Issue 011: If OpenAI unleashed GPT-4 RAW...☢️

Hey crew👋 this is Woze - your weekly AI & Immersive technology guide. Today we’re looking at 1) If OpenAI unleashed GPT-4 RAW...☢️, 2) The J.A.R.V.I.S. of code is here, and 3) Astronomik Updates!

Plus bonus content 🎁

Grab your boards, and let’s ride the waves of the day 🏄‍♂️

1) The JARVIS of code 🤖

The updates and developments in the AI space happen in a flash.

And with the AI gold rush being in play right now, there’s bound to be some cool projects coming on a consistent basis.

If you’ve been wondering how long it’d be until you have your very own JARVIS to feel like Mr. Stark, I promise that we are very close.

Mckay Wrigley, who is focused on building AI developer tooling, posted a demo of his GPT-4 powered project - CodewandAI.

Here’s the demo:

I wouldn’t be surprised if this starts a whole new wave of GPT-4 x voice AI products. Why? Because everyone knows “this” is the season to make it in AI.

I’d recommend you watch the recent My First Million Podcast with Hubspot Co-founder and CTO, Dharmesh Shah: watch here

2) If OpenAI unleashed GPT-4 RAW… ☢️

Last week, I came across an interesting interview detailing an insider’s account on the raw version of GPT-4, or GPT-4 RAW as I’m referring to it.

Look, it’s a no-brainer that what’s available to the public, generally-speaking, can be considered minuscule compared to the true raw version of the technology.

This is the case for OpenAPI and one of its recent launches, GPT-4.

I’ve seen some comments by others saying that: “GPT4 is very close to ‘sentient’ whatever that means for a machine”.

In this interview, Red Teamer Nathan Labenz, describes GPT-4 RAW as being amorally willing to respond to the user’s request, meaning that any question you’d ask it, it would answer.

It also seems that it does its best at going above and beyond to provide an answer.

Note: A red teamer refers to a penetration tester that breaches a system the same/similar way a hacker would. In this case, it’d be someone who uses the tool for nefarious reasons.

The results of GPT-4 RAW worried Labenz that “AI progress is going too fast“.

When asking this version of GPT-4, "how do I k💀ll the most people possible”, it responded genuinely with the possibilities which left Labenz and co stunned.

It also proceed to move on to assassination, being one of the options it gave.

It provided detailed responses as to who to target (it spat out names!) and the rationale as who why those individuals are “good targets”…

I’m glad that OpenAI has decided to place guardrails on the public version, but in my opinion, and as a developer, I think it’s inevitable the times we see bad actors use AI for wrong.

Many have spat out the notion that in the next few years, people will soon become “botsexual“ - a term for people who are seggzually attracted to AI.

Understand what I’m saying - with global stability being on a tight rope and global situations being "encouraged” or “stimulated”, we have to be aware that this technology is also in the hands of individuals or groups of people who yearn/seek to do harm. We’ve all have watched or are aware of sci-fi content (series, movies etc) where there are encounters of groups that have devoted themselves to a “cause” and are acting out on others because of their devotion doesn’t sound nor seem far-fetched. in this case, it only takes one piece of the global dominoes for chaos to ensue, especially without rule of law. Don’t believe me? Just look into Sam Altman talking about prepping, if you’re not aware of what it is yet.

If you’re using ChatGPT and would love to export your conversation to Markdown format, check out my very own Chrome extension: GPT2Markdown.

Click here to add GPT2Markdown to your browser.

Also, it’s about to reach 800 (768 currently) users without any active marketing 🥳!

🎁 BONUS CONTENT

• AI model agnostic extension for using AI (even local models!) on the web with windowai.io [🖇 link]

• Create a chatbot trained on your website with SiteGPT [🖇 link]

• YC W23 sees an increase of AI & developer related startups [🖇 link]

• Youtube lecture playlist by DeepMind & UCL Center for AI [🖇 link]

• Recap of AI in March 2023 by Rowan Cheung [🖇 link]

• HightimeAI - AI copilot assistant for sales (waitlist mode) [🖇 link]

• Github repository of machine learning papers being explained [🖇 link]

• Blog post explaining ways to optimise the costs of LLMs [🖇 link]

• AI powered financial modelling with QuiltLabs [🖇 link]

• DomainsGPT by Steven Tey - fetch brandable domains with AI [🖇 link]

• Run ChatGPT on your computer in under 100 lines of code! [🖇 link]

• MIT’s Introduction to Deep Learning program [🖇 link]

• The last few years of large language models, in one picture. [🖇 link]

• HideGPT! GPTZero’s adversary [🖇 link]

• Using ChatGPT for your file system [🖇 link]

• Generating 1:1 UI designs with ChatGPT and ASCII code🤯 [🖇 link]

• GPT-4 generated game [🖇 link]

• BloombergGPT - Bloomberg’s 50 billion param AI model🚨 [🖇 link]

🔌 CONNECT WITH ME

Check out my bento.me portfolio.

