What's up? You're reading Inclined AI. Remember, tacos are for Tuesdays. Thursdays are for…I’m not sure, actually.

Here's the hard and soft shell news:

• OpenAI reacts to government regulation talks

• Dolly 2.0 delivers an open-source ChatGPT alternative

• People are using ChatGPT to work multiple jobs

• OpenAI launches a new bug bounty program

Be a sponsor of our newsletter

RULES & REGULATIONS ARE COMING FOR AI TOOLING

Don’t expect the Biden Administration to make a decision tomorrow, but it’s clear that AI is becoming a priority for governments worldwide.

It only took them half a year to catch on, guys.

You’re hearing about this now and not two days ago when Biden first spoke out? Because a response was bound to come out before the end of the week. 

Sure enough, Greg Brockman, the President of OpenAI, informally responded to the news on Twitter.

Brockman acknowledged the need for regulation and welcomed the idea with open arms (and a fat stack of cash for lobbying expenses). But he was clear that pausing to wait for rules is impossible.

Every iteration of GPT gets safer and more aligned with our outlook. However, that progress does not happen in a vacuum.

Rules are essential, and the public needs to know what we can and cannot do with AI going forward, but we won’t discover the complexities of these systems by sitting on our hands. That’s the argument Brockman makes.

If you’re going to bet on how long regulation takes to implement, take the over on a year.

OpenAI will release GPT-5 by the time the government enacts a policy, and tech's influence on Congress will not help.

The United States didn’t reign in social media giants when they had the chance, and that’s the track record we’re going off of.

We all want clarity and reasonable restrictions to prevent misuse. It’s simply a matter of making it happen.

A WORD FROM OUR SPONSOR:

How to Get AI-Powered Answers from any Document:

You have a pile of PDFs to read but don’t have the time or patience to go through them all. You need answers fast but don’t want to waste hours reading everything.

What if there was a way to get answers from your documents in seconds without reading every word yourself?

Introducing Docalysis, the AI-powered tool that can answer your questions from any document. Just upload your document, type in your question, and get an instant answer—no more reading, no more searching, and no more guessing.

Don’t let your documents overwhelm you. Instead, let Docalysis do the work for you.

Try it out today at Docalysis.com!

DOLLY 2.0 AIMS TO BE THE FIRST TRULY OPEN LARGE LANGUAGE MODEL

The problem with Alpaca, LLaMA, and other open-source LLMs is that they’re not “open” commercially. The dataset used to train those models is from OpenAI.

Using them for commercial purposes is against their terms of service.

Databricks, a startup firm focused on data analytics, ran into this issue with Dolly 1.0 (which came out two weeks ago 😯).

To solve it, they had employees volunteer to create 15,000 fine-tuning records for a commercially viable dataset. The end result is an LLM which any person or company can use to generate text responses.

That’s cool, but does it work well?

Uh, that depends on what you’re using it for; since the model is not as robust as GPT-3.5, it tends to slip up more often.

If you want to use it for your company, you need excellent prompt engineering to prevent it from going off the rails. Otherwise, Dolly is prone to hallucinations and inappropriate outbursts.

The reason is that the public model they used to train the data—GPT-J-6B—scraped tons of offensive data that researchers never took out.

The lesson: open-source work is imperative to improving AI models, but they also open us all up to more cases of misuse.

You can regulate and give all the companies rules, but personal chatbots made from these open-source AI tools are much more challenging to track. 

After all, it’s hard to say with certainty what humans will do with chat AI when they know no one is watching.

Quick Nuggets

🐛 Bug bounties are launching on OpenAI, get after it

🎙️ Your favorite podcaster is getting their voice cloned by AI

🐈‍⬛ Memory misinformation: this research paper sees how AI reacts to false memories

🇨🇳 China drafts new rules for AI to reflect CCP’s ideals

🤬 Making ChatGPT offensive is a lot easier than you might think

🎮 Video game fans feel like AI is making their experiences hollow

🧑‍💻 Overemployed hustlers are grifting their jobs onto ChatGPT

🇮🇹 Italy might release ChatGPT back to the public by April if OpenAI complies

💽 Raspberry Pi is adding AI tech with the help of Sony

⬆️ 4000% bump: AI-powered apps boost user engagement by massive amounts

💥 ChaosGPT will not destroy humanity. It reasoned that controlling us is better

🧬 AI agents can help replace work done for scientific research

🤧 Ditching work for Tay Swift? AI is going to catch you slipping

🚫 Scammers fake a kidnapping using an AI voice clone of their daughter

🎵 Record labels want to stop people from scraping songs for AI cloning

💧 Gallons of water are used in trying to train AI models

📷 Generative AI images are getting so good that AI struggles to catch them

🔥 Fresh Products

• AI Scam Detective - you won’t get fooled again (link)

• Open Assistant - added an update to their conversational AI (link)

• Yeager Agent - A GitHub guide to run your own AI Agent (link)

• ChatGPT for Gmail - use this extension to autogenerate emails (link)

• AI Code Mentor - learn to write code with an AI coach (link)

• These Abs Do Not Exist - and that’s okay (link)

• ShopMate - upgrade your e-commerce store with AI’s help (link)

• Brainlox - an AI-based learn-to-code platform (link)

• Dropchat - upload an e-book or pdf and then chat with the information (link)

• TalkBerry - use your voice to talk with AI (link)

• Prompt Randomizer - to add some chaos into your typical prompt work (link)

• HyperWrite - adds a new AI Agent to surf the web through their extension (link)

Good Content, Pasta Farmer

Would you glean the gluten from these golden fields?

_{That’s it for today. I hope you enjoyed the latest edition of inclined.ai - Davis.}

Check Out Our Sunday Edition

Every week you can dive into more AI news and topics with us by subscribing to our premium edition.

We’ve written about:

• Here Is The Work AI Will Replace Soon

• Get to Know ChatGPT Through Interactive Games

• The Difficulties in AI Alignment

If you’re not already subscribed, that’s okay. We’re offering a free 7-day trial so you can read this one. That’s how excited I am to post it.

Claim the offer below to support our writing 🙂