Several Posts at Once
NeruIPS: Three weeks ago, I went with my whole family to the NeurIPS conference in Vancouver. We got to see some of Julie's Toronto cousins on the way, and met up with the family of one of Julie's close friends who I don't think we'd seen since the wedding. So that was very nice.
The conference itself was interesting, though I didn't have enough background to understand a lot of it. I'm also not used to academic conferences. There are a lot more short talks (basically lightning talks / abstracts / intros for the poster sessions), there's a lot more focus on the poster session. The talks tended to be a lot of math very fast which went over my head, the poster sessions were more interesting because I could get the presenters to tailor their explanations to my (lack of) theoretical background. Pretty cool, like a top journal digest with live author Q&A.
The conference definitely painted a very vibrant field. There are interesting strategic and tactical things going on in terms of methods for developing better expert systems, fundamental breakthroughs and detailed optimizations in the underlying algorithms. There are interesting developments in the implementing software and hardware as well, though this conference focused way more on the math as opposed to the hardware and software frameworks. It was still very applied computer science, there was a lot of empirical results and not much in the way of theorems and proofs. Wasn't exactly machine learning as the study of black-box systems, but the boxes were only slightly translucent. The implementing software (and hardware) frameworks/tools/methodologies were basically elided over. Unfortunately for me, that's the bit I'm the most interested in.
I do think that there are going to be some real big developments coming out of the area relatively soon. For one thing, I expect automated driving will be very good (i.e. quite substantial improvements in safety over human drivers while equivalent or better in other relevant ways) relatively soon. I don't know if we'll see wide-scale deployments of that technology within five years or 2-3 decades, and I don't know how incremental the change will be if it's the latter, but I'd be pretty confident on "within my lifetime".
I also found discussions of AI safety very interesting. Both "safety" in the sense of practical applications (e.g. how do you train a model to do well under conditions under uncertainty, to predict the behavior of other agents, to behave in such a way that its own behavior is predictable and to take that predictability into account when modeling how others will react; not doing this sort of thing) and in the sense of alignment (i.e. how do you know that your expert system is "pointed" in a good direction, and that it will continue to be as it improves ("capability amplification")). I don't know that I'm worried about the "strong" version of AI safety risks (e.g. AI gains a "decisive advantage over humanity" and makes a "treacherous turn", or becomes so smart and totalizing that it's a hazard to anything nearby that's made out of stuff). But the weak version (people are going to become quite good at capability amplification, if they're not good at AI alignment by that point, there will be a lot of quite-capable expert systems pursuing goals that no one would want them to pursue) seems worth concern. And of course there's the usual concern with technology, that effective tools can be intentionally turned to evil purposes which it would be good to be able to defend against.
Erica seemed to enjoy her time in "conference camp". Was a long day running around at the daycare, and she seemed to have a good time playing with the other kids. Made for Erica going "early" to bed after a late post-conference dinner. ("Early" as in still 10PM, but the journey to slumber was nice and prompt.)
Christmas: Last week, we went to visit Julie's sister's family, along with her parents, in Rockwall (near Dallas, Texas), where they moved recently in (just now realized) anticipation of the birth of two (!!) new family members. Was nice to see people, and the twins didn't arrive unexpectedly early, so we were able to see everyone before things got unreasonably hectic. Their new house / neighborhood seem really lovely.
Too Much About Cats: We went to see the movie Cats, since the cousins are big fans of the musical. Sure is something. It's... worth seeing? If you like the musical, anyways? Might be a bit much if you're not already familiar with the musical, which is quite bizarre to begin with.
I wouldn't say it's a good adaptation of the musical. Well, I don't think Cats (2019) is as good as Cats (1998), and the latter a film adaptation only trivially; it's a recording of a stage performance of the musical with some minor adjustments. But Cats (2019)'s failures as an adaptation are interesting, and there's so much talent and effort crammed into the thing and then it's still somewhat of a complete mess.
It's a setting where cats are people (obviously, given the source material), but people are also cats. (Maybe??? Or maybe it's just aspects of the "human" environment are interpreted by cats as being "for cats" in ways that are just a visual metaphor. "It's not clear what's a visual metaphor and what's part of the actually-seen environment" is a pervasive problem for the film, in part because the level of visual metaphor expected in theater is way higher than what you expect from film, where things depicted are usually things that are actually visually there, at least seen from some perspective.)
The whole thing is in the uncanny valley in so many different ways. In many instances, it seemed to have great ideas for a high-budget theatrical run and great ideas for a film adaptation, but ideas which often don't work well together. In several scenes, it would have benefited from more reserved transitions, where frenetic camera work makes apparently really-well-choreographed song-and-dance numbers hard to follow. (Unless that's all papering over those bits taking a million takes and/or being actually terrible.)
