Thanks Nexa Center for the invitation and thr great discussion on the Minimal Cognitive Grid and its applications! It was a blast: nexa.polito.it/mercoledi-172

#mimimalcognitivegrid #cognitivedesignforartificialminds #ciitlabunisa CIIT Lab CognitiveDesignForArtificialMinds

172° #nexawednesday
“ #Cibernetica : un apparato metodologico che può evitare di farci commettere errori interpretativi in merito all’Intelligenza Artificiale”, Antonio Lieto @[email protected] (@UniSalerno CIIT Lab)
➡️Partecipa online: didattica.polito.it/VClass/NexaEve…

172° #nexawednesday
“La #MCG tiene in considerazione tre variabili quantitativamente misurabili e non soggettive: il rapporto tra componenti funzionaliste e strutturaliste in un dato sistema, la sua generalità e i risultati di performance nel tempo”, Antonio Lieto @[email protected] (@UniSalerno)

172° #nexawednesday
“I principali sistemi di #IA recenti, come i #LargeLanguageModels , sono di tipo funzionalista: non seguono quindi l’ispirazione cognitiva di stampo biologico”, Antonio Lieto @[email protected] (@UniSalerno e CIIT Lab)

❗Oggi ore 17.00
172° #nexawednesday
“Evitare la trappola comportamentista e la confusione tra sistemi human-like e human-level in #IA con la #MinimalCognitiveGrid ”
➡️Con Antonio Lieto @[email protected] (@UniSalerno)
📍Centro Nexa e online: didattica.polito.it/VClass/NexaEve…
INFO: nexa.polito.it/mercoledi-172

Thanks Association for Computing Machinery! Please note that my new affiliation is with Università Salerno!

🗓️Mercoledì 8 maggio, ore 17
172° #nexawednesday “Evitare la trappola comportamentista e la confusione tra sistemi human-like e human-level in Intelligenza Artificiale con la #MinimalCognitiveGrid ”
👉Con Antonio Lieto @[email protected] (@UniSalerno)
ℹ️ nexa.polito.it/mercoledi-172

#AI #LLM #umanità

Testing day at the CIIT Lab (ciitlab.org)! We just started an extended evaluation campaign (based on several linguistic & cognitive dimensions) of a fine-tuned dialogue-assistant based on #LLMs (GPT Generative-Pretrained Transformer)! Stay tuned for more!
#AI

How to be as 'smart' as Auto-Regressive LLMs:
- memorize lots of problem statements together with recipes on how to solve them.
- to solve a new problem, retrieve the recipe whose problem statement superficially matches the new problem.
- apply the recipe blindly and declare…

Science goes on, bringing us another harvest of exciting papers in causal analysis: ucla.in/44pmJew
Among them my eyes fall on 'The development of human causal learning and reasoning' by Goddu and Gopnik, nature.com/articles/s4415…, which provides a comprehensive review of…

Molto onorato di aver ricevuto l'invito a tenere questo intervento (mercoledì prossimo, 8 maggio) al Politecnico di Torino (@DAUINatPoliTO) da parte del Nexa Center.

Università Salerno CIIT Lab

Link: nexa.polito.it/mercoledi-172

ChatGPT overuses certain words, notably 'delve'

Why?

Turns out it was trained by RLHF workers from Africa

In Nigeria, “delve” is much more frequently used in business English than in England or the US. So we are ending up with an AI system that writes slightly like an African

😧 How on earth could you say this, Mustafa Suleyman @microsoft?

LLMs and related techniques have been used to
• influence elections
• influence markets
• generate nonconsensual deepfake porn
• generate so much child porn the NCEMC is overwhelmed
• write fake books,…

Daniel Dennett (RIP) on LLMs:

'[..] their goal is truthiness, not truth. They are more like historical fiction writers than historians.'

Not a bad way to remember the form vs. content conundrums surrounding GenAI.

No. Training LLMs on purely factual data STILL WON'T cure them of 'Hallucinations' #SundayHarangue

There is a persistent myth that LLM hallucinations are just a result of them being trained on un-curated and 'non-factual' data, and will go away with high quality/factual data.…

Look forward to discussing with Roberto Navigli about the current limitations of #LLM s and the efforts of the research community to overcome them. Thanks to Università di Napoli L'Orientale for the event 'Human in the loop: human oversight of AI systems'
Link: unior.it/sites/default/…

#AI #LLM

Come on, New Scientist, “deep-learning neural networks” are not “systems modelled on the way brains work”.

They are organized piles of matrix multiplication trained on amounts of data no human is ever exposed to, with only the vaguest connection to how actual brains work.

So many people are confused about the relation between human cognitive errors and LLM hallucinations that I wrote this short explainer:

Humans say things that aren't true for many different reasons
• Sometimes they lie
• Sometimes they misremember things
• Sometimes they fail…

That wonderful phrase, 'the space of possible minds', comes from a 1984 paper by Aaron Sloman. I've used it a lot in my work, for example in my 2016 'Conscious Exotica' paper: aeon.co/essays/beyond-…

We've been looking quite a bit lately at measuring the language-understanding ability of LLMs. This is an arxiv preprint on this line of work (w/ Reut Tsarfaty Victoria Basmova).

arxiv.org/abs/2404.06283