JoelEinbinder

5 ms·

1.
by JoelEinbinder 6mo ago
My understanding is that existing rail lines aren't flat/straight enough for high speed rail. There's no point to a bullet train if it has to constantly slow down for corners/hills.
2.
by JoelEinbinder 2y ago
The benchmark shows lower ram usage on a very simple demo website. I expect that if the benchmark ran on a random set of real websites, ram usage would not be meaningfully lower than Chrome. Happy to be impressed and wrong if it remains low
3.
by JoelEinbinder 2y ago
When I've talked to people running this kind of ai scraping/agent workflow, the costs of the AI parts dwarf that of the web browser parts. This causes computational cost of the browser to become irrelevant. I'm curious what s
4.
by JoelEinbinder 2y ago
Google's monorepo of closed source code.
5.
Show HN: I made an online Python REPL(python.joel.tools)
1 points by JoelEinbinder 2y ago|0 comments
6.
by JoelEinbinder 2y ago
Seems like Hyundai own 33% of Kia, rather than it just being a brand under the same company like Lexus/Toyota. They share some things and compete on others.
7.
by JoelEinbinder 2y ago
I think Hacker News might appreciate some of the behind the scenes of this post. Getting this page to load quickly was not trivial. The initial dataset of books starting sentences was over 20 megabytes. By only sending the unique prefix of
8.
Using the Infinite Bookspace to Reason About Language Models(joel.tools)
1 points by JoelEinbinder 2y ago|1 comments
9.
by JoelEinbinder 2y ago
It would make competitive chess even more draw-ish. It is much easier to see when you accidentally get into a losing position than when you miss a winning idea. So the take back would be used defensively.
10.
by JoelEinbinder 2y ago
On the full set of 1000 questions, the language models are getting 30-35% correct. With patience, humans can do 40-50%. The language models were prompted with the text + each candidate answer, and the one with the lowest perplexity was pic
11.
by JoelEinbinder 2y ago
What scores are you getting using this technique?
12.
by JoelEinbinder 2y ago
After the quiz, the source is linked along with the full comment.
13.
by JoelEinbinder 2y ago
The prompts you see in the quiz are from real hacker news comments. Whatever word the commenter said next is the "correct" word.
14.
by JoelEinbinder 2y ago
Temperature doesn't play a role here, because the LLM is not being sampled (other than to generate the candidate answers). Instead the answer the llm picks is decided by computing the complexity for the full prompt + answer string.
15.
by JoelEinbinder 2y ago
If I used old comments then it's likely that the models will have trained on them. I haven't tested if that makes a difference though.
16.
by JoelEinbinder 2y ago
The language model generating the candidate answers generates tokens until a full word is produced. The language models picking their answer choose the completion that results in the lowest perplexity independent of the tokenization.
17.
by JoelEinbinder 2y ago
The LLM didn’t generate the next word. Hacker News commenters did. You can see the source of the comment on the results screen.
18.
by JoelEinbinder 2y ago
If you want to practice it one question at at time, you set the question count to 1. https://joel.tools/smarter/?questions=1 When I tested it this way it resulted in less of an emotional reaction.
19.
by JoelEinbinder 2y ago
That isn't how it's supposed to work. I mean sometimes you get a supper annoying prompt like ">", but if you guess the right answer it should give you the point. I just checked the two prompts like that, and they seem
20.
by JoelEinbinder 2y ago
I made a little game/quiz where you try to guess the next word in a bunch of Hacker News comments and compete against various language models. I used llama2 to generate three alternative completions for each comment creating a multiple
21.
Are you better than a language model at predicting the next word?(joel.tools)
233 points by JoelEinbinder 2y ago|103 comments
22.
by JoelEinbinder 2y ago
You can set `pipe` to true in puppeteer (default false) here https://pptr.dev/api/puppeteer.launchoptions By default, Playwright launches this way and you have to specifically enable the tcp listening.
23.
by JoelEinbinder 2y ago
I'm not going to invest a drug company with a 90% chance of failure unless I can expect to get a 10x return if it succeeds.
24.
by JoelEinbinder 3y ago
Interesting to me that WebKit gets vertical form controls before MacOS. I don't have any experience with vertically written languages. How important are these controls to computer usage in Japan? Does Windows have them?
25.
EU Poised to Enact Sweeping AI Rules with US, Global Impact(news.bloomberglaw.com)
1 points by JoelEinbinder 3y ago|0 comments
26.
by JoelEinbinder 3y ago
While it's not quite C++, here is Chromium's implementation: https://source.chromium.org/chromium/chromium/src/+/main:v8/...
27.
by JoelEinbinder 3y ago
It's been a few years since I was looking into it, so this might be out of date. But WPE is targeted at kiosks and places you might want to display web content but not have a full web browser. Igalia can sell consulting services to the
28.
Creating an Alternative Browser Engine for iOS(joel.tools)
2 points by JoelEinbinder 3y ago|0 comments
29.
by JoelEinbinder 3y ago
Assuming you are working in tech, I'd avoid investing in tech. You don't want a situation where tech does badly and you are both out of a job and out of your savings.
30.
Federal Trade Commission Launches Inquiry into A.I. Deals by Tech Giants(nytimes.com)
1 points by JoelEinbinder 3y ago|0 comments

More ›