Our Aeneas AI model gives historians valuable new insights into ancient inscriptions & ancient history that may have taken years to uncover otherwise. Published in @Nature today:
MULTIMODAL AI
-
Multimodal LLMs Enable Unprecedented Privacy Risks Through Recording Mining
By
–
The problem is not just the proliferation of devices that let you record people without their knowledge, but the fact that multimodal LLM let you use recordings in ways that neither law not society anticipated. Everyone has an easy way to mine hours of
footage. No forgetting. -
Demis Hassabis on AI, AGI, and the Future of Computing
By
–
Here's my conversation with @demishassabis, CEO of Google DeepMind, all about the future of AI & AGI, simulating biology & physics, video games, programming, video generation, world models, Gemini 3, scaling laws, compute, P vs NP, complexity, energy (solar & fusion), and much… pic.twitter.com/VhbqmBQHje
— Lex Fridman (@lexfridman) 23 juillet 2025Here's my conversation with @demishassabis
, CEO of Google DeepMind, all about the future of AI & AGI, simulating biology & physics, video games, programming, video generation, world models, Gemini 3, scaling laws, compute, P vs NP, complexity, energy (solar & fusion), and much -
Microsoft Study Shows AI Limits Physical Tasks vs Coaching
By
–
New @Microsoft paper analyzed 200,000 Copilot conversations. Because of lower trust, under-performance, or lack of physical embodiment, AI is much more likely to train, coach, teach, or advise than, say, investigate a car accident or run a 100y dash. >> If you do things on
-
VEO 3 Logo Transformation: JSON Prompts Improve Video Outputs
By
–
La transformation de logos dans VEO 3
— VISION IA (@vision_ia) 23 juillet 2025
Les gens ont découvert qu'utiliser des prompt au format JSON améliore grandement les sorties vidéos https://t.co/MkMmjuVb7L pic.twitter.com/TATUAyLD46La transformation de logos dans VEO 3 Les gens ont découvert qu'utiliser des prompt au format JSON améliore grandement les sorties vidéos
-
Hugging Face Benchmark: Vision LLMs Long Video Input Performance
By
–
Interesting new benchmark from Hugging Face testing how well vision LLMs can handle long video inputs (generally after they've been split into many thousands of images) – my notes here: https://
simonwillison.net/2025/Jul/23/ti
mescope/
… -
Replicate Partners with Runway for Video and Upscaling Models
By
–
We've partnered with Runway @runwayml to bring their video + upscale models to Replicate.
— Replicate (@replicate) 23 juillet 2025
Create cinematic shots.
Upscale up to 40 seconds of video to 4K.
Try them today:https://t.co/KRsOkGhHDUhttps://t.co/DNpVnuvI75 pic.twitter.com/wNsFBHhei3We've partnered with Runway @runwayml to bring their video + upscale models to Replicate. Create cinematic shots.
Upscale up to 40 seconds of video to 4K. Try them today: https://
replicate.com/runwayml/gen4-
turbo
… https://
replicate.com/runwayml/upsca
le-v1
… -
10x Bandwidth Communication Interface for AI Systems
By
–
need a 10x higher bandwidth way to communicate with ai
-
AI Models Still Struggle with Quality Pelican Image Generation
By
–
I'll believe they've done that when one of the models draws me a good pelican!
-
Hedra Launches Live Avatars: 15x Cheaper AI Streaming Solution
By
–
Hedra just launched Live Avatars, a new streaming avatar model in partnership with LiveKit
— The Rundown AI (@TheRundownAI) 23 juillet 2025
It allows users to give a visual identity to their AI at just $0.05/min — 15x cheaper than existing solutions — with ultra-low sub-100ms latencypic.twitter.com/rg0oUcAQmQHedra just launched Live Avatars, a new streaming avatar model in partnership with LiveKit It allows users to give a visual identity to their AI at just $0.05/min — 15x cheaper than existing solutions — with ultra-low sub-100ms latency