[preciselywibble]

Quick summary of my blogs. Please do not assume they will be SFW.

@preciselywibble is for personal things, like maths and health

@approximatelywibble is for things I see that amuse me, runs on a queue

@sometimesmyhandswork is for my drawings, usually of cute hedgehogs

[preciselywibble]

I'm making a thing

I've been working on a tool for managing health data. It's patient-focused and inherently multilingual, and it helps you keep all your health data in one place (especially useful if you have a complex medical history, or ever move between countries).

We're about to launch the kickstarter, but if you could please go to the page now and click "Notify Me On Launch" that would really help the campaign get boosted when it goes live (tomorrow!!)

Coming soon: Health Data Avatar: all your health data in your pocket

An independent secure platform for real-life, multilingual, cross-border patient-owned health data to manage and share on your terms.

[preciselywibble]

I'm making a thing

I've been working on a tool for managing health data. It's patient-focused and inherently multilingual, and it helps you keep all your health data in one place (especially useful if you have a complex medical history, or ever move between countries).

We're about to launch the kickstarter, but if you could please go to the page now and click "Notify Me On Launch" that would really help the campaign get boosted when it goes live (tomorrow!!)

Coming soon: Health Data Avatar: all your health data in your pocket

An independent secure platform for real-life, multilingual, cross-border patient-owned health data to manage and share on your terms.

[preciselywibble]

I'm not sure if "ability to gauge when the youtuber is going to bring up Deus Ex" is a skill I should have spent so much time developing.

[preciselywibble]

Quantum-Enhanced LLM Efficient Fine Tuning

Low-Rank Adaptation (LoRA) enables efficient fine-tuning of pre-trained language models via low-rank matrix approximation, which is effectiv

LoRAs (and similar techniques) let you represent a large number of numbers in a way similar to Singular Value Decomposition; you represent a big matrix by approximating it as an outer product of two smaller matrices, losing some information as you do. You want this for fine-tuning neural networks because you want to be able to impact all of the weights of the system with as few parameters as you can get away with.

This leads to a "rank bottleneck". Think of it as the bandwidth of a channel being limited by the narrowest point. One way to improve accuracy is to just use bigger matrices, which is expensive. This paper instead enhances the narrowest point of the classical channel with a quantum channel.

And then they actually went and did it. Actual usage of quantum computers. Demonstrating usefulness. Or at least demonstrating a kind of behaviour that some more engineering might make practical - the paper makes some claims about step reduction and parameter count but not about literal time and money.

[natural-blogarithm]

i fucking hate my brain. i have a stupid mashup of creep by radiohead and we are the crystal gems from steven universe stuck in my head and i don't even have the audio editing skills to inflict it on the world

[preciselywibble]

[preciselywibble]

No matter how like a young deer you are, animals will always be fauna.

[bubbloquacious]

Presented without comment.

[machine-unlearning]

Operation Mincemeat is surprisingly good for a musical that sometimes sounds like one of those TikToks parodying Lin Manuel Miranda's musical style

[preciselywibble]

I really, really recommend going to see it. It manages to be both wittier and more heartfelt than almost everything else.

[clockwork-fishbowl]

The dynamic of "scientist who wants to fuck the thing they're studying" gets exponentially better the further removed from the biological sciences you get.

Xenobiologist wanting to bang an alien is old news. Give me a programmer in love with his computer. Quantum physicist who gets horny when he gets to operate the partical accelerator. Do you understand

[preciselywibble]

I passed an IBM System One in an office lobby and was dumbfounded for a full minute.

[preciselywibble]

What do you mean it's pronounced "doej" not "doggie"??

[invisible-goats]

I am once again taking the word twat away from americans until they stop pronouncing it as twot

[preciselywibble]

"twat do I owe this honour?"

[raginrayguns]

pretty much as expected i guess that trump's attack on government inefficiency began with making people wait an additional three months for some bureaucrat to finishh their paperwork

[preciselywibble]

there's nothing as efficient as changing the targets

[medium-amount-of-gay]

[positively-knotted]

I work over non-algebraically closed fields bc I only use R and Z/2. I am upset.

[sapphic-adjunction-deactivated2]

My advice is to start slow..... set a goal this year of using \mathbb{Q}[\sqrt{2} + \imath] and Z/3..... and then go from there.

[preciselywibble]

[preciselywibble]

I have a method??

I've just been sent a photo from a conference of a researcher discussing "The [PreciselyWibble] Methodology" and I am equal parts delighted and confused that some (rather pretty) applications of Laurent polynomials to quantum computing are getting use.

[preciselywibble]

Just had a look back at the paper it's based on. It's 50 pages long, and I was in so much pain and on such horrendous meds at the time. I remember moving from the easy case (novel approach, and yields a more general framework, was enough to be interesting) to the complicated case (novel domain, about 10 people in the world might give a fuck, but kinda solves the problem outright) and realising that it was going to add 30 pages of deeply unpleasant typesetting to my life.

[preciselywibble]

I have a method??

I've just been sent a photo from a conference of a researcher discussing "The [PreciselyWibble] Methodology" and I am equal parts delighted and confused that some (rather pretty) applications of Laurent polynomials to quantum computing are getting use.

[aufenthaltsraumspizza]

“Hello Hartshorne my old friend, I’ve come to learn from you again.”

— Algebraic Geometry