Meta AI : l’ambition totalitaire, version bêta Meta triche avec ses modèles L’IA explose les compteurs carbone Shopify remplace les recrutements par l’IA Google connecte les agents intelligents WAI by @BNPParibas fête ses 10 ans Tout est dans la 35e édition
SAFETY
-
AI Company Named Top 20 in CRN AI 100 List
By
–
We’re all fired up to be named one of the 20 Hottest AI Companies in the 2025 CRN A1 100! Thank you to @CRN for recognizing our work in delivering AI that maximizes impact and minimizes risk for businesses. Read more: https://
bit.ly/3G6LG6A -
Defending Against Prompt Injection: StruQ and SecAlign Methods
By
–
New BAIR Blog Post: Defending against Prompt Injection with Structured Queries (StruQ) and Preference Optimization (SecAlign) http://
bair.berkeley.edu/blog/2025/04/1
1/prompt-injection-defense/
… Works presented led by @_Sizhe_Chen_ and @julientpiet
. -
Deepfake Pandemic Drives Development of E-Fraud Detection Tools
By
–
Deepfake pandemic triggers a rush for tools to fight e-fraud https://
search.app/X5vwf #efraud #fraud #fake #deepfake #faceswap #syntheticmedia #deepfakevideos #AI #cybersec #infosec #ML #DL #NLP #Algorithms #Security #AIEthics #EthicalAI #OpenAI #100DaysOfCode #OpenSource #bot -
Grok’s unfiltered promise questioned after left-leaning tests
By
–
A quick political detour: Remember Grok’s edgy, unfiltered promise? Musk said it wouldn’t be “woke.” But earlier Grok models leaned left in tests. It’s unclear if Grok 3 corrected that or simply rebranded it.
-
Concerns About Grok Accessing Direct Message Information
By
–
Scared when grok can reference everything it knows about your dms
-
AI Ethics: Killer Robots Over Climate Solutions?
By
–
Imagine if instead of asserting power, you could solve the climate crisis, world poverty and abundance? no, let’s make killer AI robots. Seems a bit dystopian
-
ChatGPT bias concerns and UX design challenges
By
–
This is interesting, but will make responses be more and more angled to your world view. I like the stateless part of ChatGPT, and I guess the temp chats are good for this. This is a UX problem that needs TLC
-
Moderation v7 Update: Safer Model, More Permissive Policies
By
–
We updated our moderation with v7 and it should let 70 percent more things through now because the model itself is much more safe. Still more to come but in the meantime, if you can give us lists of prompts you're having issues with that'll help! Thanks!