David Eccles 🌻🇵🇸7x🩹🛡️
LightNews LightNews
banner
gringene.org
David Eccles 🌻🇵🇸7x🩹🛡️
@gringene.org
1.6K followers 1.3K following 4K posts
Neurodivergent freelance bioinformatician, cuddler of data, and theoretical geneticist, with a love for Free Software and the hacking thereof. Rangitāne o Wairau. He/they.

https://ko-fi.com/gringer

Website: https://www.gringene.org
Posts Replies Media Videos
gringene.org
The dropped stitches (loops? / knots?) add character, and will help to remind me where I began a few years from now.

#CrochetProject
The beginnings of a pot holder crochet, after about 5 rounds. The shape is bent in the middle; the result of troubles I had early on in trying to crochet the chains together, as well as a few odd mistakes elsewhere, every now and then. The crochet end loop is filled with a 4mm wooden crochet hook that I made myself, with a smoothed quarter-cylinder handle (due to the curved wooden plank it was cut from).
January 14, 2026 at 12:52 PM
gringene.org
Round 1 done, apparently.

It was very tight in the first row, so I guess my chain was too tight. I've noticed (after the first row) that the women in the video I'm watching has an extremely loose thread, but I can't do it that loose [yet] because the crook in my hook is too shallow.
Crochet attempts 1, 2, 3, 4 and 5, in linen-coloured 8-ply cotton thread, attached with yellow stitch markers, together with crochet attempt 6 with an apple green stitch marker. Attempt six has one row of about 55 chains, and two surrounding rows of... hopefully approximately that number of crochets, making the first round. The stitch marker is attached to the first loop of round 2. The oiled 4mm hook is inside the loop at the start of round 2.
January 14, 2026 at 7:23 AM
gringene.org
I've just been working on the final data submission touches for a #Bioinformatics research project, and my hunting for metadata reminded me that the first tranche of sequencing data for that project arrived a few days before I was made suddenly "redundant".

#Triggering
My meeting notes from a meeting in which I was told that my role as a bioinformatician was being disestablished as part of a restructure that [to me] appeared to involve no material changes (and especially no improvements) beyond the disestablishment of my role. I was told that this restructure had nothing to do with my personal grievances and/or complaints against [burnt orange].
January 13, 2026 at 11:28 AM
gringene.org
After making a 4mm hook, I think I now see how the hooks are meant to function. It looks like the thread should sit in the notch and make the resultant summed diameter the same as the hook diameter.

I might file the head more on this one. It's a bit too long for lifting loops off.

#CrochetProject
4 crochet chain loops (numbered 1-4) from linen-coloured cotton thread, together with an aborted pot holder end (0.25 rows) numbered 5. There is also an oiled 5.5mm wooden crochet hook, and an un-oiled 4mm wooden crochet hook, with the beginning of another chain on it.
January 13, 2026 at 6:07 AM
gringene.org
Learning how to chain.

It feels like the hook needs some modifications to work better, but I can't work out how much of that feeling is a skill issue, a mismatched hook size, or an actual design flaw.

It's at least usable, so I think I'll leave it alone until I know better.

#CrochetProject
Three crochet chains created from linen-coloured 8-ply cotton, created as part of my learning process, together with the 5.5mm hook I used to create them. The ends of the chains are tied together using random knots, and form loops of about 10cm diameter when laid in a ring. They are labeled with yellow stitch markers, which have numbers on them (1, 2, 3) indicating the order in which they were created.
January 12, 2026 at 7:47 AM
gringene.org
Oiling is now done.

#CrochetProject
A hand-carved / shaped crochet hook (size 5.5mm) on a light blue bedsheet. The hook was created from a cherry wood plank that was first cut into a square-ish rod with a hacksaw, then shaped using a knife, sandpaper, and a polishing tool, then hook added using a mini file. The finished product was then oiled with linseed oil and walnut oil.
January 11, 2026 at 7:18 PM
gringene.org
Hook end detail, after carving and shaping with a mini file.

In hindsight, I should have shaped the end into a point *before* carving out the hook; I ended up losing a bit of wood from the hook front by doing the carving first.

#CrochetProject
Close-up photo of a wooden crochet hook end, filed to a point and including the thread catching indentation, together with the 5.5mm hole used for sizing it, and the half-round mini file used for doing most of the carving & shaping.
January 11, 2026 at 9:37 AM
gringene.org
Shaping and sizing is now done, but I still need to form the hook end, then oil it. Here's the hook thingy, together with the holes I used for getting the size approximately right.

