Karen Akers
karenny
AI & ML interests
None yet
Recent Activity
repliedto sergiopaniego's post 3 days ago
Something I really like when I study a subject is understanding its history, how it reached the point where it is today
I did that exercise for RL in post-training: from RLHF and PPO, to verifiable rewards, to the GRPO family of variants, to agents acting in environments. Everything is backed by what the labs themselves say in their public reports (DeepSeek, Qwen, Kimi, GLM-5, Nemotron, Mistral and more), in their own words
This is the companion piece to Class 3 of our Training Agents series with @burtenshaw. The class explains how GRPO works, with three hands-on experiments. The article shows where the same ideas appear at frontier scale
https://huggingface.co/blog/sergiopaniego/agentic-rl-2026 repliedto kanaria007's post 3 days ago
✅ Article highlight: *Responsibility Surfaces for Governed Intelligence Failures* (art-60-281, v0.1)
TL;DR:
This article argues that “the model failed” and “a human approved it” are usually incomplete sentences.
A governed failure often crosses several layers. 281 separates responsibility into six surfaces: causal, admissibility, control, override, disclosure, and reliance. The goal is not symbolic blame. It is to identify which layer owned which part of the failure path, with evidence.
Read:
https://huggingface.co/datasets/kanaria007/agi-structural-intelligence-protocols/blob/main/article/60-supplements/art-60-281-responsibility-surfaces-for-governed-intelligence-failures.md
Why it matters:
• separates who introduced harm from who allowed it
• distinguishes approval from live control
• makes explicit overrides visible
• treats public claims and release notes as part of system behavior
• prevents bounded audits from being reused as unlimited guarantees
• allows shared, contested, or cleared responsibility instead of forced singular blame
What’s inside:
• six responsibility surfaces: causal, admissibility, control, override, disclosure, and reliance
• responsibility-surface maps
• failure-allocation reports
• reliance-boundary notes
• a workflow that identifies surfaces before naming owners
• anti-patterns such as single-throat mythology, assessor laundering, board sanctification, and release-note innocence
Key idea:
Do not ask only:
*“Who caused the failure?”*
Ask:
*“Who introduced the harmful condition, who made it admissible, who could still stop it, who overrode the guard, who shaped the public interpretation, and who invited downstream reliance?”*
A governed failure is rarely one broken actor.
It is usually a broken path across layers.
repliedto melvindave's post 3 days ago
What's the best small model for training on tool calls?
Basically, the use case is to use it with an agent which processes a customer's inquiry and suggest a travel itinerary via an API call. I'm thinking this will just be wrapped as a tool call). It should be smart enough also to understand when to call multiple tools to achieve its goal.
Any ideas on this? I need to do a POC with a small model which can run in my RTX 3090. Planning to use unsloth for the fine-tune.
Just wanted to know which model I can use as baseline for this use case.Organizations
None yet