So GPT-5 is running on GB200, to take full advantage of it, it could be in FP4. So I wonder if this change is hiding the size of the model somewhat. If it was FP8 and on H100, it would be now running at something like 10 tokens per second rather than current 50 tps.
RESEARCH
-
OpenAI’s GPT-5-Pro: Cost Optimization Over Capability Push
By
–
Given that we saw OpenAI test better checkpoints on the arena, it does also seem like they have also optimised for cost / mass appeal rather than pushed the capabilities as far as they could. We also don't have almost any benchmarks for GPT-5-Pro, which is genuinely superb and
-
MIT AI Discovers New Antibiotic Candidates for Drug-Resistant Bacteria
By
–
Just read a fascinating MIT study: AI designed 2 brand-new antibiotic candidates for drug-resistant gonorrhoea & MRSA.
It analysed 36M compounds (many never seen before), created new molecules & killed the superbugs in lab tests.
Still years from patients, but a huge step -
AGI Predictions Fall Short: AI Still Struggles with Basic Tasks
By
–
2024: In 2025 we will have AGI
2025: “how many Bs in blueberry” “what is 8.11-8.9” -
MIT CSAIL Explores Current State and Future of AI-Powered Robots
By
–
@CBSMornings recently visited MIT CSAIL to understand the state of AI-powered robots & where we're headed.
— MIT CSAIL (@MIT_CSAIL) 14 août 2025
"The robots we have today are primarily demonstrations," says MIT Prof. & CSAIL Director Daniela Rus, but she adds that "we can choose to do extraordinary things." pic.twitter.com/m4cpwTvYYZ@CBSMornings recently visited MIT CSAIL to understand the state of AI-powered robots & where we're headed. "The robots we have today are primarily demonstrations," says MIT Prof. & CSAIL Director Daniela Rus, but she adds that "we can choose to do extraordinary things."
-

Meta releases DINOv3 computer vision model under commercial license
By
–
To help foster innovation and collaboration in the computer vision community, we’re releasing DINOv3 under a commercial license with a full suite of pre-trained backbones, adapters, training and evaluation code, and (much!) more. Find them here: https://
github.com/facebookresear
ch/dinov3
… -
DINOv3 Models Now Available in Hugging Face Transformers
By
–
We're excited to share that we have Day-0 support in Hugging Face Transformers for DINOv3 so people can easily leverage the full family of models. Find out more on @huggingface here:
-
DINOv3: State-of-the-Art Self-Supervised Vision Model
By
–
Introducing DINOv3: a state-of-the-art computer vision model trained with self-supervised learning (SSL) that produces powerful, high-resolution image features. For the first time, a single frozen vision backbone outperforms specialized solutions on multiple long-standing dense… pic.twitter.com/nwS3zFCaaN
— AI at Meta (@AIatMeta) 14 août 2025Introducing DINOv3: a state-of-the-art computer vision model trained with self-supervised learning (SSL) that produces powerful, high-resolution image features. For the first time, a single frozen vision backbone outperforms specialized solutions on multiple long-standing dense
-
DINOv3: Self-Supervised Learning for Large-Scale Vision Models
By
–
A few highlights of DINOv3 👇
— AI at Meta (@AIatMeta) 14 août 2025
1️⃣SSL enables 1.7B-image, 7B-param training without labels, supporting annotation-scarce scenarios including satellite imagery
2️⃣Produces excellent high-resolution features and state-of-the art performance on dense prediction tasks
3️⃣Diverse… pic.twitter.com/6glMC1S3KWA few highlights of DINOv3 SSL enables 1.7B-image, 7B-param training without labels, supporting annotation-scarce scenarios including satellite imagery
Produces excellent high-resolution features and state-of-the art performance on dense prediction tasks
Diverse -
LangChain Academy Launches Deep Research with LangGraph Course
By
–
🔥 Our latest LangChain Academy course – Deep Research with LangGraph – is now live! 🔥
— LangChain (@LangChain) 14 août 2025
Deep research agents are taking off – from major AI labs to companies building their own.
Research is inherently open-ended. You can't always predict whether a question needs broad… pic.twitter.com/BDyNq6zlxcOur latest LangChain Academy course – Deep Research with LangGraph – is now live! Deep research agents are taking off – from major AI labs to companies building their own. Research is inherently open-ended. You can't always predict whether a question needs broad