Making tools to fight disinformation at @AFP Medialab | @veraai_eu | Prev. InfoSci @Cornell — Pics: https://xitter.cf/t.co/DmSvqxkI9w
France
Joined May 2020
- Tweets22
- Following160
- Followers43
- Likes117
Valentin Porcellini retweeted
Finally, we provide NLP credibility signals (including persuasion techniques) in the text analysis of the Assistant (thanks to @kbontcheva @SheffieldNLP, @KInIT_sk and @ipekbrs ).
Great integration work by @_valentinp & @b_goupil !
Valentin Porcellini retweeted
We released today a new version v0.82 of our verification plugin (@InVID_EU @WeVerify @veraai_eu) with enhanced layout and improvement of synthetic image detection (using @meverteam near duplicate search to enhance results on degraded copies or screenshots of synthetic images)
We just published a new release of the Verification Plugin including several improvements.
We are planning to have even more enhancements of our beta tools running AI models in the next update. See our full roadmap here: github.com/AFP-Medialab/veri…
#InVIDWeVerify
Just before #euelections2024, a new release of our verification plugin v0.80 (@InVID_EU @WeVerify and now @veraai_eu). Improvements on beta tools [synthetic image detection, deepfake, voice cloning by Loccus.ai]
u.afp.com/plugin
We just released the 0.78.1 version of the Verification plugin, with a fix for the Forensic.
The 0.78 version also included new algorithms for the synthetic image detection tool (beta access), assistant improvements, and a new translation engine.
@veraai_eu
Valentin Porcellini retweeted
🚀🚀Introducing Ferret, a new MLLM that can refer and ground anything anywhere at any granularity.
📰arxiv.org/abs/2310.07704
1⃣ Ferret enables referring of an image region at any shape
2⃣ It often shows better precise understanding of small image regions than GPT-4V (sec 5.6)
Valentin Porcellini retweeted
📢 C'est demain : @MediasEnSeine, festival international des médias, en partenariat avec l'#AFP !
🗣 Avec les participations de @sohnic, @_valentinp et @PierreAusseill.
👉 Plus d'infos : u.afp.com/MES-2023
#MediasEnSeine #medias
Happy to drive my 1998 car:
“Mozilla found that numerous car companies collect sensitive user information like photos, immigration status, and even sexual activity”
foundation.mozilla.org/en/pr…
theverge.com/2023/9/6/238610…
Valentin Porcellini retweeted
We've examined some of the most widely shared false and misleading tweets about the war in Ukraine in recent weeks, all posted by Twitter Blue subscribers whose content is promoted by the platform.
bbc.co.uk/news/world-europe-…
We have renamed the InVID WeVerify plugin repo from we-verify-app to verification-plugin to make things a little more simple.
(@veraai_eu)
github.com/AFP-Medialab/veri…
Happy to announce the version 0.76 of the verification plugin! Arabic language, enhanced UX and more responsive UI.
This is also the first step towards a code base that is more easy to work with, which will simplify our future features developments. (@veraai_eu)
While at #GlobalFact10, we released a new version of our verification plugin (@InVID_EU @WeVerify @veraai_eu). This v0.76 has been approved by Google last night. It's a major re-factoring of the plugin by @_valentinp & @b_goupil using the new manifest v3
Valentin Porcellini retweeted
Dear verification plug-in users
Due to a server failure, access to advanced tools of the plug-in (@InVID_EU @WeVerify @veraai_eu) is not working at the moment. This should affect primarily the forensic export and CheckGIF feature. Sorry for the inconvenience.
We’re on the case!
Valentin Porcellini retweeted
Registration OPEN - register NOW for the 'Future of AI: Countering Sophisticated & Advanced Disinformation'. It takes place in Brussels on 29 June. You can also attend online. Brought to you by @ai4mediaproject @TITAN_Thinking #AI4Trust @HorizonEU & us
veraai.eu/posts/registration…
Valentin Porcellini retweeted
How rumors start. #verification
1. Said yesterday in Twitter that I am a famous banjo player, together with @JackK and @PaulMyersBBC That data came from Google Bard.
2. Today, Bing Chat (ChatGPT 3.5 + Bing) considers it to be a fact.
I am not rigging any prompts here, this was the very first try. Bard takes my prompt already as a fact and starts hallucinating. This happens often. Here are three examples. Did you now @jackk @paulmyersbbc and @henkvaness are a famous banjo trio from Saint Kitts? At least we are not gang members.
Valentin Porcellini retweeted
#FactCheckers are key in the battle against #disinformation. But their work is being hampered: an #EUFunded project shows that the recent steps taken by #SocialMedia platforms to limit access to data are preventing effective analysis.
👇
europa.eu/!39TbKc
ALT Hand is turning a dice and changes the word FAKE to FACT
Valentin Porcellini retweeted
🚨🚨@ElonMusk complied when India’s right-wing prime minister asked him to block a BBC documentary critical of him, and even deleted MP tweets interc.pt/3XsVHyN by me & @mazmhussain
Valentin Porcellini retweeted
A team led by our @dteyssou and Valentin Porcellini of @AFPcom with @kbontcheva of @shefcompsci and others looked into the challenge of preserving fact-checked disinformation items.
Another research outcome of the 2023 @digitalmethods Winter School.
veraai.eu/posts/mapping-memo…
Valentin Porcellini retweeted
Christopher Nolan says he recreated the first nuclear weapon detonation without using CGI for #Oppenheimer
(via @totalfilm)
Valentin Porcellini retweeted
Docs show Facebook and Twitter closely collaborating w/ Dept of Homeland Security, FBI to police “disinfo.” Plans to expand censorship on topics like withdrawal from Afghanistan, origins of COVID, info that undermines trust in financial institutions. interc.pt/3Dq6TDB
Valentin Porcellini retweeted
Sorry bro theverge.com/2022/10/28/2342…
Valentin Porcellini retweeted
turning all of your interests into content is exhausting. i’ve honestly fallen out of love with a lot of things simply because i believed i had to capitalize on my enthusiasm for them with videos etc.
it’s cool to enjoy shit and keep it to yourself. – at Toronto, Ontario