Trending
Dave's Avatar

Dave

@irksomelib

lib. postdoc. computer scientist (computational linguistics/natural language processing). british. likes metal, sci-fi and fantasy. extremely bad at posting. OCD-haver, ace/aro. he/him.

295
Followers
813
Following
430
Posts
15.06.2023
Joined
Posts Following

Latest posts by Dave @irksomelib

"Social Democratic Party":

13.03.2026 09:38 πŸ‘ 109 πŸ” 10 πŸ’¬ 3 πŸ“Œ 1

You know what, screw it, maybe the Tories should just go all in on embracing their inner dril

13.03.2026 12:23 πŸ‘ 0 πŸ” 1 πŸ’¬ 0 πŸ“Œ 0

Disgusting. Possibly even more cynical than my own institution which uses a subsidiary of all new staff to avoid paying a decent pension (I suspect they’ll try to do it to us all soon). The effect has been to deter all mid-career applicants but desperate young people and hourly-paid still come.

11.03.2026 09:49 πŸ‘ 55 πŸ” 23 πŸ’¬ 3 πŸ“Œ 0

"If you kill the boss, the victory cutscene automatically starts" -Machiavelli

10.03.2026 10:37 πŸ‘ 4 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

Quote skeet with a gif of a powerful woman.

08.03.2026 21:07 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

🎡 We carry the flame 🎡

08.03.2026 10:46 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

Why are so many VCs like this?!

07.03.2026 12:31 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

Pre-sliced bread is under a hundred years old.

Conversely, the first touch-screen phone was made all the way back in 1993.

07.03.2026 12:28 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

Shakespeare in Love. No idea why, I just hate that movie.

07.03.2026 11:29 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

TL;DR He joined Labour because the Greens weren't transphobic enough for him.

07.03.2026 11:20 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
Homer Simpson's Car Design

Homer Simpson's Car Design

Another downside will be a mad focus on battleship production (which tbf seems to be happening) and every future tank design will basically be the equivalent of...

04.03.2026 10:39 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

Yes, they would at least have a base understanding of logistics...are at the very least, understand that it is a term that exists.

04.03.2026 10:31 πŸ‘ 2 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

Plus I suspect a non insubstantial part of the administration are HOI 4-TNO fans, for exactly the reasons one would think.

04.03.2026 10:29 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

I dunno, there's just as big a risk they'd try to invade Russia via Alaska...

04.03.2026 10:28 πŸ‘ 1 πŸ” 0 πŸ’¬ 2 πŸ“Œ 0

Sad to report that genius can sometimes be used for evil.

03.03.2026 19:26 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
Statement by trade union leaders

Statement by trade union leaders

We stand with fellow trade unions in condemning the illegal war on Iran.

Reports of civilian casualties, including students and schoolchildren are deeply disturbing and require urgent and independent investigation.

The killing and wounding of civilians is indefensible.

02.03.2026 09:42 πŸ‘ 52 πŸ” 22 πŸ’¬ 1 πŸ“Œ 2

schools are never valid military targets, on or off base

if someone bombed a DoD school in Okinawa and killed a bunch of American 4th graders, that would be a war crime.

The murder of Iranian schoolchildren is a war crime.

01.03.2026 19:19 πŸ‘ 5349 πŸ” 968 πŸ’¬ 56 πŸ“Œ 17
Fake headline by Allister Heath saying "Why Kamelot's The Black Halo is the final nail in the coffin for Western values"

Fake headline by Allister Heath saying "Why Kamelot's The Black Halo is the final nail in the coffin for Western values"

Not everyone agrees however (last one I promise)

26.02.2026 13:59 πŸ‘ 0 πŸ” 1 πŸ’¬ 0 πŸ“Œ 0
Fake headline by Allister Heath saying "Why Kamelot's The Black Halo is the final nail in the coffin for Western values"

Fake headline by Allister Heath saying "Why Kamelot's The Black Halo is the final nail in the coffin for Western values"

Not everyone agrees however (last one I promise)

26.02.2026 13:59 πŸ‘ 0 πŸ” 1 πŸ’¬ 0 πŸ“Œ 0
The Haunting (Somewhere in Time) (feat. Simone Simons)
The Haunting (Somewhere in Time) (feat. Simone Simons) YouTube video by Kamelot - Topic

Just listened to Kamelot's The Black Halo for the first time a in a few months, reminded of what an absolute banger it is.

