Anyone else feel like Linus just totally did a "failed to read the room" like those commencement speech makers which were filmed recently being booed when trying to talk up "AI" ?

@kinnison
It's not like he hasn't been inspiring PEOPLE for the last 35 years to join the social movement of freedom, and zero purchase cost, software. Indeed he seems to be actively pushing them away.

@kinnison Yes, and it's not the first time. I really don't think he is the right person to lead the personal side of Linux, as opposed to the technical side.

@kinnison I think it might depend which room; it's certainly tactlessly put - but lots of kernel devs are finding AI useful when used carefully. (especially things like the Sahiko review system)

@penguin42 It's reasonable to say "This technology has uses" - it's not reasonable to say that claiming it's fine to use it is in any sense a-political though. It's not possible for most of us to avoid thinking of the vast ethical violations that the construction of these models has required, and to dismiss our concerns as political when Linux is, itself, political, otherwise he'd have chosen BSD/MIT not GPL.

@kinnison Well I prefer GPL over BSD because I saw the free X versions missing fixes that the proprietary ones had had for years; we're not all political animals! I separate the ethical concerns between AI itself and the current vendors; I agree many of the datacentre build outs are crazy and some of the training data as well; but I'd rather concentrate on getting locally running openly trained models working than decrying AI as a whole.

@penguin42 Are there actually any ethically sourced openly trained models though? So far I've not seen a sufficiency of evidence to convince me. Also the environmental cost of the training, whether open or otherwise is horrendous; as is the actions of the companies producing the hardware making such training possible.

This is why I think it's still not as easy as "well this isn't openai/grok/whatever" when it comes to ethical approaches to this technology.

@kinnison I think the openly trained models are getting there - I'm not sure how close to being ethically sourced they are yet; Training costs are high, true - but the algorithms and hardware are improving, so the energy use can go down. (e.g for what is now old, consider Deepseek surprising the west with what they managed on older hardware); and the MoE models are reducing the energy/compute costs at the inference side. No one tots up the worldwide cost of CI farms and says CI is political.

@penguin42 On the other hand, CI isn't being shoved down our throats harder than anything which could be construed as consensual. If in 5 years, once the most egregious violators of reasonable human decency have collapsed, and we've started to climb out of the global recession they induced, I might consider if someone could ethically apply the tech.

But for now, I don't believe that "this dumpster fire is contributing a bit less to global warming and it only smells half as bad" is sufficient.

@penguin42 and no, I'm not asserting that LLM training is a significant contributor to global warming, I just used 'dumpster fire -> warming' as an analogy.

@kinnison Yeh I agree it's being pushed stupidly hard - I completely agree about the craziness of how hard it's being pushed, some of the data centre builds, and 'token maxing' - but I just think there are uses when people are actually careful with it and I'd rather there are open tools to be able to do that rather than leaving them all in the hands of the big companies.

@penguin42 And that is a political position - my complaint was entirely around Linus' attempt to claim his stance was apolitical, or that Linux itself is anything but inherently political.

@wouter Again, just because he doesn't care doesn't mean that the Linux project as a whole is apolitical.