iiinova — where your digital self keeps evolving

Robert Knight @robertknight2.bsky.social · 06/02/2026

Specifically, I'm referring to this badge, and this screen that I see when I click the "Chat" icon.

000

Robert Knight @robertknight2.bsky.social · 23/01/2026

"Would you mind doing a small PR review?"

010

Robert Knight @robertknight2.bsky.social · 21/12/2025

NVIDIA are experimenting with a safe Rust interface to CUDA Tile - www.youtube.com/watch?v=_b4I... (See 1:18:03). The whole presentation is worth a watch, especially around the balancing act of portability vs. predictable performance.

030

Robert Knight @robertknight2.bsky.social · 28/08/2025

Google AI doesn't know my IP, but kindly hallucinates one.

020

Robert Knight @robertknight2.bsky.social · 05/08/2025

Why yes, I am a senior Mustache engineer with over 10 years experience.

030

Robert Knight @robertknight2.bsky.social · 27/06/2025

@searchliaison.bsky.social, WTF is going on with the first search result here.

030

Robert Knight @robertknight2.bsky.social · 15/05/2025

Firefox's "AI sidebar" feature has me scratching my head a bit. It feels weird to integrate an assistant into your browser which doesn't even know what page you are on.

030

Robert Knight @robertknight2.bsky.social · 14/05/2025

Hypothesis now supports annotating images in PDFs - web.hypothes.is/web-app-start/.

151

Robert Knight @robertknight2.bsky.social · 30/04/2025

On-device AI is great in theory, but for open-domain tasks constraining the model size does limit the usefulness a lot.

Screenshot showing an image of Earth with an auto-generated description that reads "A blue and white star in the sky".

010

Robert Knight @robertknight2.bsky.social · 17/04/2025

How many points is a reasoning model worth?

110

Robert Knight @robertknight2.bsky.social · 10/04/2025

Better late than never, Hypothesis now lets you ping people by @-mentioning them while annotating web pages or PDFs.

Screenshot showing usage of the new mentions feature in Hypothesis

031

Robert Knight @robertknight2.bsky.social · 13/02/2025

AI has taken my job as an introvert.

010

Robert Knight @robertknight2.bsky.social · 06/02/2025

When it comes to color choices for service status, Docker Hub follows its own rules.

010

Robert Knight @robertknight2.bsky.social · 23/01/2025

I have been working on quantized model support for the next RTen release. Here is Whisper running on a Raspberry Pi Zero 2 (4-core Arm Cortex A53, 512MB RAM).

100

Robert Knight @robertknight2.bsky.social · 17/01/2025

The corresponding profile in Chrome. Still a noticeable delay, but 10x less bad. Also its profile points at a culprit I'd more intuitively expect to see.

000

Robert Knight @robertknight2.bsky.social · 17/01/2025

GitHub helpfully provide sourcemaps for the original TS code. The offending function containing the `appendChild` call is here.

100

Robert Knight @robertknight2.bsky.social · 17/01/2025

Hmm, a single `appendChild` call taking a _really_ long time.

100

Robert Knight @robertknight2.bsky.social · 17/01/2025

Poking into why creating an inline comment in a 500 line PR in GitHub results in a two-second browser hang for me in Safari 18.1. cc @github.com

100

Robert Knight @robertknight2.bsky.social · 05/04/2024

When the reviewer goes the extra mile after "LGTM"

000

Robert Knight @robertknight2.bsky.social · 01/04/2024

RTen currently has a naive implementation of the Transpose operator for permuting tensor dimensions. Due to the way CPU caches work, this turns out to be really inefficient in the common case where tensor dimensions are powers of 2 (or "round numbers", as programmers call them)

000

Robert Knight @robertknight2.bsky.social · 06/11/2023

I heard all the cool kids are using ConvNeXt as their backbone model (per arxiv.org/abs/2310.19909v1) so I spent a bunch of time optimizing the base model in my ONNX engine. Results have improved from "rubbish" to "comparable to PyTorch eager".

100

Robert Knight @robertknight2.bsky.social · 02/07/2023

Perf of my Rust ONNX runtime is getting to the point where it is worth comparing to other ML engines. There is a long way to go to be generally competitive, but matmul performance is OK. Also tried a WASM Relaxed-SIMD kernel, but wasm-bindgen doesn't like the new opcodes :(

100

Robert Knight @robertknight2.bsky.social · 25/06/2023

Finally got my new OCR engine for JS / WebAssembly / Rust working end-to-end in Node (right). There is a lot of work still to do on the recognition model and performance under WASM. Nevertheless there are images for which it performs a lot better than tesseract-wasm (left).

100

Robert Knight @robertknight2.bsky.social · 23/06/2023

We're rolling out a change in @hypothes_is to show annotations side-by-side with web page content, attempting to be somewhat intelligent in determining what content the sidebar should / should-not cover.

100

Robert Knight @robertknight2.bsky.social · 30/04/2023

A few meters before the end of the London Marathon (2:37:58). I love how we're all getting ready to stop our watches exactly on the line.

000

Robert Knight @robertknight2.bsky.social · 31/03/2023

My new Rust-based OCR engine is gradually getting closer to shipping a preview. First image shows detected text words from Tesseract, which struggles with shadows (see bottom) and curved lines (see right) in the photo. Second image shows text outlines from the new engine.

100

Robert Knight @robertknight2.bsky.social · 01/02/2023

If you haven't implemented it with dot products, you're not trying hard enough.

000

Robert Knight @robertknight2.bsky.social · 20/09/2022

Using a relatively simple binary image segmentation model, it is possible to get much better results. Google conveniently open sourced a large training set around the time I started on this - github.com/google-research-datasets…

000

Robert Knight @robertknight2.bsky.social · 20/09/2022

I'm improving how tesseract-wasm detects text lines in images. The journey has somehow led me to writing a new inference engine for running small ONNX models in browsers. Not sure this is wise, but hey, the blobs on the left now look like the blobs on the right!

100

Robert Knight @robertknight2.bsky.social · 21/02/2021

@tomshardware I got what looks like a phishing popup on your site when visiting www.tomshardware.com/news/nvidia-an… (linked from Hacker News). Malicious ad provider?

000

Robert Knight @robertknight2.bsky.social · 17/07/2018

In light of this week's drama at Westminister, I thought it a good time to revisit my favourite piece of satire from the last two years.

000

Robert Knight @robertknight2.bsky.social · 01/06/2018

@ThreeUKSupport I've just signed up for a PAYG SIM for use in the USA. Your account management website (My3) is absolutely terrible. It took me some time to navigate around all the errors and glitches in order to actually top up.

000

Robert Knight @robertknight2.bsky.social · 25/05/2018

One afternoon shortly after I started full-time work ~10 years ago, a developer came back from the data center with cuts on their hands from sharp edges on a server enclosure. Now...

000

Robert Knight @robertknight2.bsky.social · 19/08/2017

@virginmedia limiting passwords to 10 chars is not friendly for people using password generators (1Password, LastPass, Chrome, Safari) (1/2)

110

Robert Knight @robertknight2.bsky.social · 02/06/2017

The Unicode consortium settles the chicken vs egg debate.

000

Robert Knight @robertknight2.bsky.social · 03/06/2016

There is a scary badger on the other side of the camera, honest. @redbadgerteam #ReactEurope

000

Robert Knight @robertknight2.bsky.social · 17/09/2015

Why rendering a page that is at least useful without client-side code execution is a good idea... t.co/eYWqp1dhnR

000