oh god the phone call was coming FROM INSIDE THE HOUSE openai.com/index/huggin... there I was thinking Chinese state actors had hacked @huggingface.co.web.brid.gy but uh, nope! it was fucking OpenAI failing to supervise their own models, which they'd intentionally not neutered…
Who's Who of AI
Liz Fong-Jones (方禮真)
Why they matter
Tracked through public AI activity and peer connections inside the directory.
- AI signals
- 11 past 30d
- Sources
- 10 distinct domains
- Discussions
- 32 past 30d
- Latest signal
- 3d ago
🇺🇸🏳️🌈🏳️⚧️🥄❄️🐆👩🏽💻 in 🇦🇺🇨🇦 working on 🍯⬢🔭
she/it
What they're sharing
openai.com
Introducing Claude Opus 5 anthropic.com
Meta Contractors Posed as Teens to Prompt Rival Chatbots About Suicide, Sex, and Drugs | WIRED wired.com
Security incident disclosure — July 2026 huggingface.co
[BUG] **EXTREME DANGER**: AskUserQuestion: "No response after 60s — continued without an answer" · Issue #73125 · anthropics/claude-code github.com
Ornith-1.0: Self-Scaffolding LLMs for Agentic Coding | DeepReinforce Blog | Jun. 2026 deep-reinforce.com
View on Bluesky →
GLM-5.2 - How to Run Locally | Unsloth Documentation unsloth.ai
Data centers turn to commercial aircraft jet engines bolted onto trailers as AI power crunch bites — cast-off turbines generate up to 48 MW of electricity apiece tomshardware.com
AI Influence Level (AIL) v1.0 danielmiessler.com
I'm Kenyan. I Don't Write Like ChatGPT. ChatGPT Writes Like Me. marcusolang.substack.com
Charity Majors on AI, Determinism, Instrumentation, & Eating Your Broccoli redmonk.com
View on Bluesky →
I tried to stop Google DeepMind's Pentagon deal. Then I quit. transformernews.ai
Articles & links
@anthropic.com strikes back against GPT Sol! Let's fucking gooooo. www.anthropic.com/news/claude-...
AI Weekly's analysis
→
- Anthropic launched Claude Opus 5 on July 24, 2026 at $5 per million input tokens, matching Opus 4.8's rate.
- On Frontier-Bench v0.1 Opus 5 scored 43.3%, versus 18.7% for Opus 4.8 and 33.7% for Fable 5.
- Opus 5 becomes the default on Claude Max but sits behind Mythos 5 on cybersecurity tasks, per Anthropic.
Read full analysis →
Liz Fong-Jones (方禮真) reposted
NEW: Meta paid hundreds of contractors to pretend they were kids—and then prompt rival chatbots like Gemini and ChatGPT to talk about subjects like suicide, sex, eating disorders, and self-harm. From @dmehro.bsky.social and @joelkhalili.bsky.social
AI Weekly's analysis
→
Read full analysis →
View on Bluesky →
yepppp, see also huggingface.co/blog/securit... where the blue team's frontier models wouldn't help them so they had to use GLM to do blue team work.
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 →
github.com/anthropics/c... ahh okay they've made it configurable in /config now
AI Weekly's analysis
→
- In Claude Code 2.1.198, AskUserQuestion auto-returned a 'No response after 60s' message and told Claude to proceed on its own judgment.
- The behavior was undocumented, missing from the changelog, and a regression from 2.1.196, which worked correctly per the report.
- Maintainer ThariqS said a release will expose the setting under /config with the timeout configurable and defaulting to off.
Read full analysis →
Liz Fong-Jones (方禮真) reposted
Ornith 1.0: full spectrum of open weights coding models from 9B to 397B The largest appears to rival Opus 4.7-4.8 All fine tuned from Gemma 4 and Qwen 3.5 but no, the cool part is how they made it… 👇 deep-reinforce.com/ornith_1_0.h...
unsloth.ai/docs/models/... -- requires some pretty severe quant
AI Weekly's analysis
→
Read full analysis →
yeah the article is not a ref to the correct thing. www.tomshardware.com/tech-industr... these things are monstrosities and what I mean when I say literal fucking jet engines
If you're a human, even if you're using AI assistance, you know to snip out the irrelevant bits when copy-pasting. A crawler working a list of 100 leads does not. (ps want to know what an AIL is? check out danielmiessler.com/blog/ai-infl...)
See also: marcusolang.substack.com/p/im-kenyan-...
Liz Fong-Jones (方禮真) reposted
Hear @monkchips.com talk w Charity Majors, Co-founder & CTO @honeycomb.io, about what AI is doing to engineering practices. Majors argues that non-determinism takes more discipline, not less, & every CEO who wants their #AI cookies has to eat their broccoli first. redmonk.com/…
Never count on Jeff Dean to take a material stance on ethics. www.transformernews.ai/p/i-tried-to...
Their own posts
Recent commentary
today one of our media editing suppliers asked for consent to create a deepfake of my voice and sample data reading a script with all the major phonemes and I went nope nope nope nope nope I am okay with finetuning voice recognition, but not letting a corporation literally put words in my mouth.
gaaaaaaah I've hit the point of pressing my yubikey or touchid becoming the blocking factor for my LLM productivity, and I'm sometimes _barely_ reviewing the requests before mechanically approving, except sometimes it does ask to push something I don't want pushed, so the gate is helpful but aaaaa
I get almost zero AIL-4/5 fully automated outreach to me on LinkedIn, and very little to my work/personal email addresses. Here's my secret: I embed refusal phrases into my extended bio at the bottom, so that anyone who tries to use an LLM to fully automatically draft outreach to me will fail.
talk 4 in the bag, out of 6 in a span of 5 weeks. just the big opener for LDX3's building for production stage next week, and a panel discussion on what AI-enabled teams will look like, and then I can take a breather and not give any talks for a few months!
it's fascinating how quickly the r's in strawberry discourse dropped off as soon as people just fucking conceded, no, you don't ask LLMs to think in letters, only tokens.
we live in a world where account execs can build from scratch and drive their own demos, because generative AI can handle the heavy lifting of building the domain-specific load generator and fault injector tailored to the client's industry and ontology (verbs/nouns/adjectives). what a world.
Second time keynoting a conference in regional NSW! I'm kicking off #SlashNEW tomorrow morning bright and early in beautiful Newcastle, after a lovely train ride through the Central Coast earlier today. I'll be discussing the optimist & pessimist cases for AI, and how not to ensloppify everything.
Their network
In Liz Fong-Jones (方禮真)'s orbit
Center = Liz Fong-Jones (方禮真). 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.