20 posts tagged “paul-graham”
2024
How cheap, outsourced labour in Africa is shaping AI English. The word “delve” has been getting a lot of attention recently as an example of something that might be an indicator of ChatGPT generated content.
One example: articles on medical research site PubMed now use “delve” 10 to 100 times more than a few years ago!
Nigerian Twitter took offense recently to Paul Graham’s suggestion that “delve” is a sign of bad writing. It turns out Nigerian formal writing has a subtly different vocabulary.
Alex Hern theorizes that the underlying cause may be related. Companies like OpenAI frequently outsource data annotation to countries like Nigeria that have excellent English skills and low wages. RLHF (reinforcement learning from human feedback) involves annotators comparing and voting on the “best” responses from the models.
Are they teaching models to favour Nigerian-English? It’s a pretty solid theory!
2023
Once you've found something you're excessively interested in, the next step is to learn enough about it to get you to one of the frontiers of knowledge. Knowledge expands fractally, and from a distance its edges look smooth, but once you learn enough to get close to one, they turn out to be full of gaps.
2013
Is there a compilation of all of Paul Graham’s essays available online?
Yes: his website. http://www.paulgraham.com/articl...
[... 23 words]What is the Hacker News technology stack?
It’s written in Arc, a Lisp variant created by Paul Graham. I believe it uses the file system for storage rather than a dedicated database.
[... 70 words]Why was the popular Silicon Valley incubator named Y Combinator?
From this interview with Paul Graham http://www.paulgraham.com/frinte...
[... 95 words]Does Paul Graham steal business models from teams not accepted into Y Combinator and feed them to accepted teams as pivot ideas?
No. If he did, word would quickly get around and strong teams would stop applying to YC.
[... 45 words]How does a VC decide to make an investment?
I suggest reading Paul Graham’s latest essay, How to Convince Investors.
[... 27 words]Y Combinator Demo Day and Class of Summer 2013: Is it true that YC alumni are participating in the primary selection process during S13?
Yes—this has been happening for a few years now. Here’s Paul Graham’s announcement about it back in October 2009: Announcement: YC alumni will help us read applications
[... 60 words]2012
What advice does Paul Graham commonly give Y Combinator startups?
Read the essays on http://www.paulgraham.com/—many of them are derived from advice he has already been giving YC startups.
[... 37 words]What’s the real reason for the drop in cash that Y Combinator now gives companies? What really precipitated this change? Were they not seeing the value for their money and wanted out?
I was in the first batch that got the 150k. I believe Paul Graham when he says that the amount was high enough that companies stuck around for longer than they would have otherwise (and founders were more likely to “fight over the corpse”) and that 80k would have been enough to get companies through to the point where they either raise more money or decide not to continue.
[... 124 words]2009
Programmers don't use launch-fast-and-iterate out of laziness. They use it because it yields the best results. By obstructing that process, Apple is making them do bad work, and programmers hate that as much as Apple would.
We advise startups to launch when they've added a quantum of utility: when there is at least some set of users who would be excited to hear about it, because they can now do something they couldn't do before.
What I’ve Learned from Hacker News. I’m always fascinated by online community war stories.
2007
Microsoft saw the danger of Javascript and tried to keep it broken for as long as they could. But eventually the open source world won, by producing Javascript libraries that grew over the brokenness of Explorer the way a tree grows over barbed wire.
If you don't think you're smart enough to start a startup doing something technically difficult, just write enterprise software. Enterprise software companies aren't technology companies, they're sales companies, and sales depends mostly on effort.
2005
Fixing Paul Graham’s Footnotes
I’m a big fan of Paul Graham’s essays, the latest of which is How to Start a Startup. There’s just one niggling problem with them: Paul makes extensive use of footnotes, but provides no way of jumping from the reference in the text to the footnote at the bottom of the page and back up again. Instead, you have to manually down to the bottom of the article and back up again every time you hit a footnote reference.
[... 172 words]2004
Hackers & Painters (via) Paul Graham’s new book.
2003
Fighting Filters and DDoS
Paul Graham’s essays on fighting spam are generally excellent; it was Paul who sparked the recent flurry of activity surrounding Bayesian statistical filters and inspired the creation of some of the best tools for fighting spam yet. Paul’s latest suggestion, Filters that fight back, seems to me to miss the mark in a big way. Paul suggests email servers should “follow” links in any email received. This would turn the tables on spam, as suddenly sending out a million spams would result in a million useless hits to the site being promoted, quickly brining it to its knees. It’s a great concept, until some malicious script kiddie realises that they’ve been handed a tool to run massive distributed denial-of-service attacks on any domain they care to target. Not to mention that such a feature would make many legitimate mass email tools prohibitively expensive to run.
[... 190 words]Spam conference
It sounds like Paul Graham’s Spam Conference was a huge success, with attendance rocketing to 560 from the original estimate of 50—60. Scott Johnson sings its praise and promises a full write up later on. In the meantime, webcasts of the talks are available on the conference website.
2002
A plan for spam
Paul Graham: A Plan for Spam. Paul suggests using content based filters that learn from users specifically marking messages as spam or legitimate mail. The system then picks emails apart looking for commmon terms (in both the body and the header of the message) that can then be used later on to identify spam messages. He claims his test have let through only 5 per 1000 spams, with 0 false positives
. Impressive stuff, and great reading for the excellent explanations of some advanced alogithmic and statistical techniques.