In our safety baselines, only models with a post-mitigation score of medium or below can be deployed, and only models with a post-mitigation score of high or below can be developed further. We also increase security protections commensurate with model risk.
AI
-

Safety Baselines and Governance for AI Model Development
By
–
Our new safety baselines and governance process will turn these technical findings into safety decisions for model development and deployment. This involves establishing a cross-functional Safety Advisory Group to make safety recommendations.
-
Frontier Model Safety Framework Implementation Details
By
–
We are systemizing our safety thinking with our Preparedness Framework, a living document (currently in beta) which details the technical and operational investments we are adopting to guide the safety of our frontier model development.
-

AI Preparedness Team Evaluates Model Risks Through Scorecards
By
–
Our Preparedness Team will drive technical work, pushing the limits of our cutting edge models to run evaluations and closely monitor risks, including during training runs. Results will be synthesized in scorecards that track model risk.
-
Build Chat LLM and AI Agents Apps Free Webinar Tomorrow
By
–
Learn how to build your own Chat LLM and AI Agents apps in our free webinar tomorrow at 9 am PST Some topics include:
• How to use RAG
• Fine-tuning open-source models
• Creating and deploying Chat LLMs & AI Agents RSVP here: -
Cerebras Booth Highlights: Sparsity, Inference Optimization, Multilingual Models
By
–
We really enjoyed meeting many brilliant ML researchers and model builders at the Cerebras booth last week at NeurIPS. Some topics of discussion: * Unlocking Sparsity for Training Efficiency
* Optimizing Training for Inference Efficiency
* Multilingual Models
* Training large -
Google’s Responsible AI: Human-Centered Design Approach
By
–
Responsible AI is artificial intelligence built using a human-centered design approach. Learn how Google Research approaches Responsible AI, including how it’s built, tested, and how to release AI products responsibly. Watch the full video → https://t.co/OlilKFSzxs pic.twitter.com/JNWm63qnXd
— Google AI (@GoogleAI) 18 décembre 2023Responsible AI is artificial intelligence built using a human-centered design approach. Learn how Google Research approaches Responsible AI, including how it’s built, tested, and how to release AI products responsibly. Watch the full video → https://
goo.gle/3tpDDvy -
ChatChaos: Stress Testing LLMs with Chaos Engineering
By
–
Chaos Engineering The behavior of LLMs can be unpredictable and chaotic ChatChaos is a fake LLM by @andrewnguonly specifically designed to return invalid JSON, hallucinate, and have long response times Can be used to stress test your system! Blog:
-
Understanding Log Probabilities Parameter in Language Models
By
–
Hey Amy, following up on this, the top logprobs number is just setting how many log probs you want returned, not the range. The range can be any negative number to 0 but that is different than the top logprobs parameter.
-
Poe Windows App Now Available with Multiple AI Models
By
–
You can download Poe for Windows at http://
poe.com/download. As usual, all bots from across Poe are available, including GPT-4, Gemini Pro, Claude 2, Claude Instant, DALL-E 3, StableDiffusionXL, Llama-2, PaLM, and all user-created bots. (2/2)