I don't know if it will be forgotten or somehow become a cult classic, but maybe you want to get in on the ground floor of the global phenomenon or whatever. Also, everyone should go see it so that it makes a bajillion dollars and inspires a big-budget Hollywood treatment for Starlight Express.
Also, the President Was Impeached? I haven't been talking politics here because it's hard to keep up with President Gallop and it's generally exhausting communicating across a partisan perception gap that's way bigger than it ever was with Clinton. But President Trump has been impeached, and I'm not going to let something that significant pass by without comment. Trump's been impeached for withholding congressionally-authorized funding to a ally, a portion of whose territory is currently under military occupation, on the condition that they announce a criminal investigation of the son of his leading political opponent (with concern only about this particular investigation and concern only for the announcement).
He's also been impeached for stonewalling all congressional oversight, not on the basis of executive privilege, but a theory of absolute immunity of the entire executive branch to compulsory process. Instead of the legislature as a co-equal branch of government, the Trump administration at best thinks it would take both the legislative and judicial branches together to compel the executive branch to cooperate to any degree with any sort of oversight. Of course, as soon as impeachment happened, the switch from "no testimony, no documents for the House" switched immediately to, "How can you possibly ask for testimony in the Senate, wasn't that supposed to be done in the House?" was immediate.
The Senate Majority Leader announcing in advance, "I'm not impartial about this at all," seems crazy by historical standards. Am I just wrong about this? Did Senate leaders say similar things about Clinton? Nixon? Do they intend to amend the usual oath sworn before an impeachment trial, or swear an oath they've publicly vowed to violate? Given that, it does make sense to delay the appointment of managers and the delivery of the articles of impeachment. What's the advantage of appointing someone to prosecute a sham non-trial followed by a sham quickie acquittal? Remember the witnesses Democrats want to testify are Republican appointees, members of the administration. But the Senate Republicans seem to realize that "Team Trump" doesn't equate to "team maybe go to jail for the guy" even for supporters (e.g. Sondland), and it seems like the sorts of things people might say while diligently trying to avoid perjury (e.g. the truth) might sound pretty bad.
(Side note: Is Pence really so bad from their perspective? But of course, it's not that, it's that while Republicans aren't going to find winning with Trump that easy, they've run themselves into a corner where winning without Trump (or similar?) is going to be really hard.)
It's amazing how relatively bad a position Republicans are in on this, given their Senate majority and the unity of their caucus behind Trump's whatever. It seems that normally "just calmly wait" would be a pretty good counter-strategy. Does stalling a pre-determined non-trial accomplish much? Of course, the wrinkle here is that Trump is not calm, ever. Voting out those first two articles of impeachment doesn't really limit the scope of the House oversight inquiries much, if at all. The courts still will weigh in on those subpoenas eventually. There are lots of other potentially impeachable things Trump has overtly done. There's him running the Presidency as a money-making scheme, for one. There are also things like the obstruction of justice documented in the Muller Report. And even if conduct of the executive doesn't merit impeachment or removal from office, the executive's regulatory powers are a legitimate subject of legislation. Congress therefore has a legitimate reason to access some of the Special Council's grand jury testimony, among many other things. And some of the Harm to Ongoing Matter redactions will merit unredaction once the matter (i.e. Stone) is no longer ongoing (i.e. not in jail).
(Another digression, with only a hair of descent into madness: It's weird to deal with the framing of impeachment where "cheated on his wife and formed a shell corporation to launder hush-money payments during the campaign", "already has several members of his campaign facing felony convictions for actions during and related to the campaign", "has an attitude toward celebrity of 'they let you' get away with whatever", "doesn't seem to take any aspect of the job at all deliberately", "vastly outspends the previous administration on leisure activities, in ways that line his own pockets", "runs businesses that provide an obvious way for corporate executives and foreign diplomats to get cash directly into his (or his family's) hands", "gives his children White House staff positions", and "fired the FBI director and immediately told Russian diplomats that the 'pressure' he personally faced due to the FBI investigation into Russian interference into his election, among others, is now 'taken off'", to name some examples, is just part of the deal. The sort of thing they voted for. Yes, some Democrats wanted Trump impeached immediately, but the sort of conduct that would be universally viewed as impeachable if done by a Democrat was obvious immediately. If the Republicans have become "the party of Trump" to the extent that opposition to this conduct is framed as a partisan matter and used to justify the sort of tit-for-tat where if a Republican gets impeached for any conduct, Democrats also get impeached for any conduct, Trump may well end up being the last not impeached President for a while. I've said it before, it may get crazier before it gets less crazy. Probably will. Happy new year!)
