Hardware for artificial intelligence
AI hardware doesn’t make intelligence—it makes computation cheaper, faster, and more brittle.
Artificial intelligence
The field that has been five years away for seventy years, and then was not.
- 26
- in technology
- 9:56
- average
- 258 min
- in total
- 78
- across the network
All briefs
Machine learning
Machine learning is not intelligence — it is applied statistics with memory.
Technological singularity
The singularity is not a prediction — it's a surrender note written in advance of the argument.
Backpropagation
Backpropagation is not learning—it’s bookkeeping for learning.
Convolutional neural network
It did not invent vision—it invented the first scalable, trainable scaffold for spatial pattern recognition.
Diffusion model
Diffusion models don’t generate images—they reverse entropy, one noisy step at a time.
Expert system
Expert systems don’t think — they transcribe. What they deliver depends entirely on what experts wrote down, and nothing more.
Gaussian splatting
Gaussian splatting ditches ray-marching for splatting—and gains speed by abandoning geometry.
Generative adversarial network
GANs don’t model reality—they train a mimic to beat a critic.
High Bandwidth Memory
HBM doesn’t fix memory latency—it weaponises stack height to brute-force bandwidth.
LoRA (machine learning)
LoRA doesn’t shrink models—it shrinks the cost of changing them.
PyTorch
Facebook · 2016
PyTorch is a Python-first deep learning library built by Meta in 2016 as Torch's successor. It enables rapid model prototyping via dynamic computation graphs (Autograd), tensor abstractions, and CUDA-accelerated training. As of 2025, it remains one of the most popular deep learning libraries—alongside TensorFlow and Keras—but its design prioritises researcher agility over production robustness, scalability, or deployment tooling.
Reinforcement learning from human feedback
RLHF outsources ethics to annotators—and then optimises for their rankings, not their intent.
Robotaxi
Robotaxis are running — but they are losing money, failing safety benchmarks, and failing to earn public trust.
Self-driving car
Self-driving cars are not autonomous — they are geographically gated, software-limited, and already lethal.
TensorFlow
Google · 2015
TensorFlow is a machine learning API developed by Google. It is used mainly for training and inference of neural networks. It is free and open-source software released under the Apache License 2.0. It was developed by the Google Brain team for Google's internal use in research and production. The initial version was released under the Apache License 2.0 in 2015. Google released an updated version, TensorFlow 2.0, in September 2019. TensorFlow can be used in a wide variety of programming languages, including Python, JavaScript, C++, and Java. It is one of the most popular deep learning frameworks, alongside others such as PyTorch.
AI data center
AI data centers don’t just run models—they starve the rest of computing of memory.
Vector database
It doesn’t understand meaning — it just maps distances between precomputed guesses.
ChatGPT
ChatGPT is not intelligence — it’s a prompt-to-output relay, optimised for speed, not truth.
ELIZA
ELIZA doesn’t talk — it mirrors, misleads, and exposes how little it takes for humans to believe a machine is listening.
Google Search
Google Search didn’t organise the web—it redefined authority as a function of linkage, then language, then generation.
ImageNet
Fei-Fei Li
ImageNet is a foundational dataset infrastructure — not a model, not a tool, not a service. It succeeded by making scale measurable, not magical.
Siri
Siri didn’t invent voice control—it outsourced understanding to Nuance and offloaded thinking to the cloud.
Tesla Autopilot
Tesla Autopilot passes NHTSA’s test — but only because the test measures what Teslas do, not what drivers must still do.
Amazon Echo
Amazon · 2014
The Echo is a hardware interface for Alexa—a cloud-dependent voice service. It works well for narrow, internet-reliant tasks. It fails as a private, autonomous, or universally accessible system. Its legacy is not intelligence at the edge, but the normalisation of persistent listening and remote AI dependency.
DARPA Grand Challenge
The first driverless car race didn’t produce a working robot — it produced a better way to break them.