Andres Guadamuz
LightNews LightNews
banner
technollama.bsky.social
Andres Guadamuz
@technollama.bsky.social
3.5K followers 920 following 1.5K posts

Law and technology academic. Posts about copyright law, internet regulation, AI, llamas, pandas, and cats. Dual Tico-British citizen.
https://www.technollama.co.uk/

Videos
Business 38%
Computer science 27%
Posts Replies Media Videos

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 4h
I used to enjoy spending time on Reddit, but nowadays I can feel my brains melting every time I go there, some previously thoughtful spaces have devolved into mindless meme war. Haw modern discourse degraded into endless memetic gotchas?

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 19h
Bad Bunny. Damn. O mejor dicho, carajo!

Reposted by Jonathan A. Eisen, Brian A. Nosek, Gavin L. Foster , and 20 more

theo.io
Theo Sanderson @theo.io · 1d
I made a map of 3.4 million Bluesky users - see if you can find yourself!

bluesky-map.theo.io

I've seen some similar projects, but IMO this seems to better capture some of the fine-grained detail
Bluesky Map
Interactive map of 3.4 million Bluesky users, visualised by their follower pattern.
bluesky-map.theo.io

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 2d
I notice Bluesky is having another normal one about AI (I could repost this twice a month and it would still apply),

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 3d
Really? Please elaborate.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 3d
Yes, I'm getting some sense that a lot of this stuff is automated, I mean, how did you find me in the first place? Can't you google? Even ChatGPT knows where I work 🤖

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 3d
I should add that he's communicating with me with my work email. I guess the sussex dot ac dot uk address is mystifying.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 3d
Person sends me a request for me to supervise him for a PhD at the University of Hertfordshire. I do not work there, so I tell him. He then replies, and I quote:

"Could you kindly let me know which university you are currently affiliated with?"

Not PhD material I'm afraid.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 4d
Biggest new tell-tale sign of AI use in law essays, the paper ends with a section entitled "Policy implications" or "Policy reform".

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 5d
I've been tempted...

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 5d
One trick is to see which references aren't picked up by Turnitin, most references have been cited before by someone else and are highlighted, those which are not may be hallucinated.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 5d
No resources to do stuff like that. There are a bunch of other assessment methods I may try in the future, but the essay is dead. I have to assume from now on that all essays are AI.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 5d
And finally, I haven't seen a single em-dash, it's evident that word got around that this was a common LLM trait.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 5d
Another interesting side effect, the overall quality of essays has gone up. Most years there would be a noticeable curve, with some extremely poor essays. I have yet to read a truly bad essay, AI has helped the worst students write better essays (or wrote it for them).

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 5d
Something I noticed is just how much the LLMs have improved. Long gone are the days of "delve", "intricate", "complexities" and other common words. Reasoning is more sophisticated and harder to spot.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 5d
But the biggest sign was the prevalence of hallucinations. At least 20% of essays had hallucinated references, and those are the ones I caught, I'm sure a few more may have escaped. This is the practice I punished more with points deductions as I warned them repeatedly about it.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 5d
How do I know? I ran my essay questions through ChatGPT, Claude, Gemini, Deepseek and Qwen, so I know what the LLMs will answer, and I could know with high degree of confidence which model they used. There were also some quirks, ChatGPT kept adding US law concepts to a UK question.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 5d
I've now marked enough essays this year that I can make a good assessment of where we stand with AI use. I'm pretty sure that a vast majority of students are using AI in some shape or form, but the quality of the essays has also improved in average if I'm marking the works as they are.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 7d
Home with the flu, how can my head hold so much mucus?

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 8d
Fun, not my first time, teaching an introduction to the technology for my IP PG module.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 8d
Me during the first week of teaching AI.
a close up of a man 's face with the words i 've seen things you people wouldn 't believe
Alt: a close up of a man 's face with the words i 've seen things you people wouldn 't believe
media.tenor.com

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 9d
Sony Music of course!

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 9d
My first hallucinated case of the marking season! Did you know that Sony Music sued Napster UK in 2000 and that the case was appealed in 2001?

Reposted by Andrés Guadamuz, Jason A. Heppler, James A. Benn

djlink.bsky.social
David Amador @djlink.bsky.social · 12d
recorded a Windows 95 full disk defrag to soothe your timeline.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 12d
You've hit on the problem, the work isn't being reprinted or communicated to the public, it's added to billions other works to train a model, which is not a derivative of the work by any theory of copyright.
Judgements across the world are starting to find that.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 12d
My last FOSDEM was Amsterdam 2013, wow, time flies.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 12d
Is there anything more satisfying than sending out final edits on an article or a book chapter?

I know, I lead a sad life.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 13d
We don't have rulings in other countries on this point, but we have art4 CDSM in the EU. Also private copying is aloud, and the copies aren't communicated to the public, information is extracted from them.

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 13d
"The court also agreed that buying and digitizing books is transformative because “every purchased print copy was copied in order to save storage space and to enable searchability as a digital copy.”
www.whitecase.com/insight-aler...
Two California District Judges Rule That Using Books to Train AI is Fair Use | White & Case LLP
Two days apart, two judges in the Northern District of California decided on summary judgment that two examples of using copyrighted works to train AI models were transformative, and ultimately fair u...
www.whitecase.com

technollama.bsky.social
Andres Guadamuz @technollama.bsky.social · 13d
I try my best here. www.technollama.co.uk/a-short-guid...
A short guide to the Copyright Wars
My social timeline has been abuzz with the latest copyright decision regarding the Internet Archive, which lost its appeal in the copyright lawsuit from Hachette. In the words of a colleague, this …
www.technollama.co.uk