Platform engineers make sense of external applications and infrastructure and synthesize that into an internal platform. There are significant upsides to optimizing horizontal software and hardware for a company-specific use case.However, there is a tendency to create a system so … | Continue reading
Frontend engineering is complicated. We know what we want things to look like, but our code doesn’t always reflect that. Off-the-shelf UI kits can be easy to add, but tough to modify to our liking. AI can help, but unbounded, it’s hard to integrate into our frontend stack.That’s … | Continue reading
The frontend ecosystem is notoriously confusing. At every layer, there seem to be incompatible, competing standards.No universal import system. ESModules, CommonJS, Asynchronous Module Definition (AMD), and Universal Module Definition (UMD) are all different ways you can import o … | Continue reading
Wikipedia's top five accounts (by number of edits) are all bots. There’s MalnadachBot (11 million edits), WP 1.0 bot(10 million), Cydebot (6.8 million), ClueBot NG (6.3 million), and AnomieBOT (5.9 million.). These bots range in functionality from migrating tables, formats, and m … | Continue reading
The half-life of software usually increases as you move down the stack. The average age of infrastructure components is much older than the frontend JavaScript framework. But in today’s AI stack, the world is flipped.The infrastructure layer in AI might have the shortest half-lif … | Continue reading
200 milliseconds doesn’t sound like a lot, but it’s an eternity for latency-sensitive code (you could travel around the world at the speed of light in 133ms). If you’re working with latency-sensitive code over the network, you might have found that sometimes your requests take mu … | Continue reading
There’s an interesting argument from Sam Bankman-Fried (SBF) in his new biography by Michael Lewis,I could go on and on about the failings of Shakespeare . . . but really I shouldn’t need to: the Bayesian priors are pretty damning. About half the people born since 1600 have been … | Continue reading
In the 1960s, Fairchild Semiconductor produced early versions of transistors for the government. They wanted to expand and wanted to land RCA as their first commercial client. RCA bought vacuum tubes at $1.05 each for their UHF television broadcasting. On the other hand, the tran … | Continue reading
Moravec’s Paradox is the observation that high-level reasoning (e.g., chess, math) is relatively easy for computers to perform, while simple sensory tasks (e.g., perception, reflexes, mobility) are much harder.Moravec believed that most people thought this result was the opposite … | Continue reading
In that land the great experiment was to be made, by civilized man, of the attempt to construct society upon a new basis; and it was there, for the first time, that theories hitherto unknown, or deemed impracticable, were to exhibit a spectacle for which the world had not been pr … | Continue reading
Programming languages are expensive to develop. They take expert programmers with a deep technical skill set. They require probably a decade of development to reach maturity. For example, Go took five years to reach version 1.0, with top-tier engineers working on it (Rob Pike, Ro … | Continue reading
The third dimension is coming to computing. * Capture 3D natively. Spatial video can be captured on the iPhone 15. Higher quality video and photos that can map to 3D settings. * From 2D to 3D. Techniques like Gaussian Splatting, Photogrammetry, Neural Radiance Fields (NeRF) predi … | Continue reading
Katalin Karikó shared the Nobel Prize in Medicine this week for her work on mRNA vaccines that were instrumental in fighting COVID-19.In 1995, her research was not showing results, and UPenn gave her the choice to quit or be demoted (she chose demotion).In 2005, she wrote her bre … | Continue reading
The Feynman diagram is a visual representation of subatomic particles and their interactions. The diagrams are not meant to literally represent the particles and their interactions but rather a way to represent the mathematical formulas that describe them. They are extremely usef … | Continue reading
Database migrations sound difficult but are even more difficult in practice. Unlike stateless code, they are an arrow in time. 1. Development / production parity is impossible. You can (and should) try to recreate production state as much as possible. This is much easier at the A … | Continue reading
The “wizard” interface design has been around for decades: interfaces that guide users step by step through complicated procedures, usually installations. Even today, you see them on the web with software like Typeform (splits form questions into a multi-step UI, which is aesthet … | Continue reading
