Rijul Gupta retweeted
To try out DeepSeek R-1 recently, I did the laziest thing possible, which was to open up ChatGPT and pick out a few recent prompts.
I thought the answers were remarkably similar in both content and structure. Judge for yourself:
1. What was used for refrigerants before cfcs?
Deepfakes are fucking up biometric security systems. Biometric security in face uses embedding vectors to determine facial similarity, these same models are integrating into Deepfakes. Banks, passports, etc - nothing is safe anymore.
deepmedia.ai/blog/2025-biome…
Rijul Gupta retweeted
Not everything viral is real. The burnt Oscar statue photo, shared widely on Instagram, was flagged by our DeepID platform.
See how quickly fake visuals can gain traction—and why content verification is critical—in our latest report.
deepmedia.ai/reports/lost-os…
#DeepfakeDetection #AITrust
Rijul Gupta retweeted
What makes CLIP work?
The contrast with negatives via softmax?
The more negatives, the better -> large batch-size?
We'll answer "no" to both in our ICCV oral🤓
By introducing SigLIP, a simpler CLIP that also works better and is more scalable, we can study the extremes.
Hop in🧶
Rijul Gupta retweeted
I came up with a technique for dynamic token selection in Vision-Language Models.
Instead of wasting compute on every part of an image, this method adapts the number of tokens based on the complexity of each region.
Here’s an example of how it works: 👇
theverge.com/2025/1/6/243375…
Two of these can be linked to run a 405B model like llama 3.1 (quantized). Fuckin nuts.
Curious about @huggingface SMOLLM V2 models, going to give them a shot.
Rijul Gupta retweeted
Good morning. Was watching DS9 again. Specifically eps with the Defiant being awesome.
This fell out of my brain.
"Tough little ship." ;)
Hope you enjoy it.
#StarTrekDS9 #StarTrek #StarTrekLowerDecks #StarTrekSNW #StarTrekPicard
Rijul Gupta retweeted
SmolVLM running at 80 tok/sec on a 4-year old M1 is just incredible. The future is lit
SmolVLM was just released 🚀
It's a great, small, and fully open VLM that I'm really excited about for fine-tuning and on-device use cases 💻
It also comes with 0-day MLX support via mlx-vlm, here's it running at > 80 tok/s on my M1 Max, cc @Prince_Canuma 🤯
Rijul Gupta retweeted
BREAKING NEWS: How Schools Across America Are Struggling With AI Deepfakes: How Schools Across America Are Struggling With AI Deepfakes
Authored by Aaron Gifford via The Epoch Times (emphasis ours),
Gone are the days where the biggest concern is… dlvr.it/TFg5zz
Rijul Gupta retweeted
This is so cool.
This edit was done all inside of ComfyUI! The cuts were done based on a BPM analyzer and the videos were generated with Runway API nodes
So so so clever
Rijul Gupta retweeted
It seems strange that the tools we have for creating synthetic media are much more restrictive about what we can create than the platform that host and distribute the resulting material. My column on YouTube opening its doors to deepfakes today: platformer.news/p/youtube-op…

ALT Still, it seems noteworthy that the most accessible tools we have for creating synthetic media are more restrictive than the tools we have for distributing it. Meta’s synthetic media policy, which would fit on the back of a napkin, is roughly as permissive as YouTube’s. (TikTok is the only one of the three to ban synthetic videos of non-public figures.) Perhaps all this will balance out over time. For the moment, though, it seems strange that the likeliest source of harms from AI comes from not from leading AI companies, which will block it at the source of creation, but from open-source tools used to create materials hosted and distributed by tech giants.
Great definition of a Deepfake by @DeepMedia_AI resident philosopher Ryan!
🔍 Curious about the true essence of 'Deepfake'? 🎭 Dive into the exploration by Ryan Ofman and uncover its comprehensive definition. Let's ensure we're on the same page with this evolving technology. Read more here: deepmedia.ai/blog/what-is-a-…
#deepfake #ai #ArtificialIntelligence #technology























