mastodon 4.7.3Geek, Linux user, open-source contributor and developer. Former security researcher & consultant turned data engineer. Itinerant tinkerer and knowledge junkie.
Running websites since 1999. Grew up with Games Workshop (Warhammer 4th to 6th, 40K 2nd to 4th Ed, Mordheim, WHQ), but now just a reader and occasional painter. Single-player PC gamer.
Dreams: Big. Achievements: Small.
He/him.
(Disclaimer: Comments are my own, not my employer's, boost = interesting not endorsement of the person, etc)
Corey Quinn with the smackdown on AWS announcements again:
> Seven design principles for letting AI loose in financial services. I'd have gone with one: "Don't." But sure, comprehensive observability and fine-grained access controls sound great until an autonomous agent decides to rebalance a portfolio at 3 AM and your only explainability is a CloudWatch log nobody reads.
Does anyone have any good recommendations on things to put in an `llms.txt` file on your website? Apparently it's a pseudo-standard now and LLMs may read it to get a "better understanding" of your website before processing it.
My current content is here: https://ibboard.co.uk/llms.txt
I feel like Cisco and IBM kinda have a point here in that Random Joe Public probably shouldn't be doing "right to repair" stuff on Cisco switches in a data centre.
Most companies wouldn't want their techies screwing around with ad-hoc fixes on devices in the production network anyway! That's what support contracts and hardware upgrade cycles are for.
BUT they're also wrong, because the people who DO run Cisco switches should be able to hire whoever they want who is sufficiently trusted by the current owner. Not just a Cisco technician.
Also, some geeks run them at home (especially when they're being thrown away by a company) and _they_ should be allowed to do whatever they want with them in their home labs. Even if it's meant as enterprise hardware.
Some companies are a) trusting to "AI" agents instead of hiring real people (which couldn't possibly go wrong) and b) returning to metrics that are basically "Lines of Code" but more expensive!
I can't wait for all of the providers to start charging at a rate that covers their costs and watch these idiots crash and burn 🙄
https://www.404media.co/startups-brag-they-spend-more-money-on-ai-than-human-employees/
Anyone else got DMARC reporting set up and find that Google often sends their reports (with identical IDs and content) at least twice?
I got six copies today! 😐
Me at work: Here's an example of where Copilot performed "analysis" of data based on tags and pre-existing training instead of the (actually identical) data it was given. And then it bullshitted some statistics when probed. Be careful when you trust GenAI to do analysis for you.
Colleague: Everyone knows that Copilot is rubbish.
Me, in my head: Yes, that's exactly the message to take home here 🙄
Melissa Joan Hart is apparently 50 today.
She's still a witch who explains it all to me 🤩
Today in "teenage me would never have expected to agree so much with the pope":
I saw something recently about how we're "already cyborgs" because we outsource memory and calculations and stuff to computers and…
Does that mean that we've not really been "just human" for thousands of years? We've actually been human/animal hybrids for thousands of years since we started riding horses to improve our movement speed and started training dogs to perform herding tasks for us? 🧐
(We've presumably then spent a few thousand years being human/animal/reading material hybrids since we started writing things down on stone and then paper!)
As far as I can tell, the argument here is that China is winning the "making machines that do useful and practical things" race while the USA is winning in the "dumbing down society with untrustworthy text generators" race.
Well done, USA? 🤨
Kinda cool (from a purely technical standpoint) that US export controls enforced on foreign companies required China to be creative and efficient, though. Constraints create creativity!
There's FINALLY a sensible suggestion from a one of the relevant government-related policy organisations about how to deal with kids and the Internet!
Call for parents to teach their children online privacy like they would road safety - BBC News
Ah, the joys of opening a game case for a console and finding that the game isn't inside. And isn't in any of the dozens of other places that you've then looked for it 😑
I miss the days when PC games just came on a disk, you installed it and that was it. I'm sure there was a point where you didn't even need to have the disk in the drive (or No-CD cracks were still fairly safe to download 😁)
The worst part about it being 1st April is that we've already had three months of the year and WHERE DID THEY GO?! 😭
Ugh. Some of my website config and defences are getting a bit gnarly and convoluted. I kinda want to tear it all down and start again.
Except that a) I don't have the time for that and b) it won't get any better if I don't have better testing, and I don't know how best to optimise and test most of the problem areas 😐
(Example: Big configs of SpamAssassin rules and contact form filtering and script kiddy bot blocking)
BBC News: Team Scotland condemns 'abhorrent' racist abuse aimed at its models
Will you look at that. Anonymous bigots online are being arseholes, and even Sharron Davies has weighed in with an uninformed view on "feminism". Who could have expected that from a transphobe? 😐
(The skirt isn't that short, there's a longer one available, athletes are choosing the shorter one, and the model was probably wearing the heals so that she wasn't TINY next to the tall guy!)
BBC News: Businesses scramble to get noticed by AI search
"Rather than having extensive pages of content, HubSpot are now creating smaller pieces that GenAI can ingest"
Much like the move to video on social media, it feels like HubSpot are butchering their own content just to fit in with the trendy new tech that's overall WORSE for users!
I see articles like "What Claude Code ACTUALLY Chooses" (via Last Week in AWS) and I wonder… how can ANYONE with even half an ounce professional integrity trust these GenAI systems?