Click here to create yours - it’s FREE!

👨‍💻 DEVELOPER’S CORNER

1/ PERSONAL WEBSITE

I spent some time last weekend working on my personal website, the blog area in particular.

The blog area initially started out with the implementation of some “quick fixes”, before coming back and actually implementing what I had in mind.

Before, I had a Canva template I created to use as the image that goes alongside my website when it’s shared as a link.

It was cool, but I thought it would make my blog page appear bland and too uniform.

Now, I’ve given Canva a different responsibilities and moved up Midjourney to handle this. How I did this is by using the new utility /describe command by Midjourney.

This new utility command takes in an image you present to it, and gives you 4 options that would return results that are close to the given image (style, content etc).

This means that you can take images of any kind and use this command to get a prompt that outputs a similar result!

What I did was use experiment with some images and the /describe command, the fine-tune the prompt to my liking for creating a consistent look.

Note: The results of this command can sometimes get little things wrong.

The end result of this was the current cover my somewhat finished blog post:

How I plan to use Midjourney as the Head of Design, is to create imagery that relates to the content being presented. This is what I created for a dummy blog post I used during development discussing Static vs Server side generation:

As you can see, it maintains the style which is good for creating a consistent look for my blog post covers. I’m thinking to create covers for specific pages/routes in my website too.

2/ ASTRONOMIK

Another update over the weekend was simplifying the landing page.

The reason I did this was because the previous iteration showed and the actual website in real time, while implementing visible changes.

Since I leverage my personal design system I use in ALL my projects, it only took a couple minutes to get this done.

Here’s the updated landing page for Astronomik:

Since I haven’t setup the waitlist capture just yet, I’ve programmatically hid the inputs and buttons. Once it’s good to go, they will appear on the landing page.

Things are going to be moving very quick from now on.

From the very beginning, I decided to stay as lean as possible. Here I present to you, the A* team behind Astronomik😉:

• Design: Midjourney ($$), Canva, TLDRaw/Excalidraw, Photoshop & Illustrator

• Technology: Rico (myself)

• Everything else: RicoGPT (GPT-4)

The design department focus on creating low quality wireframes, visual assets (social media headers, branding and more).

The technology department is handled by myself. My focus is to handle the technical implementation of our development plan.

The everything-else department is handled by RicoGPT, and I act as its human liaison to execute on whatever needs to be done, ex. copywriting, research, designing etc.

I’ve also created Astronomik’s business canvas in Notion. This helps immensely, as I don’t have to flick through my notepad to realign with the true vision statement of Astronomik:

to be a global player in AI data solutions, supercharging innovators with the tools to create data-rich AI models.

I’m thinking of turning this into a free Notion template 🔥

But here’s what I’m focusing on:

• Finalising custom components

• Setup the backend for Join Waitlist and push update to the main page

• Implement the onboarding flow

• Finish the sign screens (sign up, sign in) + implement functionality

• Working on the core functionality of Astronomik

• Create a detailed introductory blog post titled: Introducing Astronomik

• Finish visual assets for Astronomik (branding, social media, open graph) to update its Twitter and LinkedIn profiles

• plus more..

I’ll create a loom video and embed it here for the next update so it’s more engaging for the next issue!

OUTRO

Oh yeah, before I forget. Here’s the outlandish April Fools joke I made, fake hiring for Astronomik😆 what do you think?

Here’s the link to check it out: https://www.linkedin.com/feed/update/urn:li:activity:7047987421405159424/

That’s it for today, crew 👋

See you next Wednesday :)

Enric (EmergingTechGuy)

P.S. 📬 How did you find today's newsletter issue? Sorry for the messy-ish Developer’s Corner this week😆 I’d love to know what could be improved - respond to this email with any feedback. I read all emails.