This entry was originally posted on Dreamwith.
comments are there.
The conference itself was interesting, though I didn't have enough background to understand a lot of it. I'm also not used to academic conferences. There are a lot more short talks (basically lightning talks / abstracts / intros for the poster sessions), there's a lot more focus on the poster session. The talks tended to be a lot of math very fast which went over my head, the poster sessions were more interesting because I could get the presenters to tailor their explanations to my (lack of) theoretical background. Pretty cool, like a top journal digest with live author Q&A.
The conference definitely painted a very vibrant field. There are interesting strategic and tactical things going on in terms of methods for developing better expert systems, fundamental breakthroughs and detailed optimizations in the underlying algorithms. There are interesting developments in the implementing software and hardware as well, though this conference focused way more on the math as opposed to the hardware and software frameworks. It was still very applied computer science, there was a lot of empirical results and not much in the way of theorems and proofs. Wasn't exactly machine learning as the study of black-box systems, but the boxes were only slightly translucent. The implementing software (and hardware) frameworks/tools/methodologies were basically elided over. Unfortunately for me, that's the bit I'm the most interested in.
I do think that there are going to be some real big developments coming out of the area relatively soon. For one thing, I expect automated driving will be very good (i.e. quite substantial improvements in safety over human drivers while equivalent or better in other relevant ways) relatively soon. I don't know if we'll see wide-scale deployments of that technology within five years or 2-3 decades, and I don't know how incremental the change will be if it's the latter, but I'd be pretty confident on "within my lifetime".
I also found discussions of AI safety very interesting. Both "safety" in the sense of practical applications (e.g. how do you train a model to do well under conditions under uncertainty, to predict the behavior of other agents, to behave in such a way that its own behavior is predictable and to take that predictability into account when modeling how others will react; not doing this sort of thing) and in the sense of alignment (i.e. how do you know that your expert system is "pointed" in a good direction, and that it will continue to be as it improves ("capability amplification")). I don't know that I'm worried about the "strong" version of AI safety risks (e.g. AI gains a "decisive advantage over humanity" and makes a "treacherous turn", or becomes so smart and totalizing that it's a hazard to anything nearby that's made out of stuff). But the weak version (people are going to become quite good at capability amplification, if they're not good at AI alignment by that point, there will be a lot of quite-capable expert systems pursuing goals that no one would want them to pursue) seems worth concern. And of course there's the usual concern with technology, that effective tools can be intentionally turned to evil purposes which it would be good to be able to defend against.
Erica seemed to enjoy her time in "conference camp". Was a long day running around at the daycare, and she seemed to have a good time playing with the other kids. Made for Erica going "early" to bed after a late post-conference dinner. ("Early" as in still 10PM, but the journey to slumber was nice and prompt.)
Christmas: Last week, we went to visit Julie's sister's family, along with her parents, in Rockwall (near Dallas, Texas), where they moved recently in (just now realized) anticipation of the birth of two (!!) new family members. Was nice to see people, and the twins didn't arrive unexpectedly early, so we were able to see everyone before things got unreasonably hectic. Their new house / neighborhood seem really lovely.
Too Much About Cats: We went to see the movie Cats, since the cousins are big fans of the musical. Sure is something. It's... worth seeing? If you like the musical, anyways? Might be a bit much if you're not already familiar with the musical, which is quite bizarre to begin with.
I wouldn't say it's a good adaptation of the musical. Well, I don't think Cats (2019) is as good as Cats (1998), and the latter a film adaptation only trivially; it's a recording of a stage performance of the musical with some minor adjustments. But Cats (2019)'s failures as an adaptation are interesting, and there's so much talent and effort crammed into the thing and then it's still somewhat of a complete mess.
It's a setting where cats are people (obviously, given the source material), but people are also cats. (Maybe??? Or maybe it's just aspects of the "human" environment are interpreted by cats as being "for cats" in ways that are just a visual metaphor. "It's not clear what's a visual metaphor and what's part of the actually-seen environment" is a pervasive problem for the film, in part because the level of visual metaphor expected in theater is way higher than what you expect from film, where things depicted are usually things that are actually visually there, at least seen from some perspective.)
The whole thing is in the uncanny valley in so many different ways. In many instances, it seemed to have great ideas for a high-budget theatrical run and great ideas for a film adaptation, but ideas which often don't work well together. In several scenes, it would have benefited from more reserved transitions, where frenetic camera work makes apparently really-well-choreographed song-and-dance numbers hard to follow. (Unless that's all papering over those bits taking a million takes and/or being actually terrible.)
