Mikhail Kats

Always a bit disappointing when a proposal gets the highest possible evaluation but still not awarded..

Fortunately this one was very small, so not much harm done

Any Wiley editors here? Do you know why the decision was made to not ask for suggested reviewers?

I am going to write this up in excruciating detail on my Substack on Friday, but here's my working outline of the argument for now.

SCOTUS, tariffs, IEEPA, and the legislative veto: we're all looking for the guy who did this...

blog.mattglassman.net/the-court-th...

For the big quantum centers, the COA list for each proposal is huge, and then there are like 15 center proposals going in

I can easily imagine it being very hard to find people to review these unless you go for international reviewers, and even then

I suspect that it's an issue for big centers

National Science Board 496th Meeting YouTube video by National Science Board

National Science Board meeting starts in 20 mins—I'll try to follow along and do some live coverage.

Open session will discuss merit review and changes to agency management/decision-making.

Is there any headline that ends in a question mark that doesn't follow Betteridge's law?

I did not know this was a thing

Another aspect is if something happens to the network, it can be duplicated and the user base can easily move over

Is there an econ term for something valuable, but that can't be realistically bought/sold because the moment it is sold, it loses a ton of value (due to the value being from authenticity / credibility / independence)

That's what we need an open social network to be, and bluesky is a good shot at it

I think in some cases it's actually quite hard, but it's a problem worth commiting time and resources to solve

Unfortunately the incentives may not be aligned

Well, it took 10 months, but I'm glad the Supreme Court agrees

I would like to make congress vote now about whether this law shluld remain as is..

This would be both very helpful to verify correctness, and would enable eventual fair distribution of $ that accrues to the companies making models, back to various primary sources

Without that mechanism, we might eventually lose swaths of papers, journalism, open-source code, commentary, etc

Every output could use a toggle or secondary output where you get 100s of citations that trace back where everything came from, including (as best as possible) the knowledge that's baked into the LLM via training

Maybe this can be adversarial, with secondary agents figuring out attributions

LLM-based models/agents need much better functionality to explicitly credit their knowledge/thinking/code/etc

I know it's far from trivial for something baked into the model vs explicitly accessed via a search call, but I wish more time and attention was spent on figuring this out

On one hand, yes. On the other:

One of my students got a FINESST grant in astrophysics in the most-recent cycle, and that $ arrived okay, but I don't know anything beyond that

I'd like to rebalance, but that just does not seem to be in the cards

It's a combination of the chaos at the federal level and an expanding range of university responsibilities due to becoming more senior + senior folks retiring...

...but, over the last year, I've done way less science and mentorship than I would've want to, and way more admin/budgeting/proposals

Yep, there it is..

Hope you have the $$ in the bank to do so!

But you are probably not measuring precipitation in inches

Apparently the forecast for tonight is between 0 and 12 inches of snow, which is quite the range

Maybe one issue is that all of the models (and almost all instances of all models) have roughly the same knowledge, and that knowledge is a large fraction of all written text

Would there be many good human writers, if we all had the exact same knowledge and experience?

"Create a PDF space"

Create a what?

I'm on a faculty search committee this year, somehow for the first time in my career

It's a ton of work!

Looking at it the next day, I should probably try to have error bars (clouds instead of points) for ingredient uncertainty. And it really needs more than three dimensions (sugar?)

Should I be doing this instead of the several things on my plate that are in the top-left corner of the Eisenhower matrix?

I should not. And yet here we are

Breakfast phase diagram based on the fun idea by @moultano.bsky.social