26.02.2026 13:57 πŸ‘ 0 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0
A fake Allister Heath headline saying "We must resist the totalitarian nightmare of it being impossible to win Ura Finale with Gold Ship before we lose everything"

A fake Allister Heath headline saying "We must resist the totalitarian nightmare of it being impossible to win Ura Finale with Gold Ship before we lose everything"

*nods in agreement*

26.02.2026 12:27 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

This is not going to assist my productivity.

26.02.2026 11:36 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
Post image

Remember folks, the AI is not *actually* doing this; it is querying its training-data for statistically-likely things to appear to do under the given conditions, and replicating the highest-probability outcome.

24.02.2026 11:35 πŸ‘ 257 πŸ” 55 πŸ’¬ 26 πŸ“Œ 38

Just curious: is there a book in your field that absolutely blew you away? That changed the way you thought about and approached your subject?

24.02.2026 10:15 πŸ‘ 10 πŸ” 5 πŸ’¬ 20 πŸ“Œ 5
The cover to the second edition of An Introduction To MultiAgent Systems by Michael Wooldridge.

The cover to the second edition of An Introduction To MultiAgent Systems by Michael Wooldridge.

An Introduction To Multiagent Systems by Michal Wooldridge was absolutely instrumental in shaping my thinking in my field. It also helped that he was an amazing lecturer!

24.02.2026 12:50 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
Post image

cant.stop.laughing

24.02.2026 06:18 πŸ‘ 2871 πŸ” 879 πŸ’¬ 14 πŸ“Œ 19

Investors are never beating the "most gullible motherfuckers on the planet" allegations

23.02.2026 20:39 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
Look, I’ve really tried to make a good faith attempt to understand what it is that Claude Code does and why everybody seems impressed, and the long and short of it is that I think the media is far too willing to extrapolate outcomes based on the present state of technology.

I’ll give you an example. CNBC host Deidre Bosa used Replit to create something called β€œMarket Bling,” a website that would track stocks and notify you when they hit a certain market cap. Extrapolating further, Bosa said that β€œthe question increasingly being asked is why pay software vendors at all when your own engineers are non-technical people [and] can just spin one up using Replit or Claude Code.”

This is a ridiculous and irresponsible leap of logic. Being able to spin up a simple web app that can track stocks β€” one that looks like every other vibe-coded project (because these models are all trained on the same code) β€” is entirely different to making functional, secure and scalable software.

Similarly, Bosa’s colleague created a β€œsewing community” called β€œPattern Haven.” When I went to join and post on the community, it made me create a Replit account, and then left me with the following error:


CNBC is one of the largest financial news outlets, and it is peddling nonsensical narratives that help fuel market sell-offs in software stocks by actively misrepresenting what these models and tools are capable of. No, nobody is replacing their Salesforce or Monday.com subscriptions with a fucking vibe coded app β€” you pay a software company a fee so that they maintain the software and they make sure it’s stable and they keep everything compliant and functional. 

The only people pushing this narrative are ignorant of the very basics of software, or those who are actively incentivized to lie.

Look, I’ve really tried to make a good faith attempt to understand what it is that Claude Code does and why everybody seems impressed, and the long and short of it is that I think the media is far too willing to extrapolate outcomes based on the present state of technology. I’ll give you an example. CNBC host Deidre Bosa used Replit to create something called β€œMarket Bling,” a website that would track stocks and notify you when they hit a certain market cap. Extrapolating further, Bosa said that β€œthe question increasingly being asked is why pay software vendors at all when your own engineers are non-technical people [and] can just spin one up using Replit or Claude Code.” This is a ridiculous and irresponsible leap of logic. Being able to spin up a simple web app that can track stocks β€” one that looks like every other vibe-coded project (because these models are all trained on the same code) β€” is entirely different to making functional, secure and scalable software. Similarly, Bosa’s colleague created a β€œsewing community” called β€œPattern Haven.” When I went to join and post on the community, it made me create a Replit account, and then left me with the following error: CNBC is one of the largest financial news outlets, and it is peddling nonsensical narratives that help fuel market sell-offs in software stocks by actively misrepresenting what these models and tools are capable of. No, nobody is replacing their Salesforce or Monday.com subscriptions with a fucking vibe coded app β€” you pay a software company a fee so that they maintain the software and they make sure it’s stable and they keep everything compliant and functional. The only people pushing this narrative are ignorant of the very basics of software, or those who are actively incentivized to lie.

