James McDermott, Senior Lecturer in Computer Science in University of Galway, Ireland. I'd rather write programs that write programs than write programs. Was https://twitter.com/bleepbeepbzzz
The HUMIES are the annual competition at #GECCO for human-competitive results created by evolutionary computation methods.
***Ten thousand dollars in prizes for human-competitive results***
Submit by Friday May 29, 2026
See www.human-competitive.org for more
New special issue of GPEM on Evolutionary Computation in Art, Music and Design!
Edited by Penousal Machado and Juan Romero
https://link.springer.com/article/10.1007/s10710-025-09519-4
GPEM journal has a new special issue on "twenty-five years of grammatical evolution"!
Edited and with an introduction by Mahdinejad, Murphy and Ryan.
Special issue: https://link.springer.com/collections/ifgcbejghh
Introduction: https://link.springer.com/article/10.1007/s10710-025-09512-x
Papers follow:
@StompyRobot @helvetica the good thing about Twitter is a person from Google search showed up in the thread to explain how confused and wrong the Wired writer is.
Spawning salmon were trapped in this lake in the Yukon by a landslide long ago. After thousands of years, and thousands of generations, they still migrate from one end of the lake to the other, trying to reach the sea.
.
I still feel a strange urge when the leaves start to turn, like I'm trapped and need to go outside.
My son (9) has the memory of a large distributed and multiply-redundant system of elephants
Automating Away What is Dangerous, Dirty, Dull and Difficult - and #Drumming
https://jmmcd.substack.com/p/automating-away-what-is-dangerous?sd=pf
Does anyone manage to make #Chatgpt succeed at this kind of riddles? I've been giving it quite a hard time!
@lisperati hi! Are you going to post dog bunny puzzles on here?
@niemanlab However, I think the "friendly-fire" uses of ChatGPT will be just as influential as adversarial uses.
A lot of people who currently write derivative blog posts with little quality control will now have a slightly easier way to do the same. But gaining attention becomes harder, and *stable online reputation* more valuable. Posting vague or unreliable GPT-text does little damage to reputation now but that will change, I hope.
The Future of Trust: "we will see ChatGPT and tools like it used in adversarial ways that are intended to undermine trust in information environments, pushing people away from public discourse to increasingly homogenous communities"
Janet Haven, @niemanlab
https://www.niemanlab.org/2022/12/chatgpt-and-the-future-of-trust/
@mshanahan the paper tries to be neutral perhaps, but IMO that's not sustainable! Because some of the distinctions the paper draws, putting current LLMs on the "no REAL understanding" side of the line, can be so easily bridged.
It's 2048. A busy time in the PI business. It's Friday night, and you're a quarter way through the bottle under the desk. In walks a tall, well-dressed dame-bot. She sits down and lights up. It's been a long week. But you mostly work weekends.
My new paper is a critique of the way we talk about large language models.
https://arxiv.org/abs/2212.03551
I argue that LLM-based systems are "simultaneously so very different from humans in their construction, yet ... so human-like in their behaviour, that we need to pay careful attention to how they work before we speak of them in language suggestive of human capabilities and patterns of behaviour"
For example: the VLM (just needs an unfrozen distribution) and the LLM + wiki lookup system.
Separately: "But if we stuck with a pure LM, the only way to fully trust the arguments it generated would be by reverse engineering it and discovering emergent mechanisms that conformed to
the faithful reasoning prescription." - but then how can we trust human arguments? Don't we just evaluate them line by line?
@mshanahan I enjoyed this paper. I found myself disagreeing a lot, but probably if you'd written the opposite in many cases I'd still be disagreeing.
As a general comment, in several places the distinction you draw between current models which don't "really" understand and similar models which *would* is a very fine one which could be bridged with a few lines of code! So you see "real" understanding as imminent?
I've bieng playing around with pixel sorting and it's super fun.
Thanks to IPython, I'm able to change values around and fine-tune at low resolutions!
Anyone thinking that things are better planned at Tesla needs to have their brains checked.
$TSLA $TSLAQ
---
RT @ericfreyss
Any attempt to remove my tweets that link to my other social media accounts, not violating any law, would actually make #twitter an editorial media, and no longer a social media platform, with civil and criminal liability for *any* illegal content therein. https://help.twitter.com/en/rules-and-policies/social-plat…
https://twitter.com/ericfreyss/status/1604538198054535169
It would be great if mastodon displayed mutual followers under accounts (“this person is followed by ___ people you follow”). It’s quite helpful for finding interesting people to follow, and also checking if someone is who you think they are or just shares the same name.
