Alessandro Bozzon

Fri, 14 Jul 2023 15:04:15 +0200

The Hidden Workforce That Helped Filter Violence and Abuse Out of ChatGPT - The Journal. - WSJ Podcasts

Thu, 13 Jul 2023 21:05:10 +0200

“Low-Resource” Text Classification: A Parameter-Free Classification Method with Compressors - ACL Anthology

gZip and kNN beat deep learning approaches in out of distribution settings…

Thu, 13 Jul 2023 07:00:51 +0200

But last week I thought about the death of social networks as a hardware and energy problem: You’re storing huge amounts of media on these huge server stacks costing massive overhead on the off-chance that 0.00001% of it goes viral and propels a bunch of ads into a couple of million people’s eyes. The input/output formula just doesn’t work. The dream of converting eyeballs into $$$ has always been promissory; that one day someone would crack the perfect formula and balance the scales and so the investment had, until recently, kept flowing

Wed, 12 Jul 2023 07:19:10 +0200

About to arrive in #sardinia. Even after ~15years, every time it feels like the first time

Thu, 06 Jul 2023 23:59:58 +0200

And then a good afternoon of fun and bonding with the SDE department. I am so proud of all of them

Thu, 06 Jul 2023 23:59:13 +0200

Celebrating an extraordinary first half of 2023 with the SDE department. So much to be proud of!

Wed, 07 Jun 2023 12:10:22 +0200

Invisible Indirect Injection: A Puzzle for ChatGPT

Wed, 07 Jun 2023 09:17:39 +0200

And the 23rd edition of the ICWE conference officially starts. A lovely scientific community that I first joined in 2006. And this year hosted in the charming Spanish city of Alicante.

The Critical Engineer

Mon, 05 Jun 2023 09:48:42 +0200

An interesting manifesto from The Critical Engineering Working Group . I tried to substitute The Critical Engineer with The Designer, and it reads even more spot-on

criticalengineering.org

Mon, 05 Jun 2023 08:41:06 +0200

Only when plumbing will be automated, machine will actually have taken over the world

plumbing is now the Turing test that matters

Sat, 03 Jun 2023 08:55:21 +0200

On the existential risks of AI

We should be wary of Prometheans who want to both profit from bringing the people fire, and be trusted as the firefighters.

Sun, 14 May 2023 13:53:44 +0200

A good explanation why prompt injection in language models is a serious and tricky issue Prompt injection explained, with video, slides, and a transcript

Sat, 06 May 2023 22:27:24 +0200

www.linkedin.com

Sat, 06 May 2023 16:32:21 +0200

Mandatory picture in front of the #CHI2023 sign

On AI Alignment

Sat, 06 May 2023 16:27:04 +0200

If social systems are inherently unpredictable because of complexity (and they often are), AI is gonna have a hell of a time trying to align to that…

twitter.com