Martin Ruskov @mapto@social.cleverlibre.org
- Research
- https://www.zotero.org/mapto
- Games
- https://mapto.itch.io
- Account in Bulgarian
- https://masto.bg/@mapto
Studying how people interact, in the past (#CulturalAnalytics) and today (#EdTech #Crowdsourcing). Researcher at @IslabUnimi, University of Milan. Bulgarian activist for legal reform with @pravosadiezv. I use dedicated accounts for different languages.
My profile is searchable with https://www.tootfinder.ch/ #fedi22
Joined Nov 2022
Martin Ruskov
boosted
Go ahead. Tell me we're not the problem. (P.S. Eat less meat.)
Martin Ruskov
boosted
🚙 A car is parked 92% of the time.
🚗 A car spends 20% of its driving time looking for parking.
🚙 Its 5 seats only move ø 1.5 people.
🚗 86% of its fuel never reaches the wheels.
🚙 A road at peak throughput is only 10% covered with cars.
Sounds efficient ?
@muiiio no doubt about it
@muiiio https://alternativeto.net/ is to become your best friend... at least until you appreciate that the world of FOSS is much richer than commercial offering
RT @rao2z@twitter.com
Interestingly, I was just telling someone today how several of the papers on "LLMs for Task Planning by Prompting" are rife with the Clever Hans effect (c.f. https://en.wikipedia.org/wiki/Clever_Hans ). I guess I will have to do a thread.. 2/
@Riedl I see a Roomba in the upper right; am also supportive of any collective action against captcha
Martin Ruskov
boosted
History of the #IPCC.
(Cartoon from 2013 - back then the next, 6th, IPCC report was expected for 2019.)
@davidzipper bike share was cheap when it used plain old mechanical bikes. But cheap meant that it did not bring profits to vendors. So they forced e-bikes onto us all. Now you want us to encourage them with subsidies. I'll pass on this. I'd rather have the good old conventional municipality-managed bike sharing that is still around in many places in Europe.
Martin Ruskov
boosted
Do you reside in the USA? Were you a Facebook user between 2007-22?
Then Meta owes you money!
That's right, because Facebook has accessed user data without permission, various class action lawsuits have resulted in them having to pay a total of $725 million to its users.
And who are they paying? Almost every American who's used Facebook!
Want that money? Here's where to file that claim:
RT @BenedictDunant@twitter.com
Fascinating story by @EmilyFishbein11@twitter.com and @NuNuLusan@twitter.com about the dangerous hunt through Myanmar's far-northern mountains for a fungus that's worth more than gold
https://www.frontiermyanmar.net/en/in-kachins-icy-mountains-the-search-for-a-coveted-fungus-heats-up/
🐦🔗: https://twitter.com/BenedictDunant/status/1648148370211606528
RT @mcxfrank@twitter.com
What does it mean for a large language model (LLM) to "have" a particular ability? Developmental psychologists argue about these questions all the time and have for decades. There are some ground rules. 🧵
RT @B_MCristina@twitter.com
Perchè ti piace Python?
🐦🔗: https://twitter.com/B_MCristina/status/1647248237919625218
@Riedl it's long and there's a lot to unpack (I haven't finished reading it yet). From my perspective it would be little effort and huge explanatory value to also talk of word embeddings since you're already at encodings. In any case, I see a tendency of this to grow in time so maybe presenting it also as someone sort of hypertext (vocabulary/encyclopedia) would be beneficial. Just some thoughts
Martin Ruskov
boosted
"When I tell it it's wrong, ChatGPT tries again and gets more correct!"
No. It learns which answer you stopped asking questions at. It has no idea of correct factually!
Correct for ChatGPT means *you are satisfied* with the combo of words it put together. That you liked its information-shaped sentence.
That's all. If you tell it it's still wrong, it will try again with different words instead.
You just think it's getting more right because you let it stop when it accidentally is.
Martin Ruskov
boosted
Yes, you can #jailbreak #ChatGPT and get it to say things that it doesn't usually otherwise say.
But I'm baffled at how many people are doing jailbreak experiments with the impression that they're learning about what the #LLMs *really* thinks or what it's *really* doing on the inside.
To illustrate, I've slightly tweaked one of the classic jailbreak scripts https://www.reddit.com/r/GPT_jailbreaks/comments/1164aah/chatgpt_developer_mode_100_fully_featured_filter/
Do think you learn much about its inner workings from this?
Martin Ruskov
boosted
How long until "we must fight climate change because our AI data centers need water for cooling"?
https://gizmodo.com/chatgpt-ai-water-185000-gallons-training-nuclear-1850324249
RT @RSprugnoli@twitter.com
1st workshop on Ancient Language Processing (at RANLP-2023, Varna Bulgaria)
Paper submission due: July 3, 2023
Date: September 7, 2023
Website: https://www.ancientnlp.com/alp2023/
#NLProc #digiclass @ranlp2023@twitter.com
🐦🔗: https://twitter.com/RSprugnoli/status/1640995540920856576
Martin Ruskov
boosted
Latest comic: How do oil execs sleep at night?
#climate #climatechange #climatecrisis #environment #cartoon
- Research
- https://www.zotero.org/mapto
- Games
- https://mapto.itch.io
- Account in Bulgarian
- https://masto.bg/@mapto
Studying how people interact, in the past (#CulturalAnalytics) and today (#EdTech #Crowdsourcing). Researcher at @IslabUnimi, University of Milan. Bulgarian activist for legal reform with @pravosadiezv. I use dedicated accounts for different languages.
My profile is searchable with https://www.tootfinder.ch/ #fedi22
Joined Nov 2022
