1/ One of the issues of fully amortized inference / pretrained simulator-based inference is that you are stuck with the "prior" training distribution. What if you change your mind after training?
In PriorGuide, one of our papers at ICLR this week, we allow the prior to be changed at runtime!
Tbc, I am using pests above sarcastically, to imitate the prevalent rhetoric about them (whether it's wrong or right is different). Just that these are great examples to illustrate how unmoored the no-longer-hermetically-sealed discussion is. It may lead to unpreparedness for actual disruptions.
All three (leaving aside some labour economists) may differ on specifics, but they somehow automatically agree on "white collar jobs" (pests like business consultants, for example). To me it's incredible because they are the hardest to automate.
Much of the AI alarmism around the nature of job market disruptions overwhelmingly comes from the VC ecosystem, tech, academia. Basically the visionary class. No one could guess the reason why....
in the 90s), had this similar flavour of silly anti-anthropocentric-sublime-as-radicalism.
You become a symbol, and one no longer pays attention to what you actually said). But later anthropocene studies downstream of BL, which did not have nearly as much "impact" as the science wars (because no one cares about climate now, people did care about science
I think of Latour as a forerunner of the later day culture war social media hub-personalities (in that the core competence was in wedging in between latent divisions by cleverness, and once people got fighting over broader philosophical / aesthetic disagreements, you automatically become important.
A lot of the "debates" about whether certain llms are conscious or not, and positions around them, end up ironically revealing the intellectual poverty and pseudo-radicalism of object oriented ontology. It's a very different style, different objectives, but they have a non-trivial convergence.
*might well see, not might as well see. Typo completely changed the meaning.
A net positive of the military motivation is that countries with moribund manufacturing capacity will have to get better at it (cheaper, more scalable). The supply chain spillovers could be real.
The military uptake will lead. Because you don't need too much precision there. Just enough to cause damage and absorb risk. The bit about drones first became clear in the Nagorno Karabakh war. Having won in 1994, the Armenian side did not internalize that their conventional deterrence was obsolete.
Before such robots become useful in industrial automation, they unfortunately might as well see attention as cheap tactical asymmetries in warfare (just as drones were originally thought of narrowly as precision strike instruments, but have completely altered the nature of war).
Was trying to get a list of the ones which were autonomous. 18 autonomous teams completed in the race this time vs 29 remote-controlled. The autonomous ones include Unitree H1, Booster T2, Yuanqizai, Tiangong 1.0 Ultra.
TIL: Pagliacci is just Italian for 'Clowns.'
Sorry if my spam / posting frequency goes over the top on weekends. But this topic has been on on my mind more than usual lately.
It is not a study designed to understand the question if the vaccine improves outcomes compared to not getting the vaccine. Last year it did not occur to me to recruit a LLM to answer or opine on it for me. But this time I did and they seem to agree. But it's still something and exciting.
Interesting that a study from last year got press now, and it's easy to misunderstand, but I feel like these should be promoted as much as possible regardless (despite the hype last year). The paper is a bit humbling to try and understand, but one thing is that it's not a controlled efficacy trial.
"Pancreatic cancer mRNA vaccine shows lasting results in an early trial: Scientists caution that more research is needed, but nearly all of the patients who responded to the personalized vaccine are still alive six years later."
These are not just about "liberal thought" as one might reflexively think. But general civic thought, irrespective of political orientation. One has an element of grandeur, and of building a system on virtues. The other is about building something functional around (and using) human vices.
social and political imagination, which are almost archetypical (and amount to a rough binary classification). They are not mutually exclusive archetypes, but they differ in very distinct and fundamental ways.
Talking of books, a few months ago I finally finished reading most stuff by Judith Shklar and Hannah Arendt (beyond the most well-known works).
While I think Judith Shklar >>>> Hannah Arendt, I also think in the two you have two very different sensibilities, different styles of
For quite some time I had not been paying enough attention to them.
I still need to spend some weekends to understand how to use them right (and also the new Qwen model). I feel like it's not trivial to do it in a modular way, but now you have all sorts of tools like Hermes at your disposal. But I am very impressed by these small models.
And they are largely promoted by different corporations. Why does Google release something that can be very good (like Gemma 4) as fully open-source. I think it's great ecosystem strategy. Why does Ali Baba do it (they just had another best in class release)? Also great, but very different strategy.
I specifically ref ticks because he says he admires Gibbon, but that if anyone wrote like that now they'd be "dismissed as a pompous fart." While a spare style is wonderful, I think Gibbon was great. Both deserves and goes beyond the reputation that has come through SM pictures of the Decline by pom
I read the first three volumes back in 2017/18. But then the move to Providence / Cambridge was so disruptive that I forgot about the last two. Surprisingly, I still remember the style, and even the ticks. This a remarkable achievement by The Right Honourable Lord Sumption OBE PC FSA FRHistS.
Was posting about some books, and forgot to include the last volume. So will post about it here without context.
Ahhh I can also see these guys
to her in a different language (which she can understand and speak, but does not default to while talking at home).
I grew up with it so I did not notice it till a few years ago. But every once in a while now I marvel at it. A funny thing about my parents is that my mother talks to my father in one language (dialect, but sounds very different), which my father can only understand, but not speak. He talks
This was a theme during my postdocs (plural even though they happened at the same time). I felt the kind of problems that really mattered were out of bounds without that background. And in ML we tended to have this strange focus on tractable (interesting in a ML sense) but irrelevant bio problems.