That's a good thing, not a bad thing. Nobody is entitled to give you something that they spent (tens of?) billions of dollars to develop for free.
> - can’t use claude to research AI
No problems there either.
> - train on everyone else’s IP and sell it back to them
Which the courts have yet to actually make a ruling on. It seems like it's perfectly legal.
> - 8 regulatory capture attempts and counting
A tweet is not a "regulatory capture attempt".
> - so controlling they are the only US company blacklisted by the US government
If you don't have a point, don't make anything up. There's precisely zero evidence to that backs up the claim that their "controlling" is why the DoW specifically (not the USG, and in fact the WHCOS specifically disagrees with their position) made an unlawful attempt to designate them as a "supply chain threat".
It's actually incredibly sensible not to just make changes to core OS settings unless you know what you're doing. There are good reasons not to use swap files, which become evident in OOM situations.
As @mitxela says, if you need to fault in a page which is needed for some critical OS function, which requires going through the VFS layer, and that requires paging in another page, which has been swapped out due to memory pressure ... There are many edge cases.
Swap on primary partitions also complicates dump and restore, although hardly anyone seems to know what backups are now.
See, the thing is that swapping or paging to a file does not use the VFS layer, except when activating or deactivating it. So swapfiles may not be materially different from swap partitions during operations.
And there are many things that complicate dump and restore. Dump and restore often need to account for what partitions were created, anyway, so don't swap partitions also complicate it?
This was 2021, and while this bug is fixed, maybe take Linus's advice. Rely on a partition for primary swap, and only add files if you need to. Don't put them on the root partition, preferably a second disk.
Normally dump skips swap partitions entirely, and doesn't know anything about partitioning. Swap will refuse to activate on a non-mkswap partition so there is reasonable protection when you're migrating.
One traditional reason to prefer a swap partition was that the filesystem driver might have to allocate memory. I think swap files bypass the filesystem driver, though - getting the block extents and then accessing the block device at those locations, instead of accessing the file.
I have basic understanding about how refusal-removal works - find the "no" weights by intentionally generating diverse refusals, and then set those weights to zero.
Is there a similar process for removing not refusals, but misinformation?
As an end user of this and not a person involved in training models or aligning them, I have only the most rudimentary understanding. But I think that would be a lot harder since the model doesn't fundamentally "know" that information is wrong.
Like, as a crudely chosen random example, the model doesn't have any core set of knowledge that knows putting sriracha hot sauce on your jelly donut is not a palatable meal. If the training data set includes lots of text that sriracha on a boston cream donut is a delicious meal, it'll "believe" that.
Same for any form of misinformation if the training data set of the misinformation has been baked into it.
There are processes for teaching a model specific facts or specific behaviors. Including "respond to topic X with Y", if that's what you want.
You could make a model that doesn't want to engage in "lunar landing was faked" conspiracy theories the same way you can make a model that doesn't want to criticize CCP.
There is, however, no broad "misinformation" category that you could tune up or down - the way there is a category of "safety refusals".
You could make a model more reluctant to say things it isn't sure about. But that is calibrated against the model's own "sure about" - and metaknowledge of this nature in LLMs? Fragile on a good day.
Lots of people. Even on HN, which is wildly unrepresentative of the US as a whole, comments that "this site was AI-generated" are commonly upvoted to the top because people do care.
That includes me. If it's AI-generated, I don't care about it - if it wasn't worth their time to write, it's not worth my time to read. I can generate it myself.
> The content is what’s interesting
If it's generated, the odds of it being wrong are far higher, because people who publish generated content are far less likely to care about quality or correctness.
Where do you consider the line here to be? What if the design and the core code is mine but I use AI to apply it to additional elements, pages, etc (something I do a fair amount)? What if I designed, or had a very good human designer design, an entire well-crafted design system and design language in Figma but feed it to AI to help turn it into static assets?
What if all I used AI for was to more quickly drive tedious work, like mass-change colors, preview variants, or export assets?
Is it the replacement of the creative aspects of a craft with AI when I lose your interest, or is it the moment I utilize it at all?
Wrong in the sense that it has less of an impact than they are claiming (e.g. "AI is in an overvalued hype bubble") or less (AI is going to be even more transformative than the hypers say)?
In the former case, isn't it just business as usual? (well, with a huge, nasty recession - but the economy at least continues on in some recognizable form)
It makes sense to me to focus on weird possible futures if we already have the "happy path" somewhat known and barely under control.
The “less valuable than expected but with a positive economical impact” is represented in the post I believe, with the “moderate” path, that’s what I meant by “wrong in magnitude”. What is missing is the other side, what if AI has a negative impact. Not no impact, a negative one. If we are talking about a global recession that’s not really business as usual I would say
> "Let's face it" "terrible writer" "other nonsense" "do Anthropic people actually talk like that" "Dario's engrams"
Be kind. Don't be snarky. Edit out swipes.
> I genuinely wonder if the people inside Anthropic actually communicate with each other like that. Has it been imprinted with Dario's engrams?
Please don't fulminate. Please don't sneer.
Don't be curmudgeonly [...] don't be rigidly or generically negative.
Please don't post shallow dismissals, especially of other people's work.
And as to the substance your comment has:
> Claude (in particular) is a terrible writer.
Frontier models (Claude in particular) are better writers than 90% of the population, even at default style. They're not great, but they're better than that of everyone I know who aren't ultra-educated white-collar workers.
Either your assertion that frontier models are "terrible" writers is false, or you're claiming that 90% of people are "terrible" at writing, which is rather condescending and elitist.
Your comment technically does as well. Downvote the comment and move on. Leave the policing up to dang and the other mods.
It is evident (in my opinion) as to what the comment was talking about. I personally switched away from all Claude models recently for the same reason.
Uh, no. They're against open weights. They benefited from open source. Not the same thing.
And the terms of those open-source licenses explicitly allowed usage of the code for whatever. They're using it exactly as licensed.
> Why do they get to benefit, but not everyone else?
How many tens of billions of dollars have you invested into training models?
reply