Beyond removing refusals, I hope these techniques can be used for "latent fine-tuning" to customize models at inference time.
LLMS
-
Heretic 1.1: Open-Source LLM Abliteration Library Evolution
By
–
Abliterate LLMs with Heretic 1.1
— Maxime Labonne @ ICLR (@maximelabonne) 11 décembre 2025
It's cool to see this project evolving into a solid open-source library
The new viz feature shows how the abliteration process gradually groups the residual vectors into two nice clusters 👀 pic.twitter.com/PLYmRPuYB7Abliterate LLMs with Heretic 1.1 It's cool to see this project evolving into a solid open-source library The new viz feature shows how the abliteration process gradually groups the residual vectors into two nice clusters
-
Scobleizer predicts X understands me; more Grok and video ingestion in 2026
By
–
After having @blevlabs analyze a few of my lists here I can see how X will do all of this and more. It already knows how I think very deeply. I left off one thing of my AI predictions for 2026 for next year: we all will talk with Grok a LOT more. And the video ingestion
-
NotebookLM provides podcast, mind map, and video show of news
By
–
Here is the @NotebookLM with a podcast, mind map, and video show of today's news:
-
OpenAI hints at ‘garlic’ model ahead of GPT-5.2
By
–
BREAKING : OpenAI hints at “garlic”, a rumoured codename of their upcoming model. GPT-5.2 is expected tomorrow
-
Google upgrades Stitch Agent with Gemini 3 Pro
By
–
Google upgraded Stitch design Agent with Gemini 3 Pro, which is the new default mode now.
— 🚨 AI News | TestingCatalog (@testingcatalog) 10 décembre 2025
Testing time 👀 https://t.co/uW1w4r2fOe pic.twitter.com/oAaZrtvFTeGoogle upgraded Stitch design Agent with Gemini 3 Pro, which is the new default mode now. Testing time
-
ChatGPT’s Simple Memory Architecture Revealed
By
–
Everyone assumes ChatGPT uses some sophisticated RAG system with vector databases to remember your conversations. Not really. After reverse-engineering the entire memory architecture, this developer discovered something way simpler and frankly, smarter. ChatGPT uses four
-
Choosing Right LLM Model for Your Product Use Case
By
–
Some of our top customers are still choosing Llama 3.1 8B.
— Cerebras (@cerebras) 10 décembre 2025
For a while, we jumped to whatever hottest, latest model was taking up our twitter feed. 🙈
But as we are quickly realizing, to create a SOTA product, you need a model that fits your exact use case.
Here’s what our… pic.twitter.com/caeW3NFA8JSome of our top customers are still choosing Llama 3.1 8B. For a while, we jumped to whatever hottest, latest model was taking up our twitter feed. But as we are quickly realizing, to create a SOTA product, you need a model that fits your exact use case. Here’s what our
-
Google Adds Scheduled Tasks & Suggestions in Beta
By
–
BREAKING 🚨: Google released Scheduled Tasks and Suggestions in Beta on Jules SWE Agent, where it can proactively propose improvements to your repository.
— 🚨 AI News | TestingCatalog (@testingcatalog) 10 décembre 2025
The whole UI got redesigned as well 🤩 https://t.co/ZhJkaTMN9k pic.twitter.com/cto6tHnouqBREAKING : Google released Scheduled Tasks and Suggestions in Beta on Jules SWE Agent, where it can proactively propose improvements to your repository. The whole UI got redesigned as well
-
Grok AI walks through saving wife’s life with CPR guidance
By
–
Someday your robot might save your life. But this is a huge step. Have you ever asked AI about different crises? I have. “Hey @grok my wife just fell on the floor and isn’t breathing, what should I do?” It walks through saving her life even if I didn’t know CPR. My first