Did this newsletter write itself? [from the archives]
spoiler alert: it did not.
welcome to day to data.
Today, I’m sharing a post from my vault of articles (a much, much smaller vault than Taylor Swift) about artificial intelligence. Funnily enough, I originally drafted the article in November 2022, prior to but the same week of ChatGPT’s release. As I edited the article, the dialogue and insanity around artificial intelligence just exploded.
I’m sharing this post for a few reasons. It was one of my first posts on Day to Data and one of my favorites to write. It also tells the story of artificial intelligence, from the decades old history, a highlight reel of events, and criticisms of the field — all of which are critical areas to bolster one’s foundational understanding of today’s landscape.
This article has not been updated with present day knowledge, but I plan to share more articles on artificial intelligence and large language models in the future with my up-to-date take on the space. Stay tuned!
From Day to Data on January 16, 2022 …
Today’s we are talking about artificial intelligence, from the historical roots and some notable innovation in the space.
For your info: I actually drafted this newsletter prior to ChatGPT3’s release on November 30, 2022. While it is a notable breakthrough, I won’t be covering the product fully as there is a lot of really great articles out there already talking about the amazing things the product can and can’t do.
let’s go over some vocabulary
Training data is all of the data we plan to give a computer that we want it to learn from. An example is if we took a photo of all the clothes in our closet and had our computer use those photos as a dataset.
Machine learning is giving computers training data and allowing them to learn from the data through patterns and trends. An example is using the photos of our clothes to find which styles, colors or fits of clothing we own the most of.
A machine learning model is a program that learned from the training data and can now apply what it learned on a new piece of data. An example is using the data from our closet, we can show our model a new piece of clothing and it can predict if it is a piece of clothing we’d be interested in or not based on what we already own.
💡 Machine learning is one method to artificial intelligence.In fewer words…
When we talk about artificial intelligence, we are discussing how massively large training data sets are learning very complex patterns and then able to apply that to new data to an extent that is almost human-like. Some may say we’ve reached human, beyond just human-like. More on that below.
AI is nothing new.
In 1950, the famous mathematician and supposed father of artificial intelligence Alan Turing posed the question, “Can machines think?”. He followed this with the creation of the Turing Test (first known as “the Imitation Game”) which was a proposal of how to determine the validity of his original question. The simulation included a person, machine, and interrogator that play a “game” of sorts trying to see if the interrogator can pick between person and machine. It’s deeply fascinating to try to conceptualize living in the 1950s and being as forward thinking about the future of machinery as Turing.
a short highlight reel of innovation in the AI space:
Here’s some of the most notable moments in the timeline of AI (and I am definitely missing a lot).
1955: The term “artificial intelligence” was coined in a paper that outlined ideas about automation, computers learning languages, and more.
1965: ELIZA became the first chatbot built by MIT that used “pattern matching” to simulate therapy with users.
2000: Also at MIT, Kismet was built as an “expressive robotic creature” that was “tailored to natural human communication channels”.
2011: IBM’s Watson competed on Jeopardy, and beat the legends Ken Jenning and Brad Rutter.
2020: OpenAI GPT-3 is released as a powerful text generation language model (not to be confused with ChatGPT-3, later released in 2022).
2021: Google announces LaMDA — Language Model for Dialogue Applications — in an attempt to tackle conversational skills in the language modeling space
with every innovation comes criticism to be aware of:
AI ethicists Margaret Mitchell and Timnit Gebru co-authored a paper on the dangers of extra large language models, and their ability to mislead and/or perpetuate bias. Both also were fired from Google for apparently violating company code of conduct while a part of their Ethical AI team.
The name “Greg Rutkowski”, a Polish digital artist behind designs in Dungeons & Dragons, Forbidden West and more, has become one of the most commonly used prompts on Stability AI’s platform (90K+ over Picasso’s 2K+), generating art that takes from Greg’s classic style without his permission or attribution.
Google fired an engineer, Blake Lemoine, who claimed their language model, LaMDA, became sentient, who felt the platform possessed feelings and emotions beyond the practical intended nature, demanding more oversight by Google into ethical use of a potentially sentient program.
the true development is in how consumer-friendly AI applications are becoming.
Gitlab’s CoPilot allows programmers to type a phrase then have it written in code in seconds, like a brilliant colleague is sitting next to you writing code.
Notion announced Notion AI, an assistant who can “write, brainstorm, edit, summarize and more”. I’m typing this in Notion right now - one day maybe AI will do it for me.
Startup Alert ⚠️
Tome is carving out a new space for generative storytelling, allowing users to give a prompt, a company/product description or more and get a full pitch deck or other story spilled out in deck form in no time.
https://www.fastcompany.com/90827339/presentation-tool-tome-launches-ai-to-help-make-storytelling-simpler
As I mentioned earlier, ChatGPT-3, the internet-breaking-fastest-to-1MM-users platform, was released on November 30 by OpenAI. It’s being praised (and feared) by people everywhere, for the odd and incredible and scary things it can do. The dialogue based chat platform has a feature new to the AI space that I find to be most interesting of all - persistence. If you begin a conversation with ChatGPT-3 about how to write a letter to your mom for her birthday, than ask the platform to change the tone, add some new info, or alter the letter they gave you the first time, it’ll remember. ChatGPT-3 persists across states, rather than simply getting a prompt, returning an answer and starting that over again, it’s able to remember and build based off past interactions — just like humans. Pretty cool stuff. I can confidently disclose none of this newsletter was written by, or inspired by, ChatGPT-3 output.
What’s exciting you in the artificial intelligence space? What’s still unclear? How do you want AI to change your day to day life? Or are you scared of its impacts all together?
Several images in this article were generated using StabilityAI’s Stable Diffusion Demo on Hugging Face.