Ex-academic. Forever a computational semanticist. I care about meaning. I also care *for* meaning, here for instance: https://xitter.cf/t.co/fic4erWFIH.
Planet Earth
Joined October 2017
- Tweets431
- Following285
- Followers800
- Likes450
Here is an update on our NGI Search project, which integrates a private cloud solution (On My Disk) with a local search engine (PeARS). All open-source, of course!
@devnullxff and I were invited to tell a bit more about our project in this interview: spaces.fundingbox.com/spaces…
Chers Francophones, au cas où vous aimeriez faire une contribution à la science *et* vous amuser, jetez un coup d'oeil sur ce projet de mes collègues à l'université de Trente, qui combine l'étude des émotions et celle de la prosodie: actorschallenge.eu/ Merci!
Hello Gothenburg! Looking forward to the #LSD conference starting tomorrow! I'll be talking about breaking monolithic systems and decentralising learning... with small data of course 🙂
sites.google.com/view/learni…
I'll be sharing a stage with @EvRedecker at the re:publica in Hamburg on 23.09! We will talk about the notion of freedom in relation to sustainability, and ask whether our happiness is to be found here on Earth, beyond our planet, or (as big tech currently claims) on a GPU...
This quoted post is unavailable.
New post on Denotation IO! Casey Kennington explains how natural languages are intrinsically linked to the real world, and how humans naturally ground symbols in their physical environment... in stark contrast with LLMs.
denotation.io/blog/meaning-a…
A 🧵 on understanding, language and worlds, prompted by some tweets I saw this week. Disclaimer: I am *not* at the ACL. I am lying low in the woods and mostly getting my news from charitable friends in Toronto. /1
Do we really want to conjure up an entirely different mechanism for innovative thinking and creative language? I'll go get my Occam's razor and argue that we don't. So let's not just test understanding on the ability to regurgitate behaviours. /10
Aurelie Herbelot (@[email protected]) retweeted
Once upon a time, a group of developers set out to build a decentralised search engine, so small and efficient that it would run on any computer. They called it 'PeARS, the people's search engine'. Year after year, they occasionally tinkered with their code, very very slowly.
Aurelie Herbelot (@[email protected]) retweeted
v excited to finally announce our new work that formalizes one of the most effective practices for training LLMs—something that many industry leaders have conspicuously avoided discussing
This week on Denotation IO, 'Infinity is not Everything'.
denotation.io/blog/infinity-…
Looking at the linguistic and logical processes behind creativity, and asking to what extent they are implemented by neural net architectures.
TLDR; ⬇️ /1
Creative abduction however requires the ability to very literally 'think outside of the box'. Neural nets are trained to make inferences within a particular space of fixed dimensionality. This considerably reduces their ability to do any kind of truly interesting abduction. /6