Coding isn’t just banging out lines of Python or C++. Coding requires thought. To consider a whole bunch of things that go beyond whether an application runs or doesn’t. 

By definition, I suppose you could say that Claude doesn’t know how to code, insofar as LLMs don’t inherently know anything, nor do they β€œthink” in the biological sense. They’re making guesses based on the things they’ve seen in their training data. If it’s seen 20,000 codebases do a particular thing in a particular way, it’s going to spit out something that’s pretty much identical to that.

And sure, there won’t be much variation in how someone writes an integration test, or a unit test, besides the code that checks the thing they’re testing. And I imagine that I’d be able to trust Claude to spit out some fairly correct boilerplate code for a project. In my experiments with Claude Code in the past, I’ve asked it to do just that β€” to create the basic components of a Phaser.js game, including the file structure, the Node development server, and so on. 

This is the low-hanging fruit of software development. 

I’d argue that some of this work is stuff that would previously have been handled by IDE autocompletes, or (in the case of boilerplate code) by Git cloning someone else’s foundations. Hell, we’ve already kind-of had that in the past. If you were building an MVC app, say, using Ruby on Rails, you’d use Scaffold to generate the boilerplate code and then expand from there.

In this respect, LLMs don’t necessarily do anything new β€” except for allowing you to do it using human language, and perhaps with some customizations that you can add during the prompt β€” but that’s far from the point. It can do this stuff, and with a reasonable degree of reliability.

The problem is what happens when you stray from that low-hanging fruit.  Because Claude is simply repeating what it’s seen previously, it’s inherently thoughtless. A software developer might write code with a conscious aim to ensure it’s secure, …

Coding isn’t just banging out lines of Python or C++. Coding requires thought. To consider a whole bunch of things that go beyond whether an application runs or doesn’t. By definition, I suppose you could say that Claude doesn’t know how to code, insofar as LLMs don’t inherently know anything, nor do they β€œthink” in the biological sense. They’re making guesses based on the things they’ve seen in their training data. If it’s seen 20,000 codebases do a particular thing in a particular way, it’s going to spit out something that’s pretty much identical to that. And sure, there won’t be much variation in how someone writes an integration test, or a unit test, besides the code that checks the thing they’re testing. And I imagine that I’d be able to trust Claude to spit out some fairly correct boilerplate code for a project. In my experiments with Claude Code in the past, I’ve asked it to do just that β€” to create the basic components of a Phaser.js game, including the file structure, the Node development server, and so on. This is the low-hanging fruit of software development. I’d argue that some of this work is stuff that would previously have been handled by IDE autocompletes, or (in the case of boilerplate code) by Git cloning someone else’s foundations. Hell, we’ve already kind-of had that in the past. If you were building an MVC app, say, using Ruby on Rails, you’d use Scaffold to generate the boilerplate code and then expand from there. In this respect, LLMs don’t necessarily do anything new β€” except for allowing you to do it using human language, and perhaps with some customizations that you can add during the prompt β€” but that’s far from the point. It can do this stuff, and with a reasonable degree of reliability. The problem is what happens when you stray from that low-hanging fruit. Because Claude is simply repeating what it’s seen previously, it’s inherently thoughtless. A software developer might write code with a conscious aim to ensure it’s secure, …

Claude Code is one of the weirdest products I've ever seen, marketed and written about like the second coming of Christ based on its ability to recreate quasi-functional websites and apps from its training data and speed up writing code to an indeterminate end.
www.wheresyoured.at/premium-the-...

20.02.2026 18:56 πŸ‘ 55 πŸ” 4 πŸ’¬ 1 πŸ“Œ 1

Yep. They are literally fucking EGODYSTONIC which is why they are so bloody distressing.

23.02.2026 11:40 πŸ‘ 7 πŸ” 2 πŸ’¬ 0 πŸ“Œ 0

i do not enjoy, as a person with OCD, the idea that i should be held accountable for my intrusive thoughts

23.02.2026 10:25 πŸ‘ 166 πŸ” 9 πŸ’¬ 8 πŸ“Œ 1