ChatGPT let loose on the web and DALL•E 3 gets weird

ChatGPT let loose on the web and DALL•E 3 gets weird

Welcome to the second AI Logs weekly newsletter from me, Cory, and the amazing team at Interesting Engineering, where we discuss and highlight new AI tools, nerd out on how to use them, and more. 

After starting companies, raising funding for several, scaling them and now advising as an EIR at the Founder Institute and more, I’ve learned that the most important component of any successful venture is listening to feedback and iterating upon it

To that end, we welcome all feedback, insights, and thoughts on AI Logs as we dive into the second episode. Look out for the Give Feedback link at the bottom of this email.I hope you benefit from and enjoy it as much as I do putting it together for you.

News

🌐 ChatGPT unleashed: AI bot now has full Internet access

OpenAI has removed the limits on its viral chatbot ChatGPT, granting it unfettered access to the internet for the first time. Previously constrained to data through 2021, ChatGPT can now browse the web and provide up-to-date information for users on its Pro paid plan.

The new “Browsing” feature allows ChatGPT to search the internet via Microsoft's Bing service. Released from its 2021 dataset, the bot can now answer questions on 2022 and 2023 events, allowing it to sustain more knowledgeable and relevant conversations.

While exciting for users, this expansion also raises concerns about copyright infringement. ChatGPT has already faced lawsuits over training on authors' works without permission. Its deep integration with the web creates more opportunities for problematic content generation.

Still, this new freedom of access secures ChatGPT's position in the vanguard of conversational AI.

Other Important Updates

Prompt of The Week

This week’s “Prompt of the week” is less of an actual prompt, and more of a prompt level-up. If you’re using ChatGPT Pro and NOT using “Custom instructions”, I highly recommend checking it out. 

This function allows you to supply as much context as you need, and better define what you’re using ChatGPT to accomplish. Importantly, you can also tell it how it should respond. 

This is where I recommend placing the “Professor Synapse” super-prompt I provided here last week. To access Custom instructions, simply click the three dots “” next to your name while logged into your ChatGPT account. 

DALL•E 3 can get very weird if you ask it to

Yesterday, Alvaro Cintras demonstrated just how weird things can get on DALL•E 3 in an eyecatching thread on X. He defined a new para meter, simply called “weird,” and used it to warp simple prompts into amusing, intriguing, and downright terrifying pictures. I'm going to be very wary of watermelons from now on.

TUTORIAL

DALL·E 3 Art Generator

To use DALL·E 3 visual creator you will need:

  1. Paid access to a ChatGPT Pro account

  2. To enable plugins

  3. To select the DALL•E3 option from the GPT4 dropdown menu

(As a side note GPT4 has just got way faster at delivering results, which is great to see.)Now, just prompt it with as much detailed information as you can, with the expected style, feeling, and nuances. Keep iterating on the results until you’re satisfied. Here’s an example of a prompt I used and the result:

Please act as an iconic sports photographer who’s had your work on the front cover of the most famous magazines. Please create a digital rendition on one of your works with a hyper realistic cartoon rabbit is a basketball player smal dunking a reverse dunk with a reflection of its ear and eye in the backboard, the crowd going wild of people and different animals, and the jersey looks like a championship winning team but somehow alludes to artificial intelligence and also Coryconnects on it.

You can see the result (one of the 4 options DALL·E 3 generated) above.

You’ll notice that there’s no reflection and the crowd doesn’t appear to have animals in it, but the words are clear, and it’s fairly iconic and unique. Not bad for a first pass using generative AI!

Bonus/extras to take this tutorial two steps further

First, when prompting ChatGPT you can tell it not to keep asking before keeping on going, effectively bypassing that sometimes burdensome step. 

Secondly, once you’ve created an image in DALL•E3 in chatgpt or Dall•S2 on their site (also a OpenAI product like chatgpt is) or Midjourney (on Discord), Eluna, AI Art or others) you can upload the image to RunwayML (a tool we featured last week) and have it animated. 

Finally, you can then call on an AI called Topaz to make those videos even smoother, more realistic, and so on. I'll share more on that in future issues, but for now the AI image generator is awesome, animating them is too. 

I hope you enjoyed this tutorial. I’m considering doing some video ones in future newsletters too — please let us know if that would be beneficial.

TOOLS OF THE WEEK

📝 MemGPT (Memory GPT) is a new program that you will need to install it onto your computer. You'll need to know a little about coding to deploy it successfully, but this is worth picking up, as the implications are massive. MemGPT is a huge step for AI: it can actually remember things and so is always getting better, smarter and more tailored to the user.  

🧠 Brainner is an awesome tool that uses AI trained on your inputs or data to create your own smaller model AIs that are unique to you. Basically, it lets you build your own version of a ChatGPT trained by you on your expertise to deliver outcomes that you or potential customers of your AI would want.  It’s very easy to use. To provide a working example of a use-case, I trained mine to write a great LinkedIn headline and “About” section to optimize a profile there. The results are pretty awesome already! I highly recommend Brainner.

💬 Pi is an AI chatbot that is distinguished by its emotional intelligence and commitment to privacy. This makes it more of a conversational AI chatbox than a strictly functional one — it’s really easy to just talk to, and seems to want to be like a counselor or therapist as it evolves. It is particularly suitable for people who aren’t familiar with using AI yet. And it’s free! You can just visit Pi’s website or app.

👁 ChatGPTV has been a hot topic since being released several weeks ago, because it gives the chatbot “Vision.” You can upload pictures of anything and get advice, information, etc about it.  For example, say you take a picture of an item of unassembled IKEA furniture — it could tell you how to assemble it. You could give it a picture of a houseplant and learn how to care for it, or diagnose what might wrong with it. OpenAI has also released the voice feature (“talk and listen”) in the mobile app only. Learn more about it on OpenAI’s blog.

Written by

Cory Warfield

LinkedIn Top Voice/Influencer in AI

what else?

🚨 For IE’s daily engineering, science & tech bulletin, subscribe to  The Blueprint  

🧑🏻‍🔧 For expert advice on engineering careers, subscribe to   Engineer Pros   

🔷 For all the week’s top engineering stories, subscribe to the  Vital Component  

⚙️ To explore the wonders of mechanical engineering, get your  Gears in Motion  

🎬 For a weekly round-up of our best science, tech & engineering videos, subscribe to   IE Originals   

For our weekly premium newsletter and an ad-free experience,   sign up for IE+   

© Copyright 2023 | The Blueprint is by Interesting Engineering,Inc. 530 Fifth Ave, 9th floor New York, NY 10036, USAAll Rights Reserved

You are receiving this email because you have subscribed to our newsletter.Manage your e-mail preferences here.Unsubscribe from our emails here.See our full privacy policy or terms of conditions.