The behaviour is inconsistent. It rolls its own auth without a second thought. Sometimes it's absolutely dead-set on ONLY using the market leader and sometimes it's dead-set on using the latest who-knows-if-it'll-last library (and sometimes it flips that behaviour between model generations).
I'm sure someone will want to whine about "but they gave it an open prompt without guidance". But that doesn't FIX anything in its behaviour. It just makes some of the problems less obvious.
OMG, Kagi Translate as "LinkedIn Speak" as a target language 😆
Which apparently translated to "Thrilled to announce that Kagi Translate is officially disrupting the localization space! 🚀 Grateful for the opportunity to leverage "LinkedIn Speak" as a target language. 😆 It’s all about driving impact and fostering global synergy. #Innovation #GrowthMindset #Localization"
(Probably actually just "LLM: Generate me some corporate garbage", but it's very apt)
I wish version control CLIs had a "write commit message with diff below" option.
Yeah, sure, the list of modified files is helpful. But sometimes I've made multiple edits and it would be helpful to have the full diff available so that I can write a better summary!
(I should probably just have another tmux 'tab' and switch between the two… but I generally only keep one tab per folder most of the time!)
My parents asked if I'd watched the Hannah Fry documentary on "AI". Tonight I tried to watch it.
Normally I quite like a Hannah Fry documentary. She's a mathematician who can engage without being too deeply academic (unlike de Sautoy and al Khalili - who I also like).
But this? I should have given up when I saw the episode titles.
10 minutes in and I couldn't take any more. So much "it's so amazing and transformative" and "he exchanged FIVE THOUSAND MESSAGES with his chatbot girlfriend" (that's, what, a week of nightly chat sessions given the length of messages? And it was 2021 so we're talking "a bit better than Eliza" 🙄). And then the Replika interview was just… ugh. What crap.
I *love* it every time I get to work with Polars (like Pandas, but wrapping a Rust library) or Spark. The power and cleanliness is amazing. Such clear and efficient transforms on data 😍
I wish I needed to use it more often. And I wish I knew more of it so that I didn't have to keep digging around and finding what the right function is again.
I've probably got Pandas and Spark books somewhere in the 170+ software-related ebooks that I have from Humble Bundles. But… I just don't have a use case to make it worthwhile in my spare time.
We're going to be hiring someone new. And as the senior Data Engineer, I'm going to get to lead the recruitment.
I'm currently working out how I balance my response if a candidate asks about GenAI coding tools.
I don't want to burst out laughing. But I also don't want to run away screaming 🤔
(This is Data Engineering, Sir/Madam. We care about data quality, consistency and robustness in these parts)
Which product manager at @mozilla@mastodon.social needs to be fired for having them waste time on this rather than focus on their core responsibility: the web and email? 😒
"Our system for automatically picking targets and recommending them to military planners should not be seen as an automated targetting system" says head of Palantir UK 🙄
Seen via Corey Quinn's "This Week in AWS" with the following intro:
"Turns out you can't just announce a data center into existence. Nearly half of US data centers planned for 2026 are getting delayed or canceled because nobody stockpiled enough transformers and circuit breakers. The supply chain for boring electrical components is the actual bottleneck choking the AI buildout. Who knew infrastructure required, y'know, infrastructure."
😆
https://futurism.com/science-energy/data-centers-construction-supply
Apparently, if the on-site refund won't work for a cancelled train (because it's an Advanced ticket and there's normally no refunds _for change of plans_) then The Train Line's support options are phone call, WhatsApp and Twitter.
And by "support options", I mean "a WhatsApp chat bot" that'll quickly say "sorry, you need a human" and give a 3-5 minute wait time before they black-hole you and don't reply for at least 40 minutes (and counting) 😐
Who designs these shit interfaces?
Playing "Overlord 2". One of the loading screen tips was "Never take a succubus home to meet your mother".
It also suggests boiling a phoenix in oil instead of cooking/baking it, because the direct heat only encourages it 😁
It bothers me slightly that Overlord 2 says "Autosave" but it doesn't mean "save here". It means "saved your progress". So when you go back to the game then you go _all_ the way back to your lair, have to pick the mission from the map again, get dumped by the last portal, have to re-recruit your minions, and then wander however far it is into the level between your entry portal and where you were actually at.
All of the progress saved. All of the main objects and obstacles are moved/destroyed/etc. But you're not actually saved where you were.
"George Russell says Formula 1 “would not want to lose” Max Verstappen - but he would “understand” if the four-time world champion decided to leave the sport this year."
That feels like a bit like a coded "Please leave! I'll be glad when my role as director of the Grand Prix Drivers' Association no longer involves listening to your self-centred whining" 😆
I do love the snark in Last Week in AWS 😆
"Automate AWS Lambda Runtime Upgrades with AWS Transform custom - Imagine spending 20-30% of your engineering effort manually upgrading Lambda runtimes when you could instead spend that time learning a new AI tool that does it for you, poorly, and then fixing what it broke. AWS Transform custom: because technical debt needs its own technical debt. As a bonus, a quick look at the Transform pricing page indicates it will also incur actual debt."
🙃
