Jai

@kajsotala.bsky.social

This is a good example of why property crimes shouldn't be passively tolerated.

is fiction a superstimulus?

Okay yes, but also classic sci-fi: AI will speak in incomprehensible verse briefly punctuated by actual commands and meaningful output

O3 CoT:

A bar chart illustrates the estimated lives saved each year by various American foreign aid programs, totaling approximately 3.3 million lives saved annually. The programs listed from top to bottom include: - HIV/AIDS: 1.6 million lives saved per year - Humanitarian aid: 550,000 lives saved per year - Vaccines: 500,000 lives saved per year - Tuberculosis: 310,000 lives saved per year - Malaria: 290,000 lives saved per year At the bottom, a note indicates that the figures represent central estimates and that actual estimates may range from 2.3 to 5.6 million lives saved. It clarifies that these numbers do not encompass other vital forms of aid such as water and sanitation, nutrition, and family planning. The source of the data is credited to Kenny & Sandefur, 2025. The visual includes a label stating "Our World in Data" and is presented under a creative commons attribution license (CC BY).

✍️ New article: “Foreign aid from the United States saved millions of lives each year”

For decades, these aid programs received bipartisan support and made a difference. Cutting them will cost lives.

Positional embeddings/every alternative to positional embeddings don't feel great either

I think it might be possible to do it intentionally but not overtly. As a proof of concept, there are lots of subcultures that deliberately project off-putting elements. Hazing rituals are awful but they exist to serve the same function. There are probably other underexplored strategies.

Part of the trick is that the gates didn't feel like gates - this was just a place to go for some people who absentmindedly jumped the barriers. If someone could replicate that trick again...

You can only have things like The Old Internet by gating them. The old Internet was gated by knowledge and inconvenience for a while, but when those gates fell so did the golden age. If you want to make something like it again, you need to build it in a way that most people won't ever really use it

80% folk theory: the architecture + most foundational training data has an inductive bias for modeling rigid bodies, where linear transforms excel. Getting from there to accurately modeling e.g. fabrics is Difficult.

Holy shit the rapture was today

Ah yes, the "Preoccupied Parser" numbers

This one: www.smbc-comics.com/comic/2013-0...

Still think it's true, don't at me.

Also the establishment economist and head of the federal reserve is the most popular appointee - looks like elitist neoliberalism is back on the menu 🥰

Kind of surprised Trump is faring worse than RFK.

"fiery lawsuit"

No one has been more vindicated than normie resistance Dems, women’s marchers and never-Trump Republicans. They were right all along about the threat. The anti-anti-Trump Republicans and reactionary centrists, and the far left “it’s always been this bad” folks have been proved wrong over and over.

It's good to be able to form alliances with people who you would voraciously oppose in other contexts. If you only form alliances with people who already agree with you about everything you will find it much harder going.

His argument doesn't depend on that. You are never going to get a free society without lots of people promoting ideas that undermine that free society. His point is that you need to find a way to navigate that.

I liked it. The gist is that affirmative action isn't primarily about subsidizing disadvantaged groups, but about making universities more heterogenous, interesting, and reflective of society. From that it follows that you want AA for basically any severely underrepresented demographic.

Floppy disks were protective artifacts, everyone who touched one was able to resist the demons.

You should consider going into horseracing, because that's thorough bread content.

Based on my experience with coding agents, doing much more work than anyone asked for is a pretty common LLM pattern.

Rewatched Dr. Strangelove last week, a movie about a superpower losing control of armed aircraft and subsequently coordinating to shoot their own craft down to avert a potentially catastrophic conflict.

These may be calibration questions. You know ahead of time about how many people have these very common names, so you can use the resulting distributions to sanity check your data.

I haven't been paying attention to 2028, what seems most implausible about this distribution to you? Off the top of my head Newsom should be closer to 15%, but also that's about the level of nonsense I expect in markets without insanely high volume.

Mamdani is charismatic *and* didn't build a campaign team that systemically buried his charisma.

Although there are many bespoke, specialized approaches to learning the bitter lesson, over time brute experience tends to dominate all other pedagogical approaches.

Sample from interpolated points?

wip

fixed

actually fixed

wip

fuck, why does this keep breaking

tests (wip)