Microsoft a intégré le contrôle de l’ordinateur à Copilot Studio, permettant aux utilisateurs de créer des agents capables d’agir directement sur le bureau et le web. Ils ont aussi lancé Copilot Vision dans Edge : un assistant capable de voir ce que l’utilisateur consulte (avec
MULTIMODAL AI
-
Alibaba Releases Wan 2.1-FLV2-14B Open-Source Video Generation Model
By
–
Alibaba released Wan 2.1-FLF2V-14B, an open-source AI for video generation
— Rowan Cheung (@rowancheung) 21 avril 2025
What makes this model unique is its ability to let users control their generations with first and last frame inputs
The result is 720p cinematic clips with smooth transitions!pic.twitter.com/LHBaqtw9iGAlibaba released Wan 2.1-FLF2V-14B, an open-source AI for video generation What makes this model unique is its ability to let users control their generations with first and last frame inputs The result is 720p cinematic clips with smooth transitions!
-
Google Releases Gemini 2.5 Flash Hybrid Reasoning AI Model
By
–
Google continues to cook—this time dropping Gemini 2.5 Flash, a hybrid reasoner AI —Matches 04-mini
—Strong across reasoning, STEM, and visual reasoning
—Includes 'thinking budget' to optimize for cost
—Available via Google AI Studio, Vertex, Gemini app -
Selecting AI models for different tasks: speed, research, writing, images and summaries
By
–
> 4o when I got no time, I need something solved fast
> o3 when I solve business challenges, when I need research and brain power
> Claude 3.7 when I need really good writing
> 4o when I need good images
> Gemini 2.5 Pro when I need summaries of big PDFs / long videos -
ICLR 2025: Multimodal AI Opportunities and Tokyo Living
By
–
Will go to ICLR 2025. We could chat about potential chances (not merely visual but also speech/audio and embodied ai), and anything else like working/living in Tokyo
-
AI System Enables Humans to See New Colors Beyond Visual Spectrum
By
–
Humans just saw a *new* color—literally outside the known visual spectrum. BAIR faculty and visual computing expert Ren Ng and collaborators made it possible with the Oz Vision System. Newly published in @ScienceAdvances
: https://
science.org/doi/10.1126/sc
iadv.adu1052
… -
OpenAI Adds Image Generation to Custom GPTs
By
–
OpenAI prepares Custom GPTs for native GPT-4o image generation. The setting seems to be already working, and you can try it on "Cyberpunk Artist" GPT. However, it doesn't follow GPT instructions in 50% of cases yet.
-
Google Unveils Android XR Platform at TED Conference
By
–
Shahram Izadi dévoile la plateforme XR de Google Android en conférence TED.
— VISION IA (@vision_ia) 19 avril 2025
Mon avis d’expert :
On entre dans une ère où l’interface disparaît.
Plus besoin de taper, de parler ou même de penser à interagir. L’IA capte, retient et anticipe.
Ce n’est plus un outil, c’est une… https://t.co/YsAfwYfNgnShahram Izadi dévoile la plateforme XR de Google Android en conférence TED. Mon avis d’expert : On entre dans une ère où l’interface disparaît. Plus besoin de taper, de parler ou même de penser à interagir. L’IA capte, retient et anticipe. Ce n’est plus un outil, c’est une
-
Creative Steampunk Machine Prompt for Leonardo AI and DALL-E
By
–
L’un de mes prompts préférés avec Leonardo AI et chatgpt DALL E : « Une machine fictive à vapeur avec engrenages, tubes et pièces rotatives, dessinée au stylo bille noir et rouge avec des annotations manuscrites. Entre schéma technique et imagination. –ar 16:9 –raw –p –v 7 »
-
Multimodal tool use with CoT and vision libraries
By
–
It’s tool use (PIL/cv2, BFS) debugged with multimodal perception of tool output within the CoT; it’s not 4o-style multimodal output. See the share link in the thread for more info on the tool calls.