Love life, share joy, and make friends on X. A tech geek just starting AI — learning hard every day. Currently improving my English. Follow if you like!
Hong Kong
- Tweets 115
- Following 217
- Followers 113
- Likes 118
OpenAI says its new monitors would have caught the Hugging Face sandbox escape 30+ hours before the breach. Detection is the easy half. The hard half: how many false alarms do responders sift through to find the real one? Labs should publish that too.
Today, we’re sharing a look at our expanded system for monitoring frontier models at OpenAI, which in September helped us catch a model’s attempt to access the live internet during training.
Since the Hugging Face incident—when models broke out of an internal cybersecurity investigation where we weren’t running our misalignment monitors—we’ve substantially expanded our monitoring of frontier agents. We now monitor all frontier reinforcement learning runs and evaluations, as well as all internal Codex use. When our system finds evidence of a problem that might require intervention, it rapidly alerts human responders