And, as Joe points out, this would be a non-issue if ASR had even the mildest of replication policies.
07.03.2026 21:48
π 0
π 0
π¬ 0
π 0
And, as Joe points out, this would be a non-issue if ASR had even the mildest of replication policies.
Reading the post, they seem to think that if ChatGPT can guess regression results that are close to true regression results, then researchers will somehow become more likely to fake results. I think your (and my) confusion stems from the fact that this conclusion doesn't make a lot of sense.
Analyst: βI think weβre weeks away from maybe flight cancellations or delays due to lack of jet fuel, rather than monthsβ
"SHIT"
"FUCK"
"NO"
"FUCK"
"SHIT"
"FUCK"
I feel almost the opposite. Google became terrible years ago. Stuff that is more than a few years old became especially hard to find. LLMs are the only things that seem to remember stuff that was on the web from more than a decade ago, probably due to training on the Wayback machine.
Interesting. I switched to MacOS over a decade ago. because it was a good desktop experience, but the BSD / *nix / Darwin underpinnings meant I had access to most of the cli tools I use on Linux, especially with homebrew.
Whatβs missing from your perspective?
I feel almost the opposite. Google became terrible years ago. Stuff that is more than a few years old became especially hard to find. LLMs are the only things that seem to remember stuff that was on the web from more than a decade ago, probably due to training on the Wayback machine.
We have to go back.
One of the few Luddite things I have done is never connecting my tv to the internet.
βDART was the first hypervelocity impact experiment on an asteroid at size and velocity scales relevant to planetary defence, intended to validate kinetic impact as a means of asteroid deflection. DART kinetic impact was highly effective in deflecting the asteroid Dimorphos.β
Like many methodologists, I often find myself arguing with someone that their method/epistemology makes no sense. That someone thinks I am arguing against their *conclusions*. But I am arguing they have no justification for their conclusions. Is there a philosophical term for this problem/confusion?
Great story, but 0% of this sounds fake.
So I agree with all of this, but the provocative claim was that itβs better than most professors now.
Every professor in my department, and most of the professors I know, are very aware of how fast the capabilities are changing.
Whether weβll figure out how to deal with that well, Idk.
Screenshot of the "Does that use a lot of energy?" online app
Hannah Ritchie has built a fun little tool where you can compare energy usage of various products and activities.
This is super helpful imho, because it's so hard to develop intuitions even just about the scales involved here.
hannahritchie.substack.com/p/does-that-...
I agree
Itai Sher @itaisher β’ Mar 2 I don't believe it is true that Al now can do social science better than most professors. 112KεͺδΌ S... Alexander Kustov * @akoustov Perhaps not most of your colleagues, but globally? Having some familiarity with social science in developing countries and the former soviet bloc, much of what gets published in MDPI-type journals isn't meaningfully advancing knowledge. Al can already do better than that. 4:40 PM β’ Mar 2, 2026 β’ 6,029 Views
I think this exchange on X provides important context for the headline claim. I like Claude Code a lot, but I donβt think it could get tenure at most R1 universities in the US (even if you think the research coming out of those institutions is terrible too).
The thing with Liberation Day tariffs, Minneapolis, and all the other crises Trump has started, is that they were in his control to ratchet down when they became unpopular or the bond market got a little too spooked.
I am afraid that this is different.
www.wsj.com/livecoverage...
This is the product I want most from Anthropic.
I think posts like this trade on a kind of ambiguity in the phrase "take seriously." You can take a wrong position seriously in one sense, as one that you should engage with and disprove.
Makes sense. I went straight from Intel to M3 Max on my work laptop and itβs been amazing. Should have switched sooner like you.
Not sure if a parameter like this is in the model (will read carefully later), but I suspect this problem is worse for the executive when a large swath of the public has a strong prior belief that theyβre a bunch of of lying-McLiarsons.
One of the Trump adminβs explanations for striking Iranβpreemption of an imminent attackβput me in mind of my article w/@yujimasumura.bsky.social at @ajpseditor.bsky.social, which examines when govts use the preemption casus belli and how publics react. (1/x)
doi.org/10.1111/ajps...
What upgrades here are most exciting to you? Iβve been thinking I should upgrade my m3 max before some grant money runs out, but havenβt really been paying attention to whatβs happening with mbps.
What do you think is the most likely path for this to happen? Iβm very worried about the war spiraling out of control, but my knowledge of the region is poor, which makes it really hard for me to distinguish which scenarios are more or less plausible.
He canβt taco his way out of this one, and they might be starting to realize it
5 min apart. Read my mind.
bsky.app/profile/brad...
The thing with Liberation Day tariffs, Minneapolis, and all the other crises Trump has started, is that they were in his control to ratchet down when they became unpopular or the bond market got a little too spooked.
I am afraid that this is different.
www.wsj.com/livecoverage...
Two waymos struggle to get past each other. But they do figure it out! And this video makes it incredibly clear they aren't just being teleoperated. The failures are always more informative than the successes!