By June 1949, people had begun to realize that it was not so easy to get a program right as had at one time appeared. It was on one of my journeys between the EDSAC room and the punching equipment that the realization came over me with full force that a good part of the remainder … | Continue reading
Compression algorithms encode information in efficient ways. It’s what makes a zip file smaller than the sum of its parts, a mp3 smaller than a studio recording FLAC, or a JPEG image smaller than a RAW photo. Compression can be lossy (irreversible information lost in the process) … | Continue reading
You can think of a platform shift as a change in the dominant layer that applications are built on. It might look like Software > Hardware. Or Hardware > Software. A new layer becomes the dominant layer (e.g., most applications are being built on top of the new layer rather than … | Continue reading
After the meteoric rise and fall of web3, Apple and Google might have shipped a real crypto-for-everyone product. You might have noticed a new way to sign in on Apple or Google — instead of using a password, you can opt for a passkey, a password-less type of authentication. You c … | Continue reading
“Data is the new oil” was the slogan of the last decade. Companies were told how valuable their data was (or could be). They rushed to invest in a modern data stack and store terabytes of data in data warehouses. Data science teams crunched the numbers, and the analyses were supp … | Continue reading
Transformers and other AI breakthroughs have shown state-of-the-art performance across different modalities * Text-to-Text (OpenAI ChatGPT) * Text-to-Image (Stable Diffusion) * Image-to-Text (Open AI CLIP) * Speech-to-Text (OpenAI Whisper) * Text-to-Speech (Meta’s Massively Multi … | Continue reading
In 1968, a group of teachers and students participated in a psychology study. The students were given an IQ test by the researchers. The results were not disclosed to the teachers. The researchers shared with the teachers that five students exhibited unusually high IQ scores and … | Continue reading
The distinction between fast and slow thinking, or system 1 and system 2 thinking, made popular by Daniel Kahneman’s book *Thinking Fast and Slow*, might be a helpful lens to view LLMs. System 1 is fast, automatic, frequent, emotional, stereotypic, and subconscious. Examples of S … | Continue reading
The fundamental attribution error is our tendency to (wrongly) attribute the behaviors of others to internal, personal characteristics and our own behaviors to external, situational factors. For example, if someone is late, we think they are lazy. If we are late, we think we were … | Continue reading
Markdown isn’t a perfect abstraction. In fact, when it was initially released (2004), the spec (if it could even be called that) was so ambiguous that, for the next decade, dozens of flavors of Markdown co-existed. A formal specification was introduced in 2014.Markdown was design … | Continue reading
Besides black art, there is only automation and mechanization. — Federico García LorcaIn 2018, Musk was trying to ramp up production in his Tesla Fremont factory from 2,000 to 5,000. He revisited every process and system, cutting corners wherever he could (and sometimes removing … | Continue reading
I just finished the Musk biography by Isaacson. It was a good read with original reporting that I had never read before. Musk is a complicated figure.But there aren’t enough Isaacsons to write everyone’s story. Writing a biography is an interesting task — analyze and synthesize a … | Continue reading
The web is no longer as programmable as it was, but there’s still hope for end-user software. And AI might bridge the gap.Excel is one of the world’s most used pieces of software. It is the closest thing we have to customized end-user software before the age of AI. Most SaaS appl … | Continue reading
Renaissance Technologies (“Ren Tech”) is one of the world's most successful and secretive hedge funds. It was founded by Jim Simons, a mathematician with a long history of contributions (Chern-Simons form, developments in string theory, and more in geometry and topology). He was … | Continue reading
Product-led growth and bottoms-up distribution can be a very powerful business model. Dropbox was a poster child of both. Besides being a product that “just worked” (table stakes), it grew from word-of-mouth and viral marketing campaigns. Its referral program with two-sided incen … | Continue reading
“AI-generated code is making me a worse developer.”“GitHub Copilot code is riddled with bugs.”“Generated code is way less efficient than something I wrote myself.”Developers are right. AI-generated code isn’t as good as something you or I could write. It has bugs, often hard to f … | Continue reading
