feel like you're trying to load this question with some expectation that sex/gender should change how people react and respond. It shouldn't, at all, positively or negatively. Everyone is human (for now).
I think the poster to which you're responding is saying there wasn't any visible evidence of bias in the original behaviour.
From one point of view, you're correcting a bias (in this case, one you suspect might exist), and you believe that isn't a bias.
From another point of view, you're introducing a new bias (that definitely exists) to something that was previously (presumably) agnostic and neutral, by saying that a correction should only be applied when the question-asker meets certain criteria.
Both points of view are legitimate.
PERSONALLY I'd rather we not bring the "have you thought about how this might be discriminatory" bits into the converation unless there's at least some vague reason to think that it was, rather than presuming it's always discrimination when the genders and/or sexes and/or races line up a certain way. But that's because I think it's more important to debate the ideas in their pure form than to correct every instance of discrimination, and that's an arbitrary value judgement I made.
Best way to avoid this is to treat everyone with respect. This is a useful attitude for everyone (see Buddhism, “lovingkindness”) but it’s extra important for people in high profile positions.
To be clear I’m not saying women should be given unnecessary deference. That would harm development. I’m saying that if they’re behaving in good faith, they deserve a good faith response.
One thing I've wondered about: what are the drawbacks to using QLoRA? For example if compute is not a limit, I'm guessing one should not use QLoRA and finetune in full precision instead?
Afaik when a model is first quantized to nf4 (before finetuning begins), model performance is degraded from baseline (see https://x.com/Tim_Dettmers/status/1661482614811918338?s=20).
Dettmers shows that after finetuning wrt the dataset, the result is as good as full precision. But afaik never explored the effects outside the finetuning data. Assuming the finetuning dataset is small, the model will just be the degraded nf4 version, right? Or perhaps finetuning will even skew the model in weird ways (trying to fix quantization errors).
Anecdotally models finetuned wth QLoRA perform well. Does anyone have any papers or a careful analysis of this?
I think that is a great example of how important and useful QLoRA is. Maybe we should run a dual-GPU challenge next time not that multi-GPU is working...