LLMS

→ View original post on X — @googleai

Google Releases MedGemma 1.5 for Healthcare AI Innovation

By

@googleai

–

16 January 2026 19h21

This week, we released a pair of Gemma updates that developers can use to innovate across the healthcare and translation spaces: — MedGemma 1.5 4B: Our updated open medical AI model improves performance in 3D imaging and reasoning over medical records. We also kicked off the

→ View original post on X — @openai

ChatGPT Go launches globally with 10x more features

By

@openai

–

16 January 2026 19h00

ChatGPT Go is rolling out globally in every country where ChatGPT is available. ChatGPT Go is our low-cost subscription tier that gives you 10x more messages, file uploads and image creation vs free tier, more memory, longer context window, and unlimited use of GPT 5.2 instant

→ View original post on X — @miketamir

Motif-2-12.7B-Reasoning: RL Training Recipes Practitioner Guide

By

@miketamir

–

16 January 2026 17h37

Motif-2-12.7B-Reasoning: A Practitioner's Guide to RL Training Recipes https://
buff.ly/WtUMyq0
#AI #MachineLearning #DeepLearning #LLMs #DataScience

Claude Opus 4.5 as a genius prompt engineer

By

–

16 January 2026 12h04

Claude Opus 4.5 is s genius prompt engineer. The main skill is explaining your goal clearly My workflow is this:
> I need to solve [TASK]. Research best practices and breakdown how to solve the problem in its first principles. > As an expert prompt engineer, generate a

Pattern #8: Iterative Refinement Protocols with Specific Dimensions

By

–

16 January 2026 9h30

Pattern #8: Iterative Refinement Protocols Public: "Improve your response" Internal: Structured refinement with specific dimensions. "Iteration 1: Optimize for accuracy. Iteration 2: Optimize for clarity. Iteration 3: Optimize for conciseness. Score each dimension. Stop when

Pattern #7: Evidence-Grounded Generation

By

–

16 January 2026 9h30

Pattern #7: Evidence-Grounded Generation Public: "Use examples" Internal: Explicit evidence tagging and traceability. "For each claim, cite source [S#]. For each inference, state reasoning [R#]. Confidence: [0-1]. If confidence

Adversarial Self-Critique: Find Flaws in Own Reasoning

By

–

16 January 2026 9h30

Pattern #6: Adversarial Self-Critique Public: "Review your answer" Internal: Assume adversarial role to attack own reasoning. "You are now a skeptic. Find flaws in the previous reasoning. What assumptions are weakest? What evidence contradicts? Generate counterarguments."

Multi-Stage Reasoning Pipelines with Explicit Stage Separation

By

–

16 January 2026 9h30

Pattern #4: Multi-Stage Reasoning Pipelines Public: "Think step by step" Internal: Explicit stage separation with intermediate checkpoints. "Stage 1: Extract facts. Stage 2: Identify constraints. Stage 3: Generate candidates. Stage 4: Filter. Stage 5: Rank. Stage 6: Verify."

Constitutional Prompting: Self-Correcting Multi-Layered Principles

By

–

16 January 2026 9h30

Pattern #1: Constitutional Prompting Public docs say: "Be clear and specific" DeepMind actually uses: Multi-layered constitutional principles that self-correct. Example: "First verify this follows principle X. If violation detected, revise. Then check principle Y. Iterate