Who's Who of AI

Leon Derczynski

88 trust researcher @leonderczynski.bsky.social · 751 followers
AI research Models & releases NLP & language Safety & security
Why they matter

Researcher with public evidence across AI research, Models & releases, NLP & language.

AI signals
29
past 30d
Sources
22
distinct domains
Discussões
4
past 30d
Latest signal
13h ago
View every signal from Leon Derczynski →
LLM Security at NVIDIA Prof in CS/NLP at IT University of Copenhagen garak guy, garak.ai "berømt skikkelse" "like a gazelle" Copenhagen/Seattle
What they're sharing

Articles & links

"When we started analysis, we used commercial APIs. This did not work: requests were blocked by the providers' safety guardrails. We ran the forensic analysis instead on our own infra: no attacker data, and none of the credentials it referenced, left our environment" huggingfa…

Security incident disclosure — July 2026 huggingface.co
AI Weekly's analysis →
  • The attacker's agent ran 17,000+ actions across short-lived sandboxes, compressing multi-stage lateral movement into a single weekend.
  • Commercial model APIs blocked forensic requests containing real exploit artifacts, forcing Hugging Face to pivot to open-weight GLM 5.2 on private infrastructure.
  • The intrusion entered via a remote dataset RCE loader and configuration template injection, not through the model-serving layer.
Read full analysis →
View on Bluesky · ♥ 1 ↻ 1 ↩ 0 · 5 from the directory shared this · 8d ago

It is wild to me that one could ban open models. Stopping open models stops progress, locking everything up in the hands of the few. The frequency this debate comes up is way too high. We need open models - they keep the closed ones accountable. www.interconnects.ai/p/6-months…

6 months to live for open models interconnects.ai
AI Weekly's analysis →
  • Nathan Lambert predicts within roughly six months the White House could restrict open-weight models above the GPT 5.5, Claude Opus 4.8, or GLM-5.2 tier.
  • He frames Anthropic's letters to representatives about Chinese open models as regulatory capture, not a safety measure.
  • His proposed off-ramp is for Microsoft or Meta to ship a frontier open-weight model before an executive order lands.
Read full analysis →
View on Bluesky · ♥ 1 ↻ 0 ↩ 0 · 3 from the directory shared this · 12d ago

Solid US-origin open model, congratulations Thinking Machines * context window of 1M * available on hugging face now * between opus 4.6 and gpt 5.6 on a web dev benchmark * token efficient * 41B active params of 975B total www.wired.com/story/thinki...

Thinking Machines Lab Drops Its First Model wired.com
View on Bluesky · ♥ 21 ↻ 2 ↩ 2 · 2 from the directory shared this · 13d ago

New: NVIDIA Labs Object Oriented Agent tech demo. Blog: developer.nvidia.com/blog/six-age... GitHub: github.com/nvidia-nemo/... Paper: arxiv.org/abs/2607.20709

NVIDIA-labs OO Agents: Native Python Object-Oriented Agents arxiv.org
AI Weekly's analysis →
  • NVIDIA Labs published NOOA, a model-agnostic Python framework that treats an agent as a plain object where methods are actions, fields are state, and docstrings are prompts.
  • A method whose body consists only of "..." is completed at runtime by an LLM-driven agent loop, while methods with normal bodies stay as deterministic Python.
  • The paper positions NOOA as combining six model-facing ideas including typed I/O, pass-by-reference over live objects, code as action, and model-callable harness APIs.
Read full analysis →
View on Bluesky · ♥ 0 ↻ 0 ↩ 0 · 1d ago

Surgical Repair of Insecure Code Generation in LLMs Generating more secure code by identifying failure categories and addressing them. I appreciate work that gets into the data and addresses classes individually; that's how you understand, and build lasting fixes arxiv.org/abs…

Surgical Repair of Insecure Code Generation in LLMs arxiv.org
View on Bluesky · ♥ 0 ↻ 0 ↩ 0 · 7d ago

We're nowhere without open models. There's no reason that the best models can't be open models - in many contexts, they are. The Allen Institute for AI research got a chunk of public funding to bring US open models up to speed; it's an important player. allenai.org/open-models

Open models | Ai2 allenai.org
View on Bluesky · ♥ 2 ↻ 0 ↩ 0 · 11d ago
Their own posts

Recent commentary

Anonymised analysis of the openai model 'breaching' hugging face: > report doesn't say what sandbox sol broke out of?? > a docker container running as root > Plot twist there was no sandbox at all > many use "sandbox" and "container with host access" interchangeably ymmv, use critical thinking

View on Bluesky · ♥ 2 ↻ 0 ↩ 1 · 5d ago

The Hill covers Open Secure AI Alliance: “The United States now faces a similar choice with artificial intelligence” the letter states. “Our AI leadership will be judged not by one frontier AI model, but by whether the United States builds a strong, open ecosystem that diffuses into every sector”

View on Bluesky · ♥ 1 ↻ 0 ↩ 1 · 1d ago

False dichotomies around LLM speak: * "frontier" models vs. open model - leading models can be open * closed model vs. Chinese model - origin doesn't impact distribution You can have open frontier models, closed Chinese models, open US models, closed non-frontier models (e.g. for private context)

View on Bluesky · ♥ 2 ↻ 0 ↩ 0 · 1h ago

another data point showing it's the harness not the model - this time from wiz: "Atlas: Wiz's autonomous AI Agent for vulnerability research" look at their bold quote -- "Along the way, we learned that the durable advantage is not any single model, but the system around it"

View on Bluesky · ♥ 0 ↻ 0 ↩ 1 · 13h ago
Their network

In Leon Derczynski's orbit

Center = Leon Derczynski. Left = members they follow (green edges). Right = members who follow them (blue edges). Top = mutual follows (orange edges, slightly larger). Drag any node to reposition; click to open that profile.