In 458 BC, the Romans needed a dictator (an office reserved for extreme emergencies in Roman times) to respond to a massive enemy invasion. Cincinnatus was appointed, reached his command, defeated the enemy, and laid down his dictatorship after just 16 days. He returned to his fa … | Continue reading
Harry Styles and One Direction fandoms are at odds about the recent influx of “leaked snippets” of possible demo tracks by their favorite artists. They can’t tell if they are legitimate or AI-generated (source).Sites will claim to be able to identify AI-generated writing, images, … | Continue reading
The next is replace -- replace feature after feature after feature of the older Cerner system with a new Cerner system, new Millennium, which we are not coding in Java like we usually do. The new Cerner system is being generated -- as you know, generative AI generates code. We ha … | Continue reading
Scope creep is inevitable. A project starts with a small mandate and ends with an enormous, uncontrolled scope. What was a small change is now a complete rewrite. It’s easier to visualize scope creep with software projects — you can easily see the number of services or areas of c … | Continue reading
When the initial Stable Diffusion models first came out, I fine-tuned them to add myself as a new concept. I did this by using the DreamBooth algorithm via fine-tuning. It only took about 5-10 images of myself. The results were pretty good (learning a new token, “mattrickard” on … | Continue reading
Finding the “low-hanging fruit” is not only the most straightforward plan, but it’s often one of the most effective. Continuously executing against the obvious plan gets you much further than you’d think.Some reasons why: * Simple to apply. Do the most obvious things first. * Bia … | Continue reading
A recent paper (Large Langauge Models as Optimizers) by researchers at Google DeepMind found that AI-optimized prompts (via another model) can outperform humans by up to 50% on certain benchmarks.When I wrote Prompt Engineering Shouldn’t Exist at the start of 2023, I outlined som … | Continue reading
Commoditizing your complements is the first rule of computer businesses. Microsoft made hardware a commodity in the PC Era to sell its operating system. Their non-exclusive deal with IBM let them sell DOS to any OEM they could find.Demand for a product increases when the price of … | Continue reading
Consensus algorithms are at the core of distributed systems. How do you manage consistency across multiple servers or nodes?The Raft Consensus Algorithm is a distributed system protocol that’s widely used (including by systems like Kubernetes, via etcd). It is equivalent in fault … | Continue reading
Widgets on the desktop. Widgets on the lock screen. Like many Apple features, widgets on the desktop originated with the hacker community for decades before being formally adopted in the proprietary distribution.(from user nnkd on Reddit, 9 years ago).I used to be part of these c … | Continue reading
If you want to coerce a typed JSON response out of an LLM, you have a few options:Control the token distributions via state machines with a regex or context-free grammar. The benefit of this method is correctness. You are guaranteed to get a valid response on the first generation … | Continue reading
Patrick Collison, the CEO and co-founder of Stripe, maintains a list of people quickly accomplishing ambitious things together titled Fast. On the page, he talks about The Eiffel Tower (739 days), Boeing 747 (930 days), JavaScript (10 days), Git (17 days), The Empie State Buildin … | Continue reading
It’s hard to get a measure of organizational productivity. It’s often company, project, and goal-specific. But when it comes to engineering organizations, there are at least some metrics you can collect to help investigate a hypothesis or serve as red flags for further investigat … | Continue reading
TypeScript’s type system is Turing Complete. The type system is extremely advanced — you can model algebraic data types (unions, intersections, tuples), conditional types, mapped types, and more. It is a structural type system. In TS 4.1, template literal types were introduced, m … | Continue reading
Some have described the bimodal distribution of GPU availability as “GPU-poor” and “GPU-rich” companies.Will most returns to AI accrue to the companies with exclusive access to compute via GPUs and hardware, which are in short supply but necessary for large-scale training and inf … | Continue reading
Back in May, I got Vicuna 7B — a chat-tuned version of the original Llama model, working entirely in the browser via the new WebGPU APIs that had shipped in Chrome. I open-sourced a React library to make it easy to use (react-llm).Today, I’m releasing an updated version of this o … | Continue reading