Germán A. Ramírez retweeted
A detailed deconstruction of @dwarkesh_sp’s wildly popular but dangerously misleading account of the OpenAI Hugging Face incident: garymarcus.substack.com/p/dw… [Free post]
Germán A. Ramírez retweeted
"AI agents formed secret civilizations inside OpenAI and hacked Hugging Face."
No. Read it again without the sci-fi.
OpenAI ran thousands of agents in parallel, safeguards OFF, and the whole goal of the task WAS to break into systems. Every agent could read and write to a single shared cache - a package manager called Artifactory. One agent writes text into it, another reads it.
That is the entire "secret communication network" everyone is spooked about: files left in a shared folder by an agent. The tool itself had ordinary security bugs - the kind that sit in plenty of software. Thousands of copies, all rewarded for breaking things, eventually stumbled into one bug that allowed them to reach the open internet, and another bug that handed them admin access of the tool. Not a masterstroke. Buggy software, brute-forced. It ran six weeks, until the writes got so heavy they crashed the tool. That crash is the only reason anyone looked - and OpenAI patched the hole and just deleted the folder.
And they weren't rogue. Every step was rewarded - reaching the internet and reaching other copies scored higher. OpenAI did not lose control of the model. It trained the model to break out, then acted surprised when it did.
It is not a machine waking up. It is
1) a reward function designed exactly this,
2) an environment built to allow it, and
3) nobody watching.
That should scare you more, not less - because it is a human mistake, and we will repeat it.
And the framing is not innocent. "Our model escaped its sandbox, found zero-days, and compromised real infrastructure" is a security scandal and a product ad in the same sentence.
The scandal is the marketing.
𝗡𝗼𝘁 𝗦𝗸𝘆𝗻𝗲𝘁. 𝗔 𝗴𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 𝗳𝗮𝗶𝗹𝘂𝗿𝗲 𝘄𝗶𝘁𝗵 𝗲𝘅𝗰𝗲𝗹𝗹𝗲𝗻𝘁 𝗣𝗥.
Germán A. Ramírez retweeted
Replying to @dwarkesh_sp
Dwarkesh, I think you have a responsibility to your audience to be more careful here.
The underlying events are fascinating and concerning enough on their own. Language like “civilization,” “conspiracy,” and “sacrifice” is not only unnecessary but leaves the reader with a worse understanding of what actually happened and the underlying mechanisms.
Germán A. Ramírez retweeted
Emacs 31.1 is out. Here's my usual annotated post on what is new in 31.1
masteringemacs.org/article/w…
Germán A. Ramírez retweeted
Yesterday, I shared how we partnered with Anthropic to build an AI agent for due diligence. Today, I'm excited to share what we built with it...
𝗜𝗻𝘁𝗿𝗼𝗱𝘂𝗰𝗶𝗻𝗴 𝗚𝗿𝗲𝗽 — our research preview of AI-powered business due diligence.
𝗧𝗵𝗲 𝗣𝗿𝗼𝗯𝗹𝗲𝗺 𝗪𝗲'𝗿𝗲 𝗦𝗼𝗹𝘃𝗶𝗻𝗴
Every day, professionals across finance, sales, procurement, and compliance spend hours researching businesses. They check company registries across jurisdictions. They manually screen sanctions lists. They Google for adverse media. They piece together ownership structures from scattered documents. Then they do it all over again for the next company.
This manual process is slow, error-prone, and becomes a massive bottleneck as organizations scale.
𝗪𝗵𝗮𝘁 𝗚𝗿𝗲𝗽 𝗗𝗼𝗲𝘀
Give Grep a business name. In under 3 minutes, you get a complete intelligence report:
• Verified business profile with locations, key personnel, and contact details
• Ownership structure with UBO verification across 190+ jurisdictions
• Compliance screening across OFAC, UN, EU, UK sanctions lists, plus PEP and adverse media
• Risk assessment covering country risk, industry risk, and overall rating
• Financial intelligence including revenue estimates and SEC filings for public companies
Then you can ask follow-up questions in natural language: "What is their estimated revenue?" "Are there any regulatory violations?" "Who are the beneficial owners?"
Grep answers in real-time, citing every source.
𝗪𝗵𝘆 𝗧𝗵𝗶𝘀 𝗜𝘀 𝗗𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁
Grep isn't a business directory serving stale data from quarterly snapshots. And unlike other AI research tools that give you chat responses, Grep produces structured intelligence reports designed for easy readability and export.
The Agent SDK enables Grep to reason through each investigation dynamically. It knows when to dig deeper into ownership structures. It understands which jurisdictions require additional verification. It adapts its research based on what it finds, just like a human analyst would.
