For the past few months, I've become obsessed with the problem of how do you build an AI-powered business
AKA - "how do we actually use this stuff"
But it honestly started because at the beginning of the year, I did hundreds of AI Advising calls with businesses and while I was
I told my husband ChatGPT can write his resident call schedule in <30 minutes.
A task that he said normally takes 100+ hours
Will report back in hopefully <30 minutes
🚨 JUST ANNOUNCED: ChatGPT Plugins 🚨
You'll now be able to:
- Browse the internet
- Book travel
- Buy groceries
- Do math
+ more, as developers can now build and offer plugins for ChatGPT
Here's a "being a nerd pays off" story for ya
I've made $17K so far selling two AI related domains I bought in October/December with keywords nobody was using at the time
I thought I would use them... and bought them for like $12 each
~8000% ROI
🚨 Just announced! 🚨
Meet
@Adobe
Firefly, a generative AI art tool designed to be commercially safe. 🙌
Had fun partnering with them leading up to this launch. Expect to see more cool things from Firefly, coming soon. 👀
#AdobePartner
#AdobeFirefly
I'm technical. But I've never officially had a job with the title "Engineer".
Now with ChatGPT + products like GitHub's Copilot -- I'm building entire projects in weekends.
And it's not just me.
My sister wasn't "technical" 6 months ago.
Now is.
Yes, also enabled by ChatGPT.
Revisiting approach.
I expected GPT4 could do a half-good job giving me code to solve it. Even if it was slow, unoptimized, etc.
But it hasn't.
And its making me rethink the nuances of "GPT4 as a reasoning engine"
Will definitely be writing about this soon.
Tons of headlines are declaring a decrease in ChatGPT usage
Some even venturing to question if generative AI is a fad
I'm spending 80+ hours a week working on this problem. And let me tell you the actual issue --
AI still has a major usability problem.
Most people are still
The most interesting part of
#GoogleIO
so far --
Google will be watermarking all AI-generated content going forward
And it will remain intact, even if the content is edited
Had a really great flight today
Sat next to a doctor. We started chatting, hit it off.
~20 minutes in, she asked what I do (more specific than "tech"). To which I awkwardly said, have you heard of ChatGPT?
To which she said no 😳
I explained it at a high level, alluding to
Runway introduces Gen-2 🪄
"If you can say it, now you can see it"
Text-to-video will change content creation. Communication is about to be multi-modal.
Say goodbye to your weekend.
If you're a
@zapier
fan, they just added
@openai
whisper. 🔥
Transcribe any file. Send to any ChatGPT prompt. Then do anything.
Needed 24 hours to recover & regroup
But WOW was this weekend's
#emergencychatgpthackathon
impressive 💯
250+ people & 70+ teams all "summoned" to hack on the new
@OpenAI
ChatGPT API
Here were 3 big trends across the demos & other takeaways:
What should you use the ChatGPT API for? What are all those settings? How do you get started?
Introducing 2.0 of our Cheat Sheet answering just that 🔥
Our first was a huge hit. Updated & expanded for the latest and greatest from
@OpenAI
👇
Has anyone experienced or studied prompt drift?
When prompts were performing well at one point in time, but due to model changes, degrade in performance?
Seeing this on one of our projects.
Every data scientist needs to steal this ChatGPT workflow...
- Get analysis done 10x faster (at least)
- Make cleaner notebooks without all the work
- Really freakin fun
Here it is:
1. Upload the files you need analyzed
2. Ask ChatGPT to tell you the names of the columns in
The more complex this is getting, the more I'm describing prompting as "lets write it with more structure and be super clear, like you're writing instructions for your mom to follow"
My unpopular opinion is that most LLM use cases don’t need a fine tuned model
They actually need a semantic search system that adds context to the prompt, or can accomplish their goals with a few shot approach instead
Reasons 👇
Updates:
- GPT4 has really struggled with datetime conversions. had to jump in and fix myself.
- Now testing and there's definitely logic flaws (admittedly this is a pretty difficult algorithm, requires backtracking, etc)
And 💩
My recommendation to "learn AI" has become much more nuanced after teaching hundreds of people over the past 18 months --
Most people shouldn't
- Learn automations
- Learn advanced prompting methods
- Learn agent orchestration frameworks
ALL people should
- Learn to think in
There's a resurgence of interest in fine tuning LLMs
I've yet to see a successful public use case where fine tuning > prompting.
But here's where I see fine tuning *mattering*:
First, fine tuning is for teaching an LLM specific tasks or behaviors
Not teaching an LLM new
Extremely interesting stat shared by
@amasad
on Moment of Zen
Didn't believe it, had to go look myself
Yup - Stack overflow's traffic down 11% since ChatGPT launch (from SimilarWeb)
As a former data scientist, it's mindblowing to me how easy it is to create smart workflows that use AI. Even if you have absolutely zero machine learning background.
The biggest barrier isn't skill anymore, it's awareness.
