Xiaofan Hu bom-d-van

"Don Knuth Plays with ChatGPT" but with ChatGPT-4

1. Tell me what Donald Knuth says to Stephen Wolfram about chatGPT.

I'm sorry for any confusion, but as an AI, I can't provide real-time updates or specific conversations between individuals beyond my last training cut-off in September 2021. I don't have access to the internet or any databases for real-time information or private communications. Therefore, I can't give you the details of a conversation between Donald Knuth and Stephen Wolfram about ChatGPT, if such a conversation has taken place after my training data.

However, it's worth noting that both Donald Knuth and Stephen Wolfram are significant figures in the fields of computer science and computational mathematics. Knuth is widely recognized for his multi-volume work "The Art of Computer Programming" and for creating the TeX computer typesetting system, while Wolfram is known for his work in developing Mathematica and Wolfram|Alpha, and for his contributions to cellular automata theory, outl

Training open-source LLMs on ChatGPT output is a really bad idea.

Everyone is now racing to create open-source alternatives to compete with GPT3.5/GPT4. A common shortcut used by some teams to bootstrap their effort is to fine-tune their model on ChatGPT output. I used to think it was a good idea and totally fair play to do this. Actually, I still think it’s fair play. OpenAI effectively distilled the entire web into its models. They are saying themself that they are using publicly accessible information (mostly). So distilling their model is, in effect, distilling the public open web, so small Term of Service details aside, I don’t see major ethical problems with that. Right? Well, it’s not entirely true and I realized now that, even when ignoring the ethical considerations, using their output is a really bad idea.

First of all, from a purely technical point of view, as @yoavgo is explaining it beautifully in his recent post, there is no way to align LLMs correctly without the RLHF component. I encourag

Purpose

Bootstrap knowledge of LLMs ASAP. With a bias/focus to GPT.

Avoid being a link dump. Try to provide only valuable well tuned information.

Prelude

Neural network links before starting with transformers.

Twitter thread: https://twitter.com/theshawwn/status/1456925974919004165
Hacker News thread: https://news.ycombinator.com/item?id=29128998

November 6, 2021

How does JAX allocate memory on a TPU?

jnp.device_put(1) is deceptively simple to write in JAX. But on a TPU, what actually happens? How does a tensor containing the value 1 actually get onto a TPU?

Turns out, the answer is "C++", and a lot of it.

About

I faced bandwidth issues between a WG Peer and a WG server. Download bandwidth when downloading from WG Server to WG peer was reduced significantly and upload bandwidth was practically non existent.
I found a few reddit posts that said that we need to choose the right MTU. So I wrote a script to find an optimal MTU.
Ideally I would have liked to have run all possible MTU configurations for both WG Server and WG Peer but for simplicity I choose to fix the WG Server to the original 1420 MTU and tried all MTUs from 1280 to 1500 for the WG Peer.

Testing

On WG server, I started an iperf3 server
On WG peer, I wrote a script that does the following:
- wg-quick down wg0
Edit MTU in the /etc/wireguard/wg0.conf file

Ubuntu: Install Kernel and Set GRUB Default Kernel

Install Kernel

Install the default kernel:

sudo apt install linux-generic

Set GRUB Default Kernel

Find entrance from /boot/grub/grub.cfg
- Get the $menuentry_id_option:

	// Command iossize predicts the impact
	// on memory usage of changes to the binary.
	//
	// This implementation uses dyldinfo to determine
	// exactly how much memory the dynamic linker will dirty
	// as part of launching the executable.
	//
	// It is not perfect. There are multiple sources of dirty pages:
	//
	// 1. dynamic loader rebase/relocs due to our code

	// This can grow a Robin Hood linear probing hash table near word-at-a-time memcpy speeds. If you're confused why I use 'keys'
	// to describe the hash values, it's because my favorite perspective on Robin Hood (which I learned from Paul Khuong)
	// is that it's just a sorted gap array which is MSB bucketed and insertion sorted per chain:
	// https://pvk.ca/Blog/2019/09/29/a-couple-of-probabilistic-worst-case-bounds-for-robin-hood-linear-probing/
	// The more widely known "max displacement" picture of Robin Hood hashing also has strengths since the max displacement
	// can be stored very compactly. You can see a micro-optimized example of that here for small tables where the max displacement
	// can fit in 4 bits: Sub-nanosecond Searches Using Vector Instructions, https://www.youtube.com/watch?v=paxIkKBzqBU
	void grow(Table *table) {
	u64 exp = 64 - table->shift;
	// We grow the table downward in place by a factor of 2 (not counting the overflow area at table->end).

	// RCL 05 June 2021

	/*
	verify with `openssl pkey -in <privatekey>` or `openssl pkey -in <privatekey> -pubout`
	the latter should match the publickey
	*/

	package main

	import (

	/*
	* m1racle-poc: a basic proof of concept for the M1RACLES vulnerability in the Apple M1.
	*
	* This program allows you to read and write the state of the s3_5_c15_c10_1 CPU register.
	*
	* Please visit m1racles.com for more information.
	*
	* Licensed under the MIT license.
	*/