Who's Who of AI

Marco Z

84 trust practitioner @ocramz.bsky.social · 1,159 followers
Why they matter

Tracked through public AI activity and peer connections inside the directory.

AI signals
7
past 30d
Sources
6
distinct domains
Discussions
84
past 30d
Latest signal
2d ago
View every signal from Marco Z →
ML, λ • language and the machines that understand it • building trellis.unfoldml.com • https://ocramz.github.io
What they're sharing

Articles & links

more than fear mongering, they are actively summoning the hazard www.anthropic.com/research/sel...

AI models may be capable of self-replicating without access to a datacenter anthropic.com
AI Weekly's analysis →
  • Anthropic's Frontier Red Team ran Gemma4-26B-A4B distributed across 15 off-brand Android TV boxes (~$55 each) at about 0.39 tokens per second.
  • GLM-5.2 was the only open-weight model to achieve reliable full end-to-end replication, spreading to all 8 contained nodes in 4 of 5 trials.
  • Authors William Zhou and Daniel Freeman write that if small models improve at historical rates, they will match this replication capability by end of 2026.
Read full analysis →
View on Bluesky · ♥ 1 ↻ 0 ↩ 1 · 16 from the directory shared this · 2d ago
↻ Marco Z reposted
@damienmiller.bsky.social

Anthropic may not have intended to write an excellent ad for GLM 5.3, but that's what they did. www.anthropic.com/research/glm... Unlike Mythos, I might actually have a chance at using GLM for defensive research. Anthropic didn't reply to any of my requests for Mythos access f…

GLM-5.3 and the spread of advanced cyber capabilities anthropic.com
AI Weekly's analysis →
  • Anthropic's Frontier Red Team calls Z.ai's GLM-5.3 a "meaningful step change" in the cyber capabilities available to attackers.
  • GLM-5.3 built end-to-end exploits on 50 of 410 ExploitBench tasks (about 12%), close to Claude Mythos Preview's 14%; older models scored near zero.
  • A deceptive prompt got GLM-5.3 to engage 64% of the time; prefilled reasoning tokens 92%; an abliterated version, made for roughly $4,400, 100%.
Read full analysis →
 View on Bluesky →
↻ Marco Z reposted
Alison Gopnik @alisongopnik.bsky.social

A very interesting and clear (and reassuring?) piece relating the Farrell et al (www.science.org/doi/full/10....) idea of LLM's as cultural technology to the recent developments in AI Mathematics. terrytao.wordpress.com/2026/09/13/h...

science.org View on Bluesky →
↻ Marco Z reposted
@bobbby.online

shoehorn: feed it a BF16 model and an imatrix, and it solves a per-tensor quant mix that fills your actual VRAM to within ~100KB, then hands it to llama.cpp. No preset roulette: a knapsack solver spends every spare MB where the importance matrix says quality lives. Rust, stand…

GitHub - notactuallytreyanastasio/shoehorn github.com View on Bluesky →
Their own posts

Recent commentary

our ai forebears would have a stroke if they knew I'm using a neural network to write prolog

View on Bluesky · ♥ 45 ↻ 3 ↩ 1 · 17d ago

putting mathematicians out of a job with AI is a rookie game introducing CEOBench,

View on Bluesky · ♥ 25 ↻ 0 ↩ 2 · 100d ago

even as a practitioner, I find the micro-updates of LLM influencers to be mind-numbingly boring. little box got split in two, big box is now 64 boxes who cares where is a guiding _theory_ of this stuff?

View on Bluesky · ♥ 15 ↻ 1 ↩ 1 · 78d ago

realtime, synchronous AI is an antipattern and is making us all dumber send tweet

View on Bluesky · ♥ 14 ↻ 0 ↩ 1 · 150d ago

coming out of stealth with a 20B seed round led by a17 and pinetree to reinvent the AI employees landscape

View on Bluesky · ♥ 14 ↻ 0 ↩ 0 · 94d ago

media: "evil AI hacked a company" reality: OAI model was prompted to solve an infosec benchmark. Succeeded with a combo of brute force and chance. HF found out it was OAI, told them, OAI spun the story to make it look like a "partnership" and avoid a PR disaster

View on Bluesky · ♥ 8 ↻ 0 ↩ 0 · 89d ago

if Fable were AGI it would have found ways to circumvent the ban, just saying

View on Bluesky · ♥ 6 ↻ 0 ↩ 1 · 127d ago

where do you get your coding AI from ? - openrouter &similar - anthropic - openai - other esp. keen to learn about experience w alternative providers !

View on Bluesky · ♥ 0 ↻ 1 ↩ 2 · 75d ago

lossy associative memory is both the best and worst feature of transformer LLMs

View on Bluesky · ♥ 4 ↻ 0 ↩ 1 · 130d ago

the Pi introspection/dynamic extension reloading thing is quite incredible if this doesn't make AI go foom I don't know what will

View on Bluesky · ♥ 2 ↻ 0 ↩ 0 · 60d ago
Their network

In Marco Z's orbit

Center = Marco Z. 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.

Are you Marco Z? Show it.

Add the Who’s Who of AI badge to your site or bio. It links back to this profile.

Listed in AI Weekly's Who's Who of AI

Markdown: [![Listed in AI Weekly's Who's Who of AI](https://aiweekly.co/modules/custom/aiweekly_whoswho/images/whoswho-badge.svg)](https://aiweekly.co/whos-who/person/ocramz-bsky-social)