#CrochetProject
A mostly-finished crochet hook, without the hook, fashioned out of sun-dried cherry. Shown here with a hardboard slice that has two holes in it, one with 5.5 next to it, and one with 6 next to it.
January 11, 2026 at 9:06 AM
gringene.org
#AISlop which car park is the central electric charger servicing?
An image of two car-like things on top of a parking-like thing with odd lines. The parking markings have inconsistent dimensions. There are three chargers and two parks.
January 5, 2026 at 8:22 PM
gringene.org
Oh, that's... bad.
Manage My Health has identified a cyber security incident involving unauthorised access to our systems. The matter is under active investigation and containment steps have been taken. We are working with our partners and relevant authorities and will provide further updates through formal statements as information is confirmed.
December 31, 2025 at 2:43 AM
gringene.org
Straw man in what way?

How is it evasion?

I made a statement about research in general, rather than a statement specifically about *this* research.

Human research should be avoided until hypotheses are robust and well-formed. It's not appropriate to jump straight into human medical research.
Profile clip for Kevin Ashton in which it is declared "I do not argue, debate, or engage" (in direct contradiction to what is now happening)
December 29, 2025 at 7:21 PM
gringene.org
What were you trying to demonstrate with this paper? I do not think it's saying what you think it's saying.

This paper is taking particular issue with amyloid cascade mouse models, rather than mouse models in general.
Excerpt from the referenced paper: "Mice are valuable for investigating the fundamental mechanisms.... A reconceptualisation of murine models is needed. A question remains whether mice engineered with humanized genes will prove valuable for modeling human AD. Many of us are hopeful, but...."
December 29, 2025 at 7:17 PM
gringene.org
Not too surprising, given that Māori ancestors came from southeast Asia only a few thousand years ago.
This map, created by me (Rangitāne o Wairau) is based on genetic, archaeological, and radiocarbon dating data and traces the migration routes of the Polynesian population, including the discovery of New Zealand by Māori. It traces Polynesian migration from Taiwan (before 3000 BC) through the Pacific, and then to Aotearoa (1200 AD)
December 27, 2025 at 10:26 PM
gringene.org
In any case, let's see if this claim stacks up for the longest assembly...

Nope. It's a patchwork assembly with lots of breaks, most of them being on the edge(s) of long tandem repeats.
Repaver plot (k=50) for ENA|OZ289192|OZ289192.1. There are dozens of contig breaks, and some are not obviously in repetitive regions. Repaver plot (k=125) for ENA|OZ289192|OZ289192.1. The increased kmer length makes it more obvious that some of the breaks are in non-repetitive regions. Most breaks *are* on the edge of large tandem repeat regions, typically with a repeat unit length of over 1kb.
December 27, 2025 at 1:24 AM
gringene.org
LLMs are bullshitting plagiarism machines, made worse by the deliberate opacity of their training data.

These tools *could* provide whakapapa: a summary of their algorithmic ancestry; contexto does it.

But the owners choose not to do this, because it would expose their crimes in bright sunlight.
FAQ from the word game contexto: [How is the word order defined?] The game uses an artificial intelligence algorithm and thousands of texts to calculate the similarity of the words in relation to the word of the day. Not necessarily it is related to the meaning of the words, but to the proximity in which they are used on the internet. For example, if the word of the day were “infinite”, words related to “love” or words related to “universe” might be close to the word of the day because “infinite” is commonly used in those two different contexts. In similar reasoning, if “tv” and “television”, for example, are in very different positions, it means that they are used differently in relation to the word of the day, despite being the same object. [Credits] The data used to calculate the proximity between words come from Stanford's NLP Group.
December 26, 2025 at 2:34 PM
gringene.org
* sigh *

I thought we were out of the woods this time round.

Time for another go round the death and disability train.
SARS-CoV 2 in Wastewater and Reported COVID-19 case numbers, showing a gradual increase in wastewater counts since October 2025, from a low point of about 1M genome copies per person per day up to about 4M genome copies per person per day at the most recent measurement (week ending 21 December).
December 25, 2025 at 8:07 AM
gringene.org
Don't trust the image ==> don't trust the article
Image of fridge contents with the appearance of a generative algorithm image, showing something that evokes the idea of egg boxes, but without the actual form of an egg box.
December 23, 2025 at 9:04 AM
gringene.org
#AngelineEra I'm really glad that those last shooters were . 1048 deaths.

#HanTanisChallenge
Found Scale: 11; at the apex of Ballyspire 2: Blue Leap
December 23, 2025 at 8:16 AM
gringene.org
#AngelineEra Surprise ramming dinosaurs got me again. 1048 deaths.

#HanTanisChallenge
A red-mist kill scene in Angeline Era, caused by being rammed by a ramming dinosaur high up in the clouds.
December 23, 2025 at 8:07 AM
gringene.org
#AngelineEra Some meditative jumping... how hard can it be?

... a sniper!

1036 deaths.

