bjourne

48 ms·

961.
by bjourne 2y ago
Proofreading. You have to be careful though, sometimes the model just bullshits.
962.
by bjourne 2y ago
Yes, but there are two easy solutions: 1. Larger nurseries to minimize churn in older generations. Fragmentation is a symptom of too many "false survivors". 2. Concurrent heap compaction.
963.
by bjourne 2y ago
You have private, fixed-size, copy-collected spaces for each thread.
964.
by bjourne 2y ago
But this only happens if the copying collectors' space is growable and discontinuous. I think a better scheme is to just use copying for the young generation and mark and sweep for the older ones. Structuring the heap into blocks seem
965.
by bjourne 2y ago
The smaller the tokens the more guesses the LLMs need for generating text. LLMs are really good at guessing, but the more they have to guess the likelier for them to lose their train off thought.
966.
by bjourne 2y ago
Labour won 33.7% of the popular vote and 63.5% of parliamentary seats. The British political system is just as broken as the US one. In 2019 Labour won 32.6% of the popular vote so they have actually not grown that much. It is the Tories th
967.
by bjourne 2y ago
Good writeup and commendable of you to make your benchmark so easily repeatable. On my 16-core Xeon(R) W-2245 CPU @ 3.90GHz matmul.c takes about 1.41 seconds to multiply 8192x8192 matrices when compiled with gcc -O3 and 1.47 seconds when co
968.
by bjourne 2y ago
You're conflating data misses with instruction misses. Besides, the article is about hundred-megabytes-binaries (forgot to strip debug data?), which the Linux kernel isn't.
969.
by bjourne 2y ago
If it increases performance by 5% then 5% of execution time was spent missing branches and instruction cache misses... Which sounds utterly implausible. Conventional wisdom has it that instruction caching is not a problem because whatever t
970.
by bjourne 2y ago
Population growth rate. Globally, we're about 0.9%/year right now.
971.
by bjourne 2y ago
But is the evidence sufficient to refute the null hypothesis, that coffee had no effect on the American civil war? Since the 19th century is behind us, randomized controlled trials is of course impossible, but there may be other evidence. D
972.
by bjourne 2y ago
But the images your network generates look nothing like MNIST digits. There is also no variance. For example, all 9s are identical.
973.
by bjourne 2y ago
Couldn't the overdetermination problem be solved by tagging intervals with their ids? E.g., "x*x" should be resolved as "(x, i[-3, 3])*(x, i[-3, 3])" and "i[-3, 3]*i[-3, 3]" as "(nil, i[-3, 3])*(nil,
974.
Ask HN: Why does office printing suck so much?
12 points by bjourne 2y ago|10 comments
975.
by bjourne 2y ago
Do not give anyone the wrong idea that "it will work exactly like before - only better!" Migrations are always bumpy roads and there are always people who hate new things. You don't want to give people who complain about stuf
976.
by bjourne 2y ago
> If employer A has profit to spare and could make even more profit if he had more employees, what logic would lead him to NOT raise wages and steal workers from employer B? Cause employer B is his buddy and they regularly play together
977.
by bjourne 2y ago
15 years ago I ran a website (Django+Postgres+memcached) serving 50k unique daily visitors on a dirt cheap vps. Even back then the scalability issues were overstated.
978.
Neural Architecture Search Papers(github.com)
2 points by bjourne 2y ago|0 comments
979.
Apple Matches Worker Donations to IDF and Illegal Settlements, Employees Allege(theintercept.com)
3 points by bjourne 2y ago|0 comments
980.
by bjourne 2y ago
Fool me once, shame on you. Fool me twice... you can't fool me twice!
981.
by bjourne 2y ago
If you define a "move" as a piece moving from one square to another, which I think is the correct definition, then the rarest move is probably an illegal move. Like black king jumping from b4 to h8 or something. Or a piece moving
982.
by bjourne 2y ago
Here is something I often do with ChatGPT: "Proof-read this text for me: <paste text>" And it obeys! The hype is well-deserved imo.
983.
by bjourne 2y ago
While that is true, for correctness appends should supplant updates. On HN, many users (like myself) have posted comments for over a decade. Suppose I changed my username from "bjourne" to "SpongeBob"... Should the comm
984.
by bjourne 2y ago
A key is any set of unique columns. The primary in primary key just hints to the RDBMS how to store the data. Both URLs encode queries on the data. The first says "give me the 2023 review(s) of McDonald's in Berlin" (i.e., it
985.
by bjourne 2y ago
Eh, that's a feature of natural keys. If you try to fuck up your model your RDBMS will complain.
986.
by bjourne 2y ago
But what's the point? You're complicating the data for no apparent gain. Only terrible RDBMSes which don't support multi-column keys require surrogate keys.
987.
by bjourne 2y ago
I think you are confused about the terminology because a primary key is a uniqueness constraint. They can be changed in any RDBMS worth its salt. Keys should not be random and your URL example is a case in point. The url /germany&#x2
988.
by bjourne 2y ago
Hard disagree. You create tables for them to be queried by applications. For the Restaurants example the application clearly wasn't designed to handle Restaurant entries whose only field differs is rank. Consequently, you shouldn'
989.
by bjourne 2y ago
I can tell you about the software side on which I have some experience. Spiking neural networks are strictly more powerful than conventional neural networks which in turn are strictly more powerful than hand-coded rules engines. So the idea
990.
by bjourne 2y ago
Afaik, this technique is called threading ( https://en.wikipedia.org/wiki/Threaded_code ) and has been used in most bootstrap compilers for decades.

More ›