Jacob's A.I. Journal is a collection of articles on A.I. topics from Impower.ai's Principal ML/AI Architect, Jacob Haning
A.I. in Business
Some companies are still trying to make self-driving cars work—Uber and Motional launch robotaxi service in Las Vegas.
They continue the slow and disappointing uphill battle of making self-driving cars a viable business. Ford and VW shut down their joint-backed venture Argo AI in 2018, and Uber shut down its Advanced Technologies Group years ago after one of its test cars struck and killed a pedestrian. Companies across the sector are missing expectations and timelines and have yet to turn any profit. If you've been in my life over the past five years, you've heard my opinion on self-driving cars (sorry to my family). When I say self-driving, I mean level 5 of the SAE standard (explainer video). Assuming the technology was ready, it's not, and people wanted them, they don't, our existing legal and economic systems are entirely inadequate in their current form to tackle the challenges this technology presents. Outlook: I'll counter this by saying that a functioning autonomous system is likely the safest and most effective form of individual travel. With investors pouring billions into the market, we will likely move closer to that reality in my lifetime. But I will also echo the words of Ford CEO Jim Farley in saying that we are "a long way off."
A.I. in Art and Science
Why do I keep hearing about ChatGPT? Is it hype or a massive shift?
This MIT Technology Review article is a longer read, but does a great job of chronicling just how much has happened with A.I. in a matter of months.
If you've been following along with this newsletter, you've seen DALLE-2, MakeAVideo, or GPT3 described as interesting tools or, in some cases, nuisances to the human laborers in those fields. These are generative models, meaning they generate content. DALLE-2 creates images, MakeAVideo creates short animations, GPT3 generates text and is the basis for the chatbot ChatGPT. These products have registered millions of users in record time and have altered workforces, industries, and educational institutions.
All indications point to a massive shift. The general public has responded to this product at a previously unseen pace. The currently free price tag is forcing the world's largest tech companies to take notice. Microsoft has announced massive investments in the technology to the tune of billions, while Google has adjusted its strategy declaring a "Code Red." Schools at every level are considering their options for the road ahead, and marketing professionals have a new secret weapon for writing copy.
Why is it different?
Several people have asked me this question in the last month or so and have had different answers depending on context. Here is a quick summary:
Math: Not markedly different. There are no new theorems being used here.
Architecture: The model architecture is similar to other large language models but scaled. ChatGPT has 175 billion parameters. It was trained on the internet including conversations.
Interface: Chatbot text interface
Context: Seemingly unlimited. I have yet to find a topic that ChatGPT fails to respond to.
This tool is not fundamentally different from others we've seen, but it is much larger, uses a familiar chat interface, and can contribute to a wide range of topics.
A word of caution: While immensely helpful, there are some drawbacks to using these tools. Built to mimic human output, generative models rarely have any mechanism to verify facts. If you're using a text generation tool, you should always validate numbers, dates, and proper nouns due to the high likelihood that these are incorrect. Also, you should always advocate for your personal data privacy. Many of these tools collect the prompts for analytics, so be careful what you write.
A.I. in Healthcare
Top 5 AI use cases in healthcare and its limitations in 2022
AI has massive potential in the healthcare field. There are plenty of examples of AI and machine learning causing harm. Still, in healthcare, we have a real opportunity for good. From improving the patient experience to reducing costs and assisting in diagnoses and treatment options, technology could help all of us. For example, Google Research and DeepMind have launched an AI-based healthcare platform called MedPaLM to assist physicians in answering patient-related questions. Here are some of the other top use cases in healthcare:
Improve diagnostic accuracy - Spot fractures, classify brain tumors
Transform patient experience - AI-powered assistants
Automate repetitive tasks - Doctors spend 16 minutes per patient entering EHR info
Facilitate clinical trials - Design, Recruit, and Monitor
Catch prescription errors
Random
AI can now make us immortal but should it?
[Try it out] tetris themed word association game where AI is the making the associations
[Try it out] Haryy Potter style moving photos.
Totally Random
This twitter account collates Medieval art's strangest creatures
These comedy wildlife photos are a good time
These renowned geniuses also had time for hobbies
Happy birthday to Earth's oldest living land animal