This blew my mind The model literally builds graphs of hypotheses nodes for ideas, edges for experiments. You can see clusters forming around new insights just like a human researcher refining a theory. That’s not prompting that’s cognition.
LLMS
-
A 4-Stage Iterative Research Loop for AI Reasoning and Control
By
–
The Deep Research Loop: The paper starts with this core diagram: a 4-stage research loop (Observe → Hypothesize → Experiment → Revise). Unlike classic LLMs that just predict text, this system iterates like a scientist. Every loop improves reasoning and robot control
-
Another example: MiniMax-M2
By
–
Another example: MiniMax-M2 https://github.com/MiniMax-AI/MiniMax-M2
… -
PyTorch MPS Backend Bug: Silent Tensor Contiguity Failures
By
–
Beautiful technical debugging detective longread that starts with a suspicious loss curve and ends all the way in the Objective-C++ depths of PyTorch MPS backend of addcmul_ that silently fails on non-contiguous output tensors. I wonder how long before an LLM can do all of this.
-
SNNs and Neuromorphic Computing: Energy-Efficient AI Alternative
By
–
This is the way forwards for AI – SNNs or Neuromorphic Computing are highly energy efficient vs traditional LLMs, lower energy consumption = lower running costs + lower carbon emissions. Plus they can learn on the fly. Vast potential in SNNs including for logical reasoning with
-
Meta AI Gearing Up for Major UI and Model Feature Upgrades
By
–
BREAKING : It seems like Meta AI is gearing up to bridge the feature gap with frontier AI Labs in one shot. Numerous new labels are emerging in the revamped UI. Many of these features were planned to work with Llama 4, but were abandoned. Meta Superintelligence is coming?
-
Deep Generative Modelling: Comprehensive Lectures on Models and Applications
By
–
Great lectures on deep generative modelling, covering a whole bunch of topics, model families, and learning algorithms in generative models space. And applications in vision and natural language processing, and reinforcement learning.
-
Google Gemini to introduce Visual Layout feature
By
–
BREAKING 🚨: Google is gearing up to release the Visual Layout feature on Gemini. Visual Layout can synthesise an interactive layout to represent information on a given topic.
— 🚨 AI News | TestingCatalog (@testingcatalog) 25 octobre 2025
Gemini Pulse 👀 pic.twitter.com/Tv8W4aZhjhBREAKING : Google is gearing up to release the Visual Layout feature on Gemini. Visual Layout can synthesise an interactive layout to represent information on a given topic. Gemini Pulse
-
GPT-5.1-mini model discovered in OpenAI GitHub repository
By
–
Seems like GPT-5.1-mini is real GPT-5.1-mini has been mentioned in the official "openai-agents-js" github repository. And here is a potential naming convention for upcoming GPT models. h/t @scaling01