Good morning! This is The Neuron.
The deal is simple: We're your second brain about AI. We mind-meld for 3 minutes every morning. You go and rule the day.
Today in AI:
- Google's Big AI Reveal Coming In May
- ChatGPT Pro Launches at $42/Month
- What You Should Know About "Constitutional AI"
- Around the Horn
- Leo Sends His Regards
Google's Big AI Reveal Coming In May
They're dialing it up.
Google's been in the spotlight ever since ChatGPT launched. With Microsoft's slough of industry-moving announcements, people have been eyeing Google, asking, "So...what's up with you?"
It's happening. The New York Times reported that the annual Google conference running in May 2023 will be jam-packed with AI products. Executives have been scrambling to accelerate and prep their products for launch.
These include:
- AI image generators (competing with DALL-E and Stable Diffusion)
- "Create your own AI tools" platform
- Code completion tool powered by PaLM (competing with GitHub Copilot and GPT)
...alongside other image/video projects not necessarily driven by AI.
Google's ChatGPT competitor for search is expected sometime this year, though it's unclear if that will come during May's conference.
Feeling dangerous. Google's products currently lag OpenAI on safety-related metrics relating to hateful, toxic or violent content. That's not stopping launch. Instead, they're pivoting from preventing issues to merely curbing them.
OpenAI may not kill Google, but it's certainly forcing Google to wake up.
ChatGPT Pro Launches at $42/Month
Take my credit card, Sam. Just take it.
10 days after they told us they wanted to launch a Pro plan, OpenAI has turned it on at $42/month.
What you get on Pro:
- Availability even when demand is high
- Faster response speed
- Priority access to new features
The upgrade option is coming in waves, so you might not see it in your sidebar yet. Early adopters have confirmed that ChatGPT Pro feels significantly faster.
At $42/month, ChatGPT Pro needs 200,000 subscribers to hit $100 million ARR.
ChatGPT has millions of users, leaving some wondering if 200,000 upgrades could happen in a matter of weeks or months. That would make it the fastest software product to go from $1 million to $100 million ARR.
It's gonna need that cash and more. OpenAI projections said it lost $500 million+ in 2022 alone.
What You Should Know About "Constitutional AI"
Ready to sound smart at work?
OpenAI faced controversy last week over how they developed training data for ChatGPT.
That left some concerned.
Did this mean that we'd always need some people to read traumatic material as part of training the AI to not say those things, no matter how impactful AI could be or how "worth it" the process is?
We asked some folks at the Scale AI hackathon this past weekend. One of them pointed us to Claude, the ChatGPT competitor being built by Anthropic.
Anthropic is training Claude to not say bad stuff using something called Constitutional AI:
- write down a list of principles
- ask the AI to answer a prompt multiple times
- ask the AI to choose the answer that best reflects the principles
An analogy: You want a child to learn not to swear. OpenAI's approach was like giving the child a list of swear words and saying, "Don't say these."
Anthropic's Constitutional AI is like telling the child, "It's important to be polite." And when they swear, you ask them, "Was that the most polite way to say that?"
Why this is important: Besides reducing the mental burden that OpenAI subjected its workers to, writing down our values and asking the AI to self-reflect is oddly close to how Sam Altman suggests we keep our AI overlords in check when they get here.
Around the Horn
- Can't code? Agent allows you to build apps using AI. Just type in what you want.
- InstructPix2Pix: Edit images using instructions like "wear a suit" or "make him wear VR glasses"
- Extrapolate shows you what you'll look like when you're older.
- NVIDIA Eye Contact: AI helps you appear engaged while multi-tasking.
- We had a blast at the Scale AI hackathon this weekend. Here are some of the projects that people worked on.
- AI in healthcare will necessarily take longer because of things like data privacy, infrastructure and how fast healthcare moves to adopt new tech. Still Microsoft Azure OpenAI is now HIPAA-compliant for anyone exploring.
- The AI art controversy is extending into the manga community.
DM me links on Twitter: @nonmayorpete
Are you new to all this AI stuff? Here's The 3-Minute Guide to Slaying Your Dinner Convo About AI to get you up to speed. Or at least smart enough to impress your family.
Leo Sends His Regards