#HanTanisChallenge
A tower with a lot of different jumping puzzles. The visible platforms show a button that can be pressed, a sniper, and the sniper's target.
December 23, 2025 at 5:10 AM
gringene.org
Got it out of my head. I left some of Bo Burnham's original lyrics in because they fitted well enough.

bsky.app/profile/grin...

Disclaimer: no LLM or other form of artificial bullshit generation was [knowingly] used in the creation of these lyrics
Welcome to the internet, popups are abound We will watch your usage, for the scams that can be found We've got mountains of content, and it's getting worse Everytime you're interested, we'll go for your purse Welcome to the shitternet, come and take a seat Would you like another ad? We've got a hundred as a treat There's no need to panic, this isn't a test, haha Just nod or shake your head, and we'll do the rest Welcome to the internet, what would you prefer? Would you like a bar of Disney soap, here's Pinterest selling fur Subscribe to Prime, we'll get you on the latest new craze We've got a milion different ways to get paid Welcome to the internet, with you we'll go far Here's a child with cancer, here's a lacey push-up bra We've got Netflix, and Hulu, and Disney, and shorts And that website you just visited was created so Google could harvest your data for profits! ** Could I interest you in advertising all of the time? A little bit of advertising, all of the time Apathy's our duty and attention is all mined Advertising anything, all of the time ** Welcome to the shitternet, click and join my Patreon A scammer plies on Facebook selling pictures of your mom They are grainy, and off-putting, and clearly AI Don't act suprised, it's your compulsion to buy We need your private number so that you can play this game, Show us pictures of your children, we can sell you all that fame, Here's an update, Buy or donate, Sext your roommate, Click to get a Gaming prize crate, Or try out this Plastic shake weight, You are too late for your due date... Here's a healthy breakfast option, oops we broke your phone, Here's why Apple steals your data, and the apps you never owned, How can AI help you? Take this quirky quiz! ChatGPT just told you ICE is coming for your kids! Could I interest you in AI functions all of the time? A little bit of AI sprinkles, all of the time We're stealing all the power for our simulated mind AI into everything, all of the time ...
December 23, 2025 at 3:23 AM
gringene.org
1. Except... the tolerance *is* included in the formal proof (or to be more correct, the sketch of the proof that is included in the main text of the paper). They discuss the use of a probabilistic method, rather than a discrete result:
After running the subroutine that simulates M, the algorithm will evaluate the quality of the resulting learned algorithm A by using additional samples from D and counting the number of situations s in which the algorithm A returns an appropriate behaviour b∈Bs. As this evaluation only involves executing A and counting, it can be done in polynomial time. The algorithm runs the simulation subroutine several times, and for each run of the subroutine, it evaluates the resulting learned algorithm A. From all of these runs, it picks the algorithm A⋆ that achieved the highest accuracy in the quality evaluation. Based on how well the most accurate learned algorithm A⋆ performs, the algorithm will give an answer for the input of Perfect-vs-Chance. If A⋆ performs non-negligibly better than chance, then the algorithm will answer Yes, and otherwise, the algorithm will answer No. --- By having the algorithm run the simulation a large enough (yet polynomial) number of times, and testing the output of each simulation a large enough (yet polynomial) number of times with new samples, we can ensure that the algorithm outputs a correct answer for Perfect-vs-Chance with high probability. --- A careful probabilistic analysis shows that because the learning mechanism M performs non-negligibly better than chance on AI-by-Learning, the resulting algorithm for Perfect-vs-Chance gives the correct answer with high probability (see Appendix for full details). This contradicts the known intractability of Perfect-vs-Chance, hereby concluding the proof that AI-by -Learning is intractable.
December 22, 2025 at 6:58 AM
gringene.org
A "discrete result underneath the noise floor" does not appear to be the high-level actual demand. From the paper, it already assumes that there is some tolerance for accuracy:

doi.org/10.31234/osf...
AI-by-Learning (informal) Given: A way of sampling from a distribution D. Task: Find an algorithm A (i.e., ‘an AI’) that, when run for different possible situations as input, outputs behaviours that are human-like (i.e., approximately like D for some meaning of ‘approximate’).
December 22, 2025 at 6:58 AM
gringene.org
#AngelineEra I never thought before that sheep could be so deadly.

... well, except for that one Peter Jackson film.

... and the exploding sheep in Worms.

1021 deaths.

#HanTanisChallenge
Found Scale. There are a few sheep in the background, and an eye / spikey virus thingy, and an exploding raddish.
December 22, 2025 at 5:37 AM
gringene.org
#AngelineEra Getting a hitless run was nice. Not my first try, but it's still a good feeling. 1001 deaths.

#HanTanisChallenge
Tet smashes an orb at the end of a level, showing full health, and full ammo.
December 21, 2025 at 10:39 AM