The AI Stack, explained
The Technical Underbelly
🔥Enough to be Dangerous
Talk tech, don’t code it, so you can talk about all kinds of other stuff.
The Technical Underbelly
Bite-sized technical web, SaaS, or cloud concepts
Artificial Intelligence (AI)
Artificial Intelligence (AI) is the science of creating computer systems that can perform tasks that typically require human intelligence, such as decision-making, problem-solving, and learning.
Applied AI refers to the application of AI to solve real world problems. It’s where AI comes out of the lab and into the real world - autonomous vehicles, facial recognition, bots, and disease diagnoses.
Machine learning, natural language processing, and generative AI are all subsets of AI.
Machine learning is the process of feeding computers tons of data, and training algorithms to detect patterns in the data. This enables them to make decisions without being explicitly programmed to do so.
NLP deals with understanding and working with human language and is central to machine translation like Google Translate and DeepL.
Generative AI is the current hype, and it focuses on algorithms that create original content from minimal input: text to image, image to text, the creation of sounds, music, stories and … deep fakes.
Elements of AI : If you’re looking to improve your AI fluency, I recommend this very aesthetically pleasing course from MinnaLearn + U. Helsinki.
AI: Key Takeaways
AI is one of the most awe-inspiring, god-fearing technologies we hath yet wrought.
It feels magical and fantastical even when we catch it making sh*t up or ignorant to anything that happened prior to 2021 that wasn’t logged to the internet.
AI is not just software, it’s not just algorithms and code.
a16z divides the generative AI tech stack into three layers. I think it’s a good framework with which to look at what is involved in AI systems in general.
The generative AI tech stack
“Applications that integrate generative AI models into a user-facing product, either running their own model pipelines (“end-to-end apps”) or relying on a third-party API
Models that power AI products, made available either as proprietary APIs or as open-source checkpoints (which, in turn, require a hosting solution)
Infrastructure vendors (i.e. cloud platforms and hardware manufacturers) that run training and inference workloads for generative AI models”
Microsoft Azure’s cloud infrastructure powers every query on ChatGPT. Microsoft’s cloud is a collection of servers in physical data centers here on earth hooked up to networking hardware, cooling systems, power systems, etc.
CPU are the main processors (chips) critical to the running of a computer. GPU are specialized for enhanced graphics in gaming, while TPU (T for Tensor) are specially designed by Google for AI neural networking machine learning.
Human beings are involved in training certain data sets. According to a recent Time investigation, Kenyan data labelers were paid a wage between $1.32 and $2 per hour in this effort to make OpenAI safer for the public. Please read the content warning at the top of the article as it is very distressing.
In Case You Missed It
Last week’s Shortlisted: The Great A.I. Awakening from New York Times Magazine
That’s all for now. Thanks for reading!
Up next week, it’s Stay Savvy - curated links and commentary at the intersection of tech, business, and culture.
Alice Egan, Founder & Educator, SaaS Savvy.
I teach B2B SaaS salespeople the technical SaaS + cloud concepts they need to sell SaaS smarter + talk tech with confidence. Learn about my online course.