medvezhenok

6 ms·

1.
by medvezhenok 14d ago
I think you’re missing that MTP can predict more than 1 token in advance.
2.
by medvezhenok 3mo ago
Not necessarily. House occupancy laws also have a role here - some of the living arrangements that ended up in higher density in the past are illegal today, forcing population decline in a given area (which is not necessarily equivalent to
3.
by medvezhenok 3mo ago
You can get AppleCare+ in the U.S. for $149/year which is just as good (or better) than any warranty.
4.
by medvezhenok 4mo ago
Yup, especially data backups (although I wouldn't store critical backups like this, only nice-to-have ones). One minor note is that parquet file sizes / compressed sizes can be sensitive to ordering, so you can try different sort
5.
by medvezhenok 4mo ago
Yup. And an extra benefit that you can treat any file like a table, so you can also do something like UPDATE my_table SET x = file1.x, y = file2.y FROM 'first_file.csv' file1 LEFT JOIN 's3://my_bucke
6.
by medvezhenok 4mo ago
Claude code can write exploratory queries for you to give you a quick rundown on the shape of the data set, frequencies, missing values, etc etc (without having to load it into a more persistent data store or writing custom python scripts).
7.
by medvezhenok 4mo ago
Basically like a locally hosted Snowflake - it only shines if you have enough data to analyze (100 MB - 100 GB is probably the sweet-spot range - less than that and the benefits are small, more than that and you risk flying too close to the
8.
by medvezhenok 4mo ago
DuckDB has been probably my most used tool in 2026 - if you're comfortable with SQL it's incredible at quickly prototyping and slicing / dicing data. I do a lot of experiments with regexes, and if you get used to the RE2 synt
9.
by medvezhenok 4mo ago
Soviet Union had childlessness tax, but you were exempt if you could prove infertility
10.
by medvezhenok 8mo ago
What exactly would that evidence look like, for you? It definitely increases some types of productivity (Opus one-shot a visualization that would have likely taken me at least a day to write before, for work) - although I would have never w
11.
by medvezhenok 8mo ago
Short answer: there is none. You can't get frontier-level performance from any open source model, much less one that would work on an M3 Pro. If you had more like 200GB ram you might be able to run something like MiniMax M2.1 to get la
12.
by medvezhenok 10mo ago
I'm curious about book recommendations on this (as someone raising kids in the US but originally from Russia)
13.
by medvezhenok 10mo ago
Yeah, I wonder if you plotted crime rate vs time spent outside or something like that (car accident rates are usually reported as an average of an accident / # of miles, since how much you drive changes your likelihood of being in an a
14.
by medvezhenok 10mo ago
The core premise (benchmarks are broken), might be correct, but the poverty benchmark he uses is a bad example. The OPM and SPM (supplemental poverty measure, developed in 2009-2012), disagree by less than 10%; and the latter takes into acc
15.
by medvezhenok 10mo ago
The core premise (benchmarks are broken), might be correct, but the poverty benchmark he uses is a bad example. The OPM and SPM (supplemental poverty measure, developed in 2009-2012), disagree by less than 10%; and the latter takes into acc
16.
by medvezhenok 10mo ago
The strongest argument is probably that for someone subsisting on the minimal wage, the CPI is not a good representation of their consumption basket (whereas it might be for someone close to the median). Therefore the adjustment should prob
17.
by medvezhenok 11mo ago
The increasing levels of abstraction work only as long as the abstractions are deterministic (with some limited exceptions - i.e. branch prediction/preloading at CPU level, etc). You can still get into issues with leaky abstractions, b
18.
by medvezhenok 11mo ago
Human societies? No. Subcultures? Some are at least trying to (i.e. rationalists), though imperfectly and with side-effects.
19.
by medvezhenok 11mo ago
Sure, but LLMs tend to be better at navigating around documentation (or source code when no documentation exists). In agentic mode, they can get me to the right part of the documentation (or the right of the source code, especially in unfam
20.
by medvezhenok 11mo ago
Indeed. But the unintended consequence (perhaps) of LLMs making things easier to use is that more people will use them - basically Jevons paradox. I would expect that this will cause certain programs to see more demand than the creators ant
21.
by medvezhenok 1y ago
I think this was true before AlphaZero; now, I'm not sure there's any benefit to having a human + engine versus just an engine alone. From my understanding the AlphaZero based engines (i.e. neural nets) do not really benefit from
22.
by medvezhenok 1y ago
I think they said M4 Ultra Studio is not going to happen, have to wait for M5 Ultra...
23.
by medvezhenok 1y ago
Race and culture are not the same thing and it’s a big pet peeve of mine that Americans always link the two. A Slav that was born and grew up in China would be Caucasian but culturally Japanese. Accusations of racism should stay limited to
24.
by medvezhenok 1y ago
Right. I would say the most apt analogy is from poker. In Poker, luck plays an integral role in the outcome of any specific game or match, but skill does show up when collected over a large enough sample (that's why they say you can&#x
25.
by medvezhenok 1y ago
As someone with diagnosed ADHD, and prescribed something similar: It's not that they make you superhuman - I don't think I'm any "smarter" on them. It's just that without them, my "productive" bursts
26.
by medvezhenok 1y ago
Whether its genetic or environmental doesn't matter here. Existence of a correlation is enough reason to break down any analyses by demographic data to have a clearer picture of what's going on. That's just basic data science
27.
by medvezhenok 1y ago
Yup, more article slop without accounting for demographic data. Same with the constant drumbeat of "Americans are getting shorter".
28.
by medvezhenok 1y ago
Indeed. The LLMs have been pretty useful for greenfield projects & one off scripts for a while, but GPT-5 was the first time I've found a model to be quite helpful on large-scale legacy code (>1M LOC).
29.
by medvezhenok 1y ago
I think if we agree that there are individual differences in predisposition towards juggling aptitude, and that the predisposition is mediated genetically somehow, and if juggling (in this hypothetical) is biologically advantageous for surv
30.
by medvezhenok 1y ago
I understand the argument, but I think you're missing the nuance somewhat. There are a series of things that are learnable mental tricks; I have read Moonwalking with Einstein and am well aware about rhyme techniques, memory palace tec

More ›