site banner
PaperclipPerfector  11mo ago (text post)   4829 thread views

Small-Scale Question Sunday for December 10, 2023

Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?

This is your opportunity to ask questions. No question too simple or too silly.

Culture war topics are accepted, and proposals for a better intro post are appreciated.

3
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

The NSA has known about using language models to generate text embeddings (or some similarly powerful form of search based on semantic meaning rather than text patterns) for at least 15 years. This is why they needed absolutely massive amounts of compute, and not just data storage, for their Saratoga Springs data center way back when.

0.1% that they were, like, 5 years ahead of the public state of the art IMO. So much of deep learning progress has been based on 'more compute', and moore's law in terms of FLOPS has been advancing for so long, that it just doesn't work. However the idea of neural networks for semantic classification or machine translation or similar has been known for a very long time, so I could totally see them trying to use the (quite meh) state of the art at the time with a lot of compute.

Unelected leaders of some US agencies sometimes lie under oath to Congess.

Probably over 10%? A lot of people, including people with power, say things that are various degrees of lies.

The Omicron variant of covid was intentionally developed (by serial passaging through lab mice) as a much more contagious, much less deadly variant that could quickly provide cross immunity against the more deadly variants.

<1%? My vague memory is that there were a lot of variants, and that in general 'virus mutates to spread more and be less harmful' is fairly common, so imo there's not that much reason to believe this.

0.1% that they were, like, 5 years ahead of the public state of the art IMO. So much of deep learning progress has been based on 'more compute', and moore's law in terms of FLOPS has been advancing for so long, that it just doesn't work.

It's an offshoot of the widely-reposted AI Twitter claim that 'we could have trained GPT-2 in 2004' (or with 2003 levels of supercomputer compute). And that might well be true, idk. Here's one of the biggest sources.

What's less believable is that nobody involved in this hypothetical effort at the NSA decided to just get rich in the private sector after coming up with technology decades ahead of the competition.

Guess i was wrong! I'd actually read that post before, seems I forgot.

Sometimes my "real" justifications build on a lot of accumulated knowledge and ideas, and writing those all out would take longer than I wanted, so I don't, and substitute for something shorter instead. Sometimes the shorter thing is wrong, though. So my 'real' reason for saying .1% was something about how mathematics and coordination and coming up with ideas is hard, and as we observe society develop we're seeing the best of everyone we have slowly stumble into being more and more correct, and it's almost impossible to beat that privately on something as big as 'GPT' because you have to do all of the research work that tens of thousands of the brightest machine learning researchers did in public over the past few decades. Like, the manhattan project was secret, but it used all of the best people we had and wasn't secret forever. The NSA can keep some cryptographic techniques secret, but not the entire concept of cryptography secret.

[Omicron]

<1%? My vague memory is that there were a lot of variants, and that in general 'virus mutates to spread more and be less harmful' is fairly common, so imo there's not that much reason to believe this.

For a random variant I'd agree. But omicron was really weird in a lot of ways though, and I'd actually put this one at more like 30% (and 80% that something weird and mouse-shaped happened).

  1. Omicron was really really far (as measured by mutation distance) from any other sars-cov-2 variant. Like seriously look at this phylogenetic tree (figure 1 in this paper)
  2. The most recent common ancestor of B.1.1.529 (omicron) and B.1.617.2 (delta, the predominant variant at the time) dates back to approximately February 2020. It is not descended from any variant that was common at the time it started spreading.
  3. The omicron variant spike protein exhibited unusually high binding affinity for the mouse cell entry receptor (source)
  4. Demand for humanized mice was absurdly high during the pandemic - researchers were definitely attempting to study coronavirus disease and spread dynamics in mouse models.

The astute reader will object "hey that just sounds like a researcher who couldn't get enough humanized mice decided to induce sars-cov-2 to jump to normal mice, and then study it there. Why do you assume they intentionally induced a jump back to humans rather than accidentally getting sick from their research mice". To which I say "the timing was suspicious, the level of infectiousness was enormously higher in humans which I don’t think I'd expect in the absence of passaging back through humanized mice, and also hey look over there a distraction from my weak arguments".

What is this place?

This website is a place for people who want to move past shady thinking and test their ideas in a court of people who don't all share the same biases. Our goal is to optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.

The weekly Culture War threads host the most controversial topics and are the most visible aspect of The Motte. However, many other topics are appropriate here. We encourage people to post anything related to science, politics, or philosophy; if in doubt, post!

Check out The Vault for an archive of old quality posts. You are encouraged to crosspost these elsewhere.


Why are you called The Motte?

A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently, it's an element in a rhetorical move called a "Motte-and-Bailey", originally identified by philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial but high value claim to a defensible but less exciting one upon any resistance to the former. He likens this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired propositions to which one retreats when hard pressed."

On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.


New post guidelines

If you're posting something that isn't related to the culture war, we encourage you to post a thread for it. A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a submission statement. A submission statement is required for non-text sources (videos, podcasts, images).

Culture war posts go in the culture war thread; all links must either include a submission statement or significant commentary. Bare links without those will be removed.

If in doubt, please post it!


Rules


Recommended Posts And Communities

Recommended Realtime Chats