Newest thoughts →

July 2024

Top tip: if you want to parse SQL dumps at speed in python - say for conversion to JSON for Elastic ingest - use sqloxide. I got a 2hr 30min parse with sqlparser down to 6 mins. It uses a Rust binding.

Truly disturbing but an increasingly common view.

"A few months ago a “friend” told me that he felt disabled people should be denied access to healthcare so “survival of the fittest” could happen faster." https://threadreaderapp.com/thread/1816315498746753350.html

"Said that being forced to mask was the most egregious violation he had experienced" but doesn't care about the "egregious violation" of infecting others. Absolutely incredible.

Was having a frustrating problem of tf-idf vectorization and cosine similarity producing high similarity scores on very different (long) documents. Stopwords were the answer. #facepalm

Continuing my exploration of all modern document similarity methods. So far, it seems that the AI/LLMs are slow, expensive, and don't necessarily do as good a job as much older algorithms, depending on what you need to look for in "similarity" (the AI is good at, say, synonyms).

Today is physiotherapy, dialysis, and immunoglobulin. I continue to try to look on these as treatment blessings that I am lucky to receive, although each of them is difficult, time consuming, and unpleasant in various ways.

Boycott Routledge over AI training? It's probably in the contract, which allows unlimited worldwide reproduction "in any format". https://eve.gd/2024/07/20/boycott-routledge-over-ai-training-its-probably-in-the-contract

I think the other side of this is that academics think that they have an "unspoken" agreement with publishers and that the contract is just a formality. A note: you don't. The contract is the be-all and end-all. Particularly with a for-profit commercial publisher.

I've been set weekend "homework" by the nurses at the hospital to "watch videos of people self-needling for dialysis". This is gonna really screw up my YouTube recommendations list...

A day off work today. Think I'll treat myself with a trip to the hospital to continue learning to needle myself. Then we have the dog behaviourist coming to see Luka. Hope she can advise.

"Behind every man now alive stand 30 ghosts, for that is the ratio by which the dead outnumber the living" - Arthur C. Clarke. A statement that analogously shows the problem for archival preservation systems. The afterlife of "dead" documents will be longer than their lives.

Today I implemented Broder et al., ‘Syntactic Clustering of the Web’, Papers from the Sixth International World Wide Web Conference, 29.8 (1997), pp. 1157–66.

Despite its age, it does a good job of clustering "roughly similar" documents. Next up: vector space techniques.

This afternoon, I began research into the state of the art (and history) of document similarity detection as part of a move on content drift. Wondering also whether various authorship algorithms might raise the same signals.

I wrote a chapter, in English, for a German publication. Now they keep asking me to sign a contract in German, which I cannot read. But I am being clear: I will not sign a contract that I do not understand and cannot read. This is a basic legal principle.

One of the frustrating things about home haemodialysis is setting everything up if you're tired or feeling under the weather. It takes about 45 minutes to get it ready and I sometimes think: I really wish a nurse could do this!

It seems pretty mean that the winner of wheelchair tennis singles at Wimbledon gets pretty much the same prize money as a first round singles player.

Oh dear. I have overdone it. I can't walk today, at all. And that's not an overstatement. I really can't take a single step. Supposed to be at the hospital today. Have cancelled. We were planning a pub lunch outside too on a day off work - but I can't even get to the car.

Well, I did it! I managed to get a needle into the fistula. I failed the second needle - got the angle wrong - but the dialysis nurse says that I made great progress. Some patients can't even put the needle in on the first try, apparently.

Well, here goes. Today I have to start learning to put my own needles in. They're inch-and-a-half long monsters; so long that you get given an injectable anaesthetic before they go in! Anyway, I'm nervous, despite injecting myself sub-cutaneously for years.

Interestingly, ChatGPT can solve and explain some cryptic crossword clues.

But its description of the wordplay is massively out

At #sharp2024 and hearing Ellen Forget talk about OCR processes for Braille books. Many complications in training a model for this task.

Good progress at hospital today. We got 2x 1.5 inch needles into my arm and didn't use the central line, for the second time in a row. If we can continue this for 3 weeks, we can take the central line out of my chest. Next week, I will do a needle myself.

Tonight, on dialysis, I finished Tender is the Night, which I'd never read. I'm afraid I disagree with the Hemingway blurb and really did not enjoy it. I found it well, if slightly over, written but deeply tedious. #sacrilege

People often think that the scariest part of home haemodialysis is needling yourself. It isn't. It's the precarity of knowing that if you miss a session, you could kill yourself. Don't get sick because you might not be able to dialyse and then you have a heart attack.

Quick crosswordy lunch break

A classic in Private Eye this month:

Labour leader fared badly in hot temperature -- which he strongly discourages (4, 4)

Nightmare tech evening. Upgraded a remote linux server and it wouldn't boot, dumping to emergency console. Eventually hacked ssh and multi-user runlevel and ripped the contents off the box to rebuild it.

Laughing at the letter from my Conservative MP. Reads like a domestic argument: "I know you may be very angry at the Conservatives right now". Yeah mate, you didn't respond to any of my letters asking for your help as my supposed representative. So bye bye.