Are You AI Proof?

Hurdling through realms unknown, a frenzied chase for elusive mastery. In a whirlwind of synapses, we tread the enigmatic path to General Intelligence 🤖🕰️

[Are You AI Proof? @AIProofYourJob]

"If you want to build AGI, regardless of how you define it, you need users... and you need humans in the loop" - David Luan @jluan @AdeptAILabs @EricNewcomer

https://t.co/zZ48Olc5J7

[Are You AI Proof? @AIProofYourJob]

"We have entered the industrial age of AI... its now time to build factories" - David Laun @AdeptAILabs @EricNewcomer

https://t.co/VFOYVkhyo3

[Are You AI Proof? @AIProofYourJob]

Great interview by @EricNewcomer with the CEOs of @replit and @huggingface
https://t.co/YAP14JfqsN
"We automate the things that are automatable so that we can go on to creative things" - Amjad Masad @Replit

[Atlas3D @Orwelian84]

can confirm, almost 40 something having been on the web since the early 90s. Shit doesn't just move faster, its like the expansion of the universe. Every point is accelerating away from every other point all at the same time. AI to tear down and generate disinformation is going to super charge all of this. Everyone will want to us AI to prove their pet theory. Going to be great time to be alive again, iroincally it will be those dusty old books and in person meat interactions that the bots can't fake.

[Ali Ghodsi @alighodsi]

Free Dolly! Introducing the first *commercially viable*, open source, instruction-following LLM. Dolly 2.0 is available for commercial applications without having to pay for API access or sharing data with 3rd parties. https://t.co/HtRVSmArFU

[Paul Millerd @p_millerd]

With AI the value of asking good questions increases.

Based on my work training consulting teams and knowledge work teams, people are very bad at this.

I think it’s not even something that can taught as much as experienced and dependent on actual curiosity

[Are You AI Proof? @AIProofYourJob]

Thank you for this. Opens up tons of doors.

[LMQL (Language Model Query Language) @lmqllang]

🚀 Excited to announce the first release of https://t.co/phw1uWlPEF, a novel open source programming language and platform for language model interaction!

Combining prompts, constraints & scripting, LMQL elevates the capabilities of large language models.

🧵1/6 A quick tour.

[Greg Schoeninger @gregschoeninger]

Diffusion Models getting down to 1.7MB 🤯 smaller models mean more and more companies are going to start to own their own their end to end infrastructure.

Paper: https://t.co/9gfwT1iRQy

[Greg Brockman @gdb]

A coworker on GPT-4's mistakes with detailed list of instructions: "it makes the same variety of mistakes that a team of 1,000 different humans would make". That is, it covers the full distribution of reasonable interpretations rather than picking one — a strength and a weakness!

[elvis @omarsar0]

"The hottest new programming language is English"

engshell is an application, powered by LLMs, that allows you to interact with your shell using natural language.

Supports tasks such as recording screen, compressing files, checking the weather, getting recommendations, generating pictures, reading news, doing maths, managing files, and more.

https://t.co/p8Bktah532

[Arena.ai @arena]

We’re releasing Vicuna-7B: small, efficient, yet capable.

💻 MacBook users can simply "pip install fschat" and run Vicuna-7B with GPU acceleration on M1 chips!

code: https://t.co/cPdf1rjaPv
weights: https://t.co/MnNa9zaNUI

[Manish Singh @refsrc]

Anthropic in its Series C deck, seen by TechCrunch: "Within the next ~2 years, we expect to raise ~$3-5b from strategic and/ or institutional investors to stay at the frontier."

[Aran Komatsuzaki @arankomatsuzaki]

Cerebras-GPT: Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster

- Trains Cerebras-GPT Pile following DeepMind Chinchilla scaling rules and µP
- Releases pre-trained models and code

https://t.co/I0k2b7TV45
https://t.co/VQ3KxsrOJT

[AK @_akhaliq]

SegGPT: Segmenting Everything In Context

abs: https://t.co/7O6jyDv0t9
github: https://t.co/ycb37HER46
@Gradio demo: https://t.co/KgmsUJIMTL

[Eric Park @whereisdemusic]

Otto: System using a large language model & vector database for self-governing digital orgs. Great jump-off point to adding more complexity. Credit to the original commit of @yoheinakajima.

Equal agent participation in management.

.https://t.co/ny4AJ2ZAsz

[Joshua Browder @Joshuabrowder]

Here it is! Using the DoNotPay ChatGPT Plugin, I asked the A.I to cancel my gym membership.

It immediately drafted a legal notice to cancel, e-signed it and connected with USPS to mail it, all without leaving the conversation.

The first subscription cancelled using ChatGPT!

[Scott Swingle @biobootloader]

Why Google should hire a million writers:

Chinchilla scaling laws demonstrate that training data, not parameter count, is the bottleneck for LLM performance

Instead of trying to squeeze more high quality data from the web, what if Google just created it?

The math checks out:

[Andrej Karpathy @karpathy]

The analogy between GPTs of today to the CPUs of early days of computing are interesting. GPT is a funny kind of programmable text computer. Have to think through it more 🤔 but e.g.:

## Memory
GPT-4 RAM is ~log2(50K vocab size)*(32K context length)/(8 bits/byte) ~= 64kB, roughly a Commodore64. Just as then, optimizing this precious resource is critical.
GPT registers are the residual stream. There are d_model of them, e.g. GPT-3 has ~12K registers. VLIW architecture vibes.

## CPU
The LOAD instruction is the Attention mechanism, except it can address by both location and/or content.
The STORE instruction is forced every n_layer number of clock cycles.
The ALU are the MLPs + LayerNorms. Awkwardly, as their params are not shared across layers, the ALU changes at each clock cycle. Optionally the MLPs may also be interpreted as supporting a kind of fixed knowledge database lookup.
The programs always takes the form [[LOAD, ALU]*N, STORE]*M, where N is n_layer and M is num_tokens.

## Architecture
GPT feels closer to a fixed-function than stored-program computer because the number of parameters is so large. In contrast, the description length of a CPU is very low and all the action is in the memory configuration.
Another way to look at it is that GPT is a much more bloated/complex computer. Which is fine because it is not engineered but optimized and the upshot is that the programs can be shorter.

[Zain Kahn @heykahn]

If you're not using AI, you're falling behind.

20 AI tools to future-proof yourself:

[Mckay Wrigley @mckaywrigley]

Prompt
—
I don't know how to code, but I want to understand how this works. Explain the following code to me in a way that a non-technical person can understand. Always use Markdown with nice formatting to make it easier to follow. Organize it by sections with headers. Include references to the code as markdown code blocks in each section. The code: