DeepSeek’s specific trained model is immaterial—they could take it down tomorrow and never provide access again, and the damage to OpenAI’s business would already be done.
DeepSeek’s model is just a proof-of-concept—the point is that any organization with a few million dollars and some (hopefully less-problematical) training data can now make their own model competitive with OpenAI’s.
Video of censored answers show R1 beginning to give a valid answer, then deleting the answer and saying the question is outside its scope. That suggests the censorship isn’t in the training data but in some post-processing filter.
But even if the censorship were at the training level, the whole buzz about R1 is how cheap it is to train. Making the off-the-self version so obviously constrained is practically begging other organizations to train their own.
Making the censorship blatantly obvious while simultaneously releasing the model as open source feels a bit like malicious compliance.
I’m not sure I understand the distinction they’re trying to make between the connectome and the projectome.
The typical web crawler doesn’t appear to have a lot of logic. It downloads a URL, and if it sees links to other URLs, it downloads those too.
So it has nothing to do with “AI training” in the usual sense.
For over a decade, complexity scientist Peter Turchin and his collaborators have worked to compile an unparalleled database of human history – the Seshat Global History Databank. Recently, Turchin and computer scientist Maria del Rio-Chanona turned their attention to artificial intelligence (AI) chatbots, questioning whether these advanced models could aid historians and archaeologists in interpreting the past.
Peter Turchin and his collaborators don’t have a great record of understanding human history themselves—their basic shtick has been to try to validate an Enlightenment-era, linear view of human history with statistics from their less-than-rigorous database, with less-than-impressive results. I wouldn’t necessarily expect an AI to outperform them, but I wouldn’t trust their evaluation of it, either.
One under-appreciated aspect of Docker is that it forces you to document all your setup steps in your dockerfile and docker-config files.
“We’ve got now three humans with Neuralinks implanted and they’re all working well,” Musk said
I thought the first one had more or less stopped working.
I haven’t tried it because I’ve read a lot of negative discussions of it—and because (by my understanding) the only reasonable use case would be if there were a large number of users and each user is likely to have copies of the same files but don’t want to expose their files to each other (so you can’t just manually de-dupe).
Between the various cortical layers and white matter, what part of the brain’s structure do these implants typically target? Do they sit on top of the outermost layer of some specific region of the cortex (effectively creating a new layer), or do they make long-distance connections to other brain structures?
Seems like they could have avoided this by having the Sandy Hook families join the bid with an arbitrarily high dollar amount—which they’d immediately get back as creditors.
Obligatory Saturday Morning Breakfast Cereal.
My understanding of quantum algorithms is that they set up parallel computations in such a way that incorrect solutions cancel out and correct ones reinforce each other. They indicate the existence of multiple universes to the same extent that the double slit experiment does.
One thing to bear in mind is that, whenever someone accustomed to one platform explores another, they’ll tend to ascribe any differences between the communities to the other platform being an echo chamber of some kind.
Presumably there was something in the training data that caused this pattern. The difficulty is in predicting beforehand what the exact effect of changes to the training data will be.
Is it a simple error that OpenAI has yet to address
Since even their own creators can’t understand or control LLMs at that level of granularity, that seems to go without saying. Although calling it an “error” implies that there’s some criterion for defining correct behavior, which no one has yet agreed on.
or has someone named David Mayer taken steps to remove his digital footprint
Sure—someone gained a level of control over ChatGPT beyond that of its own developers, and used that power to prevent it from inventing gossip about himself. (ChatGPT output isn’t even a digital footprint, it’s a digital hallucination.)
A quick Google search of the name leads to results about British adventurer and environmentalist David Mayer de Rothschild…
Oh, FFS.
It’s to be expected that an industry would want to study the safety of its own products—although it could point to the need for more independent research. What is a cause for concern is this:
Another study published in Scientific American found that meta-analyses by industry employees were 22 times less likely to have negative statements about a drug than those run by unaffiliated researchers.
That does suggest that the meta-analyses (as opposed to primary studies) are being used more for marketing than for product improvement.
Enshittification is defined as the gradual deterioration of a service or product brought about by a reduction in the quality of service provided, especially of an online platform, and as a consequence of profit-seeking.
I think that’s overly broad in comparison to Doctorow’s original meaning (which they also cite in the article). The critical element missing from their definition is that the enshittified product/service never had a viable business model to begin with: it uses the hype cycle to sell users and investors on an unsustainable mirage before inevitably collapsing.
It’s so secret, it’s already scrubbed itself from the internet.
Right—by “take it down” I just meant take down online access to their own running instance of it.