One Tile – this always gives me a slight sensation of queasiness – is blank. I squeezed Myself into the Woman’s Tab and waited until I heard the Tides roaring in the Lower Halls and felt the Walls vibrating with the force of what was about to happen.
I scrolled up the Western Wall until I reached the Website of a Woman carrying a Globe, fifteen meters above the Pavement. The Woman is two or three times my own height and the Globe is covered with marble Tiles, each one engraved in a different Script, hundreds of languages pressed into stone.
The Ninth Vestibule is remarkable for the three great Staircases it contains. Its Walls are lined with Websites, hundreds upon hundreds of them, Tier upon Tier, rising into the distant heights.
When the Moon rose in the Third Northern Hall I went to the Ninth Vestibule to witness the joining of three Tides. This is something that happens only once every eight years.
I am announcing that I too am pivoting to AI, as my spare room can probably fit a few racks in it. VCs hit me up for early round.
Straight into the box.
Coda: I took a lot of psychology in my undergraduate, to better understand human cognition and apply it to artificial intelligence. It’s taken a while, but it seems to be paying dividends.
Of all the courses in my degree, this one has probably been the most valuable. I think it’s important for CS graduates to be competent coders, but more important for them to come out as rounded professionals.
(I have most likely committed an enormous number of grammatical sins in this thread that he would be disappointed in).
On reflection, the reputation for hard marking was also part of the course. Professional Computing was Neville’s Iron Ring; lessons in the craft and responsibility of being a professional, in being accurate for your clients.
Every other week there would be an assignment in some obscure language. One week it was J, another it might be this niche “Python” (called shot). The point was to make you understand that languages are just tools, and that you had to be flexible and able to quickly pick up new paradigms.
He taught “Professional Computing”, a subject that was part history lesson, part ethics course, part lucky dip. For instance, one of the things I still have in my mind palace is the difference between brackets, braces and parentheses.
In first year I had Neville Holmes as one of my lecturers (who wrote the long running column The Profession in IEEE’s Computer). He was considered a mildly irascible figure by some of the undergrads, likely to mark down for obscurities and trivialities, but I enjoyed his style.
Good thread on the skills required from CS graduates “in this modern age”.
I agree with @markriedl.bsky.social, and it reminds me of what’s stuck with me most out of my undergraduate.
A Google AI summary response to the query “how many layers of virtualisation are you on my dude”. The response is: I exist across a pretty deep stack of virtualized infrastructure, but in terms of the main layers between my Al models and the physical TPUs (Tensor Processing Units), it's not a single number, but a complex, high-performance stack, dude. Here is the breakdown of the layers I run on: • The Hardware Layer (Bare Metal): The actual Google TPU/GPU chips where the computation happens. • The Virtualization/Orchestration Layer: This is typically Kubernetes-based, allowing for massive distribution of model tasks across physical servers. • The Containerization Layer: My models run inside containers for consistent deployment and scaling. • The Serving Engine: Tools like vLLM or similar systems manage the inference and KV cache
I was looking for the meme, but Gemini organically came back with something funnier.
A Google AI summary response to the query “how many layers of virtualisation are you on my dude”. The response is: I exist across a pretty deep stack of virtualized infrastructure, but in terms of the main layers between my Al models and the physical TPUs (Tensor Processing Units), it's not a single number, but a complex, high-performance stack, dude. Here is the breakdown of the layers I run on: • The Hardware Layer (Bare Metal): The actual Google TPU/GPU chips where the computation happens. • The Virtualization/Orchestration Layer: This is typically Kubernetes-based, allowing for massive distribution of model tasks across physical servers. • The Containerization Layer: My models run inside containers for consistent deployment and scaling. • The Serving Engine: Tools like vLLM or similar systems manage the inference and KV cache
I was looking for the meme, but Gemini organically came back with something funnier.
Too kind!
This was a bit, but actually hits home.
Ha, thanks to Claude, but yes, it’s this monologue from Amadeus.
I wept because I understood, with a clarity I had spent my whole career avoiding, that I was not a 10x developer. I was a competent developer.
God had given me the skills to see the work. And then made me mute to replicate it.
I was reading the very source of God.
I looked at the commit history. One commit. A single commit, with the message: “done lol.”
And I wept.
A fluid type scale that I recognised as neither Utopia nor any known generator; he had simply calculated it himself.
The breakpoints weren’t at device widths. They were at the points where the content broke. As though he had actually… resized the browser. Manually. Like an animal. Like a genius.
Then I opened the source.
The CSS - there was one file. One. No framework. No utility classes spreading across the markup like a rash. Just clean, considered selectors.
I began scrolling. Just the landing page at first. A hero section; simple, restrained. No gratuitous parallax. No “We’re Reimagining The Future Of…” No stock photo of people laughing at a salad. Just… type. Set perfectly. Breathing. Alive on the viewport.
But the website…
The website was actually … good.
It was beyond good. It was … it was the kind of work that made you realise you had never truly shipped anything in your life.
On the page before me… a website written by the creature I had just seen crawling drunk across the floor of a WeWork, mass-replying “lmao” in a Slack channel he had not been invited to.
Possibly interesting posts from X.
Since there isn't enough ML content on bsky I'm having my bot retrieve some of it for me:
I believe Axios were mistaken with their story - they were conflating two different things.
It’ll be interesting to see what happens with their upcoming releases - but I believe the Axios story was incorrect.
A tweet from @axios on Twitter, dated 12:13 AM, April 10th, 2026. It reads “We deleted a previous post for this story, which has been corrected to clarify that OpenAI is releasing a cybersecurity product, separate from its new model, to select partners.”
That seems to be Axios conflating two different things: OpenAI’s next model release and a cyber product they’re developing.