I don't know if it will be forgotten or somehow become a cult classic, but maybe you want to get in on the ground floor of the global phenomenon or whatever. Also, everyone should go see it so that it makes a bajillion dollars and inspires a big-budget Hollywood treatment for Starlight Express.
Also, the President Was Impeached? I haven't been talking politics here because it's hard to keep up with President Gallop and it's generally exhausting communicating across a partisan perception gap that's way bigger than it ever was with Clinton. But President Trump has been impeached, and I'm not going to let something that significant pass by without comment. Trump's been impeached for withholding congressionally-authorized funding to a ally, a portion of whose territory is currently under military occupation, on the condition that they announce a criminal investigation of the son of his leading political opponent (with concern only about this particular investigation and concern only for the announcement).
He's also been impeached for stonewalling all congressional oversight, not on the basis of executive privilege, but a theory of absolute immunity of the entire executive branch to compulsory process. Instead of the legislature as a co-equal branch of government, the Trump administration at best thinks it would take both the legislative and judicial branches together to compel the executive branch to cooperate to any degree with any sort of oversight. Of course, as soon as impeachment happened, the switch from "no testimony, no documents for the House" switched immediately to, "How can you possibly ask for testimony in the Senate, wasn't that supposed to be done in the House?" was immediate.
The Senate Majority Leader announcing in advance, "I'm not impartial about this at all," seems crazy by historical standards. Am I just wrong about this? Did Senate leaders say similar things about Clinton? Nixon? Do they intend to amend the usual oath sworn before an impeachment trial, or swear an oath they've publicly vowed to violate? Given that, it does make sense to delay the appointment of managers and the delivery of the articles of impeachment. What's the advantage of appointing someone to prosecute a sham non-trial followed by a sham quickie acquittal? Remember the witnesses Democrats want to testify are Republican appointees, members of the administration. But the Senate Republicans seem to realize that "Team Trump" doesn't equate to "team maybe go to jail for the guy" even for supporters (e.g. Sondland), and it seems like the sorts of things people might say while diligently trying to avoid perjury (e.g. the truth) might sound pretty bad.
(Side note: Is Pence really so bad from their perspective? But of course, it's not that, it's that while Republicans aren't going to find winning with Trump that easy, they've run themselves into a corner where winning without Trump (or similar?) is going to be really hard.)
It's amazing how relatively bad a position Republicans are in on this, given their Senate majority and the unity of their caucus behind Trump's whatever. It seems that normally "just calmly wait" would be a pretty good counter-strategy. Does stalling a pre-determined non-trial accomplish much? Of course, the wrinkle here is that Trump is not calm, ever. Voting out those first two articles of impeachment doesn't really limit the scope of the House oversight inquiries much, if at all. The courts still will weigh in on those subpoenas eventually. There are lots of other potentially impeachable things Trump has overtly done. There's him running the Presidency as a money-making scheme, for one. There are also things like the obstruction of justice documented in the Muller Report. And even if conduct of the executive doesn't merit impeachment or removal from office, the executive's regulatory powers are a legitimate subject of legislation. Congress therefore has a legitimate reason to access some of the Special Council's grand jury testimony, among many other things. And some of the Harm to Ongoing Matter redactions will merit unredaction once the matter (i.e. Stone) is no longer ongoing (i.e. not in jail).
(Another digression, with only a hair of descent into madness: It's weird to deal with the framing of impeachment where "cheated on his wife and formed a shell corporation to launder hush-money payments during the campaign", "already has several members of his campaign facing felony convictions for actions during and related to the campaign", "has an attitude toward celebrity of 'they let you' get away with whatever", "doesn't seem to take any aspect of the job at all deliberately", "vastly outspends the previous administration on leisure activities, in ways that line his own pockets", "runs businesses that provide an obvious way for corporate executives and foreign diplomats to get cash directly into his (or his family's) hands", "gives his children White House staff positions", and "fired the FBI director and immediately told Russian diplomats that the 'pressure' he personally faced due to the FBI investigation into Russian interference into his election, among others, is now 'taken off'", to name some examples, is just part of the deal. The sort of thing they voted for. Yes, some Democrats wanted Trump impeached immediately, but the sort of conduct that would be universally viewed as impeachable if done by a Democrat was obvious immediately. If the Republicans have become "the party of Trump" to the extent that opposition to this conduct is framed as a partisan matter and used to justify the sort of tit-for-tat where if a Republican gets impeached for any conduct, Democrats also get impeached for any conduct, Trump may well end up being the last not impeached President for a while. I've said it before, it may get crazier before it gets less crazy. Probably will. Happy new year!)
This entry was originally posted on Dreamwith.
comments are there.