What's possible. How to get started. What are the
We're underestimating how big of a deal "Enterprise ChatGPT" is going to be
I talked to 3 people this week, different companies, now using ChatGPT at work
They are nearly speechless about how useful it is
If you've already had a fairly liberal ChatGPT policy for your
now debugging. trying to use GPT for everything.
random thought: error messages often suck. but even more clear that they suck when you're asking GPT to debug and fix them
It’s hard to not believe LLMs are reasoning engines after playing with AutoGPT.
Haven't played with AutoGPT yet?
Here's the start of it recommending & buying wine for me:
I want the ability to switch between models within on thread.
Let me do all the easy work with GPT-3.5.
Then when it's time for the big guns, switch to GPT-4.
Yesterday, GPT-4 was announced and it's possible this completely changed the game.
Why? It can take in both text and images as prompts.
Here are 12 potential GPT-4 use cases to get your mind racing (THREAD) 👇
Excited to launch
@theaiexchange
- a curated collection of AI products & use cases.
🚀 New listings added daily
🪄 Get a personalized recommendation with our matching quiz
🕺 Save hours of research, 100% free
by myself &
@kdowns_22
Yes AI can generate content at a huge scale. But it lacks is talent. The touch, the eye, the vibe.
I believe a creative with talent will outcompete AI.
But a creative with talent that adopts AI also outcompetes the one who doesn’t.
The next wave of AI tools *actually* worth using are going to lean on LLM's ability to reason, more than their ability to generate.
Especially for tasks where perfect decision-making doesn't matter nearly as much as the speed in which it can be done.
That's going to be a
People misunderstand the point of using ChatGPT to help with content
It's not about never writing anything again
It's about having more time to create something worth reading in the first place
Wow this really took off.
I was doing this to push the bounds of what GPT can actually do for businesses right now.
In case you want to read the takeaways/result when I publish on my newsletter:
(I send free 2x per week)
Elon has made some VERY interesting moves lately
The Tucker Carlson interview was a total softball. There should have been more dots connected...
A little recap 👇
Oct '22: Buys Twitter
Feb '23: Shuts down free Twitter API
Mar '23: Signs pause AI letter & quietly starts his
The AI wars continue.
Google announces generative AI added to Google Docs, Gmail + the rest of Workspace
Timely ahead of Microsoft's Thursday event, which is rumored to cover the 365 Suite.
AI is quickly becoming table-stakes for our productivity stack.
All it's going to take is one product announcement from a well established no code app builder (Bubble, Retool, et al) that says
"Text to App is here"
For everyone to realize how much software engineering has changed.
After spending a big chunk of the weekend playing with all the new ChatGPT plugins, I have a few observations that are shaping what seems to be ahead for the "everything app" (aka ChatGPT)
disclaimer, these are WIP thoughts...
First:
I've found that plugin usage feels more
AI will have a profound impact on what it means to create content.
And of course we're pushing the envelope. AIxBot debuted in our subscriber slack today ✨
Trained on our knowledge base. Your friendly bot to help you adopt AI in your business.
AI is coming to every product near you, and fast.
This update was intended to be monthly, but the list below is all of the new products launched just since March 1st 🤯
Not including startups, rumors or non-product adoption
If there was just one piece of advice I wish more people got re: learning about wine...
Focus on stories and people. Not tasting notes and memorization.
AI will have a profound impact on what it means to create content.
And of course we're pushing the envelope. AIxBot debuted in our subscriber slack today ✨
Trained on our knowledge base. Your friendly bot to help you adopt AI in your business.
If you're not talking to businesses about their AI adoption all the time, this will probably surprise you:
Businesses big and small are concerned about data privacy and IP in an LLM world.
And it's probably the
#1
thing holding back adoption.
More than use cases, clear value,
I recently learned of AI companies that are working on the problem of 'life after death'
Basically building persona-based chatbots from someone's life, add a deep faked video clone...
And you can now talk to your loved one, long after they're gone.
3 clear UX patterns are emerging in text-based AI apps
- Copilots
- Workflows
- Chatbots
Going deeper on how to think about which works best for your use case in Tuesday's newsletter.
If you're building an AI product...
Decide whether you're enabling experts to be more efficient.
Or helping novices unlock a new level of expertise.
Those are two very different products.
Try to do both and you'll probably do neither.
The average american drinks 20 bottles of wine per year.
They aren't memorizing tasting notes. Or paying attention to AVAs.
But they do love a good story.
Tell the 🍷 story.
Last week I used
@Adobe
Firefly to help me create ad concepts 👇
The ease of use and transparent model training in Firefly really changes the game for commercial use. Like running newsletter ads 😉
I've had the same convo 5 times this week.
We need more women at AI events
@mickeyxfriedman
@hackathongoofer
and I are starting a mega-list to better syndicate event invites.
Sign up if you're a woman in AI. Any city. Non-devs welcome!
woke up to this 🤗
This was definitely not what I expected when I started a nerdy TikTok account in November talking about startups and AI
also if you’re not on TikTok, I post to Instagram now too: