Embedded Files
PROFILE
• Journalist with over twenty years of experience. Supervised 30+ editorial staff. Covered international summits and European Councils. Currently at RAI television.
• AI researcher at Hermes, a joint research group at VUB and University of Namur focused on explainable language models, i.e. on how to let machines learn language in a way more similar to the way humans do.
• MSc in Data Science and AI, MA in International relations, MA in journalism.
STORIES
A selection of the stories I wrote and recorded for the Italian National television RAI (in Italian) about AI:
• "Innamorarsi di ChatGPT? Nessuno può competere con un partner disponibile 24 ore al giorno"
(“Falling in love with ChatGPT? No one can compete with a partner available 24 hours a day” - interview to Anthony Elliot, Professor of Sociology at the University of South Australia)
• Centro per l'intelligenza artificiale, dove sarà e cosa farà
(“Center for Artificial Intelligence, where it will be and what it will do”)
• Intelligenza artificiale, i timori dell'esperta
(“Artificial intelligence, expert's fears”)
• Macchine che sbagliano a contare, come ChatGPT replica gli errori umani
(“Machines miscounting, how ChatGPT replicates human errors”)
• L’intelligenza artificiale deve anche saper dire ‘non so’
(“Artificial intelligence must also be able to say 'I don't know'”)
• L'esperto di intelligenza artificiale: "Alcuni modelli non distinguono il falso dal vero"
(“Artificial intelligence expert: "Some models do not distinguish false from true.”)
• L'Intelligenza Artificiale impara a leggere le emozioni
(“Artificial intelligence learns to read emotions”)
SIGNIFICANT WORKS
• My interview to the former Italian prime minister and former EU Commissioner Mario Monti (11 June 2024): https://www.youtube.com/watch?v=XIt7ryy3-o4&t=110s
• My interview to former president of European Commission and former Italian PM Romano Prodi (18 March 2022): https://www.rainews.it/tgr/piemonte/articoli/2022/03/prodi-gorbaciov-mi-disse-che-putin-avrebbe-unito-russa-ed-europa-adf04f33-a1d9-4ff5-832e-277306c7fa43.html
• A selection of my stories for RAI (in chronological order): https://www.google.com/search?q=%22fabio+de+ponte%22+site:rainews.it&num=10&client=firefox-b-d&sca_esv=42b57a03e57cf1c0&tbs=sbd:1&tbm=nws&sxsrf=AHTn8zrTkL793c9wRJu6dADeQsqlZ8p84g:1740499826640&ei=cuu9Z8HpJvSKi-gP49it-QQ&start=0&sa=N&ved=2ahUKEwjB6pePm9-LAxV0xQIHHWNsK084KBDy0wN6BAgEEAY&biw=1512&bih=747&dpr=2
PAPERS
• De Ponte, F.: Jokes Aside: Measuring the Semantic Distance of Double Meanings, Graduate Student Award for best paper at the annual conference of the International Society for Humor Studies, Kraków, Poland, 7‑11 July, 2025. Available at https://github.com/fabiodeponte/JokesAside
• De Ponte, F., Rauchas, S.: Grounding Words in Visual Perceptions: Experiments in Spoken Language Acquisition. In: Proceedings of the Sixth Workshop on Natural Language for Artificial Intelligence (NL4AI 2022) colocated with 21th International Conference of the Italian Association for Artificial Intelligence (AIxIA 2022), Udine, Italy, 28 Nov. 2 Dec., 2022. Available at http://sunsite.informatik.rwth-aachen.de/Publications/CEUR-WS/Vol-3287/paper6.pdf
• De Ponte, F.: Grounding Words in Visual Perceptions: Experiments in Spoken Language Acquisition. Poster presentation at the SoCal NLP Symposium 2022, Santa Barbara, CA, USA, 18 Nov., 2022.
MSc THESIS
• De Ponte, F.: Grounding Words in Visual Perceptions: Experiments in Spoken Language Acquisition. Final Project in MSc in Data Science and Artificial Intelligence, University of London, 20 Sept., 2022. Available at: drive.google.com/file/d/1m5p7WZTn_4UOJ97mVENDqmm_HDjjA7SV/view?usp=sharing
DATASETS ON KAGGLE
• Objects, actions and spoken utterances (2022)
36K video clips for symbol grounding: visual and co-occurring audio elements
https://www.kaggle.com/datasets/fabiodeponte/symbolgrounding
• Shapes (2024)
100K geometrical shapes for attributes detection (num. sides, size, etc.)
https://www.kaggle.com/datasets/fabiodeponte/shapes
LECTURES
• In December 2023 and March 2024, I taught modules (six and ten hours, respectively) on Natural Language Processing techniques,
language models and journalism at the School of Journalism of the University of Turin.
• In collaboration with the University of Turin and journalist’s union Usigrai, organized two trainings for journalists: ”AI and Journalism: a
compass for charting courses in the storm” (30 June 2023) and ”Generative AI in the newsroom” (28 Maj 2024).
The recordings:
"The hidden costs of AI", Guido Boella - https://www.youtube.com/watch?v=ELrbt7phFPk
"What is Computational Linguistics", Luigi Di Caro: https://www.youtube.com/watch?v=kJ_6HFfkDnE
"The philosophical roots of AI", Fabio De Ponte: https://www.youtube.com/watch?v=kgTexDEEbs0
"Journalism and AI", https://www.youtube.com/watch?v=xrfwAcLq7-E
• In February 2023, initiated a working group of more than 60 RAI journalists and technicians dedicated to the applications of artificial intelligence and NLP techniques to journalism. We met weekly with the RAI Research Center, an internal body within the company, to design new tools for editorial needs.
Here are two of the lectures I have recorded for the group (in Italian):
Language models: https://www.youtube.com/watch?v=gy4WwZBuV0k
Embeddings: https://www.youtube.com/watch?v=hA9fQSjaGKo
INTERVIEWS
• Interviews to Adn Kronos newswire: 9 April 2025 ("Piano Ue positivo ma c'è un problema di numeri" - "EU's AI strategy welcomed, but figures remain problematic") and 8 February 2025 ("IA, Ue indietro, mancano fondi e politica industriale" - "EU lacks both resources and strategy for AI industry").
• Interview (Nov 2023) for Global Careers Calls of University of London Careers Service: "How AI will revolutionize the journalism sector"
• Interview (Nov 2024) for "Il salotto dell'imprenditoria innovativa" ("The Lounge of Innovative Entrepreneurship") of Foudation GB Mossetto: "My Journey From Journalism to AI"
BOOKS
• Bill Gates: le due vite dell’uomo che ha creato Windows ("Bill Gates: The Two Lives of the Man Who Created Windows"), Editori Internazionali Riuniti 2012 (https://amazon.it/Bill-Gates-delluomo-creato-Windows/dp/8835991366)
[Download the pdf: https://drive.google.com/file/d/15kdVsAqu-rtupe2C9gf2A3zG3P6H-vLy/view]
• La Mela bacata: le contraddizioni del sogno di Steve Jobs ("The Wormy Apple: The Contradictions of Steve Jobs' Dream"), Editori Internazionali Riuniti 2011 (https://www.amazon.it/mela-bacata-contraddizioni-sogno-Steve/dp/8835990599/ref=cm_cr_arp_d_product_top?ie=UTF8)
[Download the pdf: https://drive.google.com/file/d/1YATqUQ08jcMMDDS02W-giaft57UZKsB9/view]
Here a review of both: https://st.ilsole24ore.com/art/notizie/2012-08-26/ecco-perche-eterna-sfida-175051.shtml (translated in English)
CURRENT AND PAST PROJECTS
Rai Media Lab
In 2023, I initiated a working group of more than 60 RAI journalists and technicians dedicated to the applications of artificial intelligence and NLP techniques to journalism. We used to meet weekly with the RAI Research Center, an internal body within the company, to design new tools for editorial needs.Yattaphone.org - www.yattaphone.org
In 2021, I assembled a Linux mobile phone based on Raspberry and wrote the code to allow the user to make phone calls. The project gathered a little community of interested people who have been exchanging ideas and experiments.Joshua Influencers Hunter - hunter.joshua.earth (not working anymore due to Twitter API end of support)
In 2018, I developed a tool (hunter.joshua.earth) that looks for tweets about a specific topic, in a specific geographical area and a specific language by a user with a number of followers over a specific threshold.Joshua Twitter Scanner - twis.joshua.earth (not working anymore due to Twitter API end of support)
In 2015, I developed a tool (twis.joshua.earth) monitoring politicians’ Twitter profiles and sending short news stories to LaPresse newswire’s supervisors’ desk.Editorial tools - joshua.earth/lapressetools
Between 2012 and 2015, I developed various tools for the editorial staff at LaPresse newswire, including a shared phonebook, an XML creator to deliver graphic contents to clients, a customized graphics creator for European elections and World Football Cup, and an advanced data sheet to organize work shifts.Neural Notes Blog - appuntineurali.blogspot.com
Between 2006 and 2013, I kept a blog (first on Splinder and then on Blogspot) titled ”Appunti neurali” (”Neural notes”). It was dedicated to neural networks, artificial intelligence and sometimes observations on the cognitive development of my own daughter in her first years.Freelliance - dismissed
In 2009, I founded Freelliance in Rome, Italy. The mission of the company (the name stood for ”Free Lance Alliance”) was to gather freelance journalists through an online platform that offered a reward on a pay‑per‑story basis. I created the platform, a CMS that was able to send news through SMSs to users. Unfortunately, I could not find investors willing to finance the company.Face Recognition - appuntineurali.blogspot.com/2013/12/facebook-plans-for-artificial.html
In 2004, while attending the School of Journalism in Bologna (Italy), I coded a neural network in C language for face recognition that was able to identify my roommates. It was trained on a set of 40 pictures taken at home (years later I published the code on my blog on artificial intelligence titled ”Neural Notes”).Online journal of the School of Journalism of Bologna - joshua.earth/lastefani - dismissed
In 2004, I also developed a news website for the online journal of school. For a few years on, I taught lectures to new students in Bologna on how to manage the system.Radiodigitale.info - www.radiodigitale.info - dismissed
In 1999, I co‑founded Radiodigitale in Turin, Italy. At the best of my knowledge, it was the first Italian online radio. I set up online transmissions, studio and developed a CMS in PHP based on text parsing that did not need a database. It lasted until 2013, it is currently dismissed.
FUTURE PROJECTS
Conflicts Predictor
I propose to quantitatively assess the global conflict levels over time by conducting sentiment analysis on historical news archives. The aim is to answer questions like: How does the current level of public discourse compare to past decades? Can we graphically represent the "temperature" of public debate from over the years? Is there a correlation between peaks in this graph and major global conflicts? A description of the project can be found here.An Alternative to Tokenization
In recent years, Natural Language Processing models have shown compelling progress in generating and translating text. Yet, the symbols that are manipulated by these models are not produced within the models themselves. On the contrary, they are externally given in the form of tokens. The models only measure the probability that specific tokens follow others, in order to generate new sequences. The semantic interpretation of inputs and outputs is completely invisible to the system. Even recent progress on multimodality relies on the insertion of special tokens and it does not address the problem of the creation of symbols, which remains exogenous. The aim of this project is to address this issue, by designing a new method for tokenization, inspired by cross-situational word learning, to allow the creation of an open-ended set of symbols, on the basis of perceptions and experiences. The idea is to map audio features directly onto video features, through a multi-output neural network that generates a unique identifier for each audio-video pairing. These identifiers would then be used in place of traditional tokens in language models but they could also provide the basis for Fluid Construction Grammar applications. The model aims to develop an internal symbol-creation mechanism, thereby moving a step closer to a system that can understand and interpret semantic content in a more human-like manner.
Page updated
Google Sites
Report abuse