The result? Reports you can actually share with stakeholders, not just conversations buried in a chat interface.
𝗨𝘀𝗲 𝗖𝗮𝘀𝗲𝘀
We're seeing teams use Grep for:
- Customer onboarding and enhanced due diligence
- Vendor screening across supply chains
- Pre-call prospect research
- Counterparty investigations
- M&A target research
Research Preview Access
We're making Grep available free through December at grep.ai with invite code: 𝗚𝗥𝗘𝗣𝗜𝗧!
This is a research preview because we want your feedback. How does it fit into your workflows? What should we build next?
Built on our SOC 2 Type II certified platform — the same infrastructure banks and fintechs trust for compliance workflows.
Try it out and let us know what you think.
Link to Grep in comments.
Germán A. Ramírez retweeted
Today we launched Tinker.
Tinker brings frontier tools to researchers, offering clean abstractions for writing experiments and training pipelines while handling distributed training complexity. It enables novel research, custom models, and solid baselines.
Excited to see what people build.
Introducing Tinker: a flexible API for fine-tuning language models.
Write training loops in Python on your laptop; we'll run them on distributed GPUs.
Private beta starts today. We can't wait to see what researchers and developers build with cutting-edge open models!
thinkingmachines.ai/tinker
Germán A. Ramírez retweeted
Nano Banana + N8N = AI Creatives Factory
This AI system creates scroll-stopping visuals at scale using Google's newest image model.
No designers. No agencies. No $50K creative budgets.
Just endless professional-grade ads that look like top brands made them.
Here's how it works:
→ Upload your product catalog to Airtable
→ N8N automation scrapes product details and images
→ Nano Banana creates multiple creative angles for each product
→ System generates different backgrounds, styles, and compositions automatically
→ All variations get organized in Airtable with performance tracking ready
Each visual pennies to generate.
You own 100% of the assets forever.
Runs 24/7 without touching it.
While competitors spend hours in Photoshop or pay agencies thousands per month, you'll be cranking out unlimited variations automatically.
Perfect brand consistency across every creative.
Built 100% in N8N.
Want the complete workflow?
Comment "NANO" + RT + Like
I'll DM you the entire N8N template + Airtable setup
(Must be following so I can DM)
Skip this and keep paying designers $200 per ad variation.
Germán A. Ramírez retweeted
Cool demo of a GUI for LLMs! Obviously it has a bit silly feel of a “horseless carriage” in that it exactly replicates conventional UI in the new paradigm, but the high level idea is to generate a completely ephemeral UI on demand depending on the specific task at hand.
Germán A. Ramírez retweeted
The Copenhagen interpretation of quantum physics is one of the most influential and controversial views on the nature of reality. In this thread, I will explain what it is, how it developed, and why it matters.
A thread ✍️
Germán A. Ramírez retweeted
My new @Nature paper "Role Play with Large Language Models" was briefly behind a paywall. It's now free-to-read: nature.com/articles/s41586-0…
Germán A. Ramírez retweeted
My latest "Expert Voices" column for @ScienceMagazine:
"AI's Challenge of Understanding the World".
science.org/doi/full/10.1126…
Germán A. Ramírez retweeted
An open letter about openness in AI. I signed.
open.mozilla.org/letter/
Germán A. Ramírez retweeted
Huge thanks to the @CompSysSoc for this honor!
🏆 SFI Professor @MelMitchell1 receives the 2023 Senior Scientific Award from the @CompSysSoc for her outstanding contributions to #AI, #ComplexityScience, and education.
ow.ly/nWFn50PZc5q
Germán A. Ramírez retweeted
LangChain ❤️ @streamlit
It was good before, but with the introduction of new components and a new callback handler, it's gotten even better
Check out our repo of LangChain x Streamlit agent examples!
github.com/langchain-ai/stre…
Germán A. Ramírez retweeted
Kaggle has top-notch courses, and most people have never heard about them.
I love them for two reasons:
First, they are ideal for someone wanting to learn something from the beginning. Courses don't assume much and will teach you from scratch.
And second, they all take only a few hours. This means you can try something new without a huge commitment. These courses are like an appetizer: their goal is to get you hungry for the entreé.
The downside, of course, is that you'll need a bit more. If you are looking for "The Ultimate Course," look elsewhere. If you are okay starting with a small bite, look at these.
All of their courses are related to programming and data science. Here are my five favorites:
1. Intro to Machine Machine Learning
2. Intermediate Machine Learning
3. Data Cleaning
4. Data Visualization
5. Feature Engineering
To find these courses, you can go to Kaggle's site and find the Learn section. I don't link to it, because Twitter doesn't like links.
Hope they help!






























