shivam mamgain shivamMg

Maybe you've heard about this technique but you haven't completely understood it, especially the PPO part. This explanation might help.

We will focus on text-to-text language models 📝, such as GPT-3, BLOOM, and T5. Models like BERT, which are encoder-only, are not addressed.

Reinforcement Learning from Human Feedback (RLHF) has been successfully applied in ChatGPT, hence its major increase in popularity. 📈

RLHF is especially useful in two scenarios 🌟:

You can’t create a good loss function
- Example: how do you calculate a metric to measure if the model’s output was funny?
You want to train with production data, but you can’t easily label your production data

First of all, please note that token expiration and revoking are two different things.

Expiration only happens for web apps, not for native mobile apps, because native apps never expire.
Revoking only happens when (1) uses click the logout button on the website or native Apps;(2) users reset their passwords; (3) users revoke their tokens explicitly in the administration panel.

1. How to hadle JWT expiration

A JWT token that never expires is dangerous if the token is stolen then someone can always access the user's data.

Quoted from JWT RFC:

Hello software developers,

Please check your code to ensure you're not making one of the following mistakes related to cryptography.

I. General Mistakes

Writing your own home-grown cryptography primitives (For example: Mifare Classic)
- Exception: For the sake of learning, but don't deploy it in production.
Using a fast hash function (e.g. MD5, SHA256) for storing passwords. Use bcrypt instead.
Not using a cryptographically secure random number generator

This is just a collection of thoughts and feelings about the technology industry and guidelines I feel should be upheld.

Public Speaking

Don't Present Original Research at Expensive Events

If a minimum wage employee cannot reasonably afford to attend an event (e.g. saving $300 for DEFCON is probably the upper limit), original research should NOT be presented at that event.

Presenting cutting-edge ideas to the wealthy only serves to insulate the fat cats from the disruptions of the poor. There are plenty of other researchers that hunger for career progression that will serve the whims and aims of the upper class that can afford to drop several thousand dollars on a technology conference.

This is one chapter of my "Chrome Extension Workshops" tutorial, see the rest here: https://gist.github.com/caseywatts/8eec8ff974dee9f3b247

Unrelated update: my book is out! Debugging Your Brain is an applied psychology / self-help book

Making Bookmarklets

I'm feeling very clever. I've got this sweet line of javascript that replaces "cloud" with "butt". My mom would LOVE this, but she doesn't computer very well. I'm afraid to show her the Developer Console and have her type/paste this in. But she IS pretty good at bookmarks, she knows just how to click those!

A bookmark normally takes you to a new web page. A bookmarklet is a bookmark that runs javascript on the current page instead of taking you to a new page. To declare that it is a bookmarklet, the "location" it points to starts with javascript:.

	import numpy


	def wagner_fisher(s: str, t: str):
	"""
	Computes the Levenshtein distance between the two strings. Returns a tuple containing
	the distance itself and also the entire matrix for further processing.
	See: https://en.wikipedia.org/wiki/Wagner%E2%80%93Fischer_algorithm
	"""
	m, n = len(s), len(t)

	class E(BaseException):
	def __new__(cls, args, *kwargs):
	return cls
	def a(): yield
	a().throw(E)

	/**
	* = Recursive descent parser =
	*
	* MIT Style License
	* By Dmitry Soshnikov <dmitry.soshnikov@gmail.com>
	*
	* In this short lecture we'll cover the basic (non-predictive, backtracking)
	* recursive descent parsing algorithm.
	*
	* Recursive descent is an LL parser: scan from left to right, doing

	from __future__ import with_statement

	from kombu.common import maybe_declare
	from kombu.pools import producers

	from queues import task_exchange

	priority_to_routing_key = {'high': 'high',
	'mid': 'mid',
	'low': 'low'}

	-- show running queries (pre 9.2)
	SELECT procpid, age(clock_timestamp(), query_start), usename, current_query
	FROM pg_stat_activity
	WHERE current_query != '<IDLE>' AND current_query NOT ILIKE '%pg_stat_activity%'
	ORDER BY query_start desc;

	-- show running queries (9.2)
	SELECT pid, age(clock_timestamp(), query_start), usename, query
	FROM pg_stat_activity
	WHERE query != '<IDLE>' AND query NOT ILIKE '%pg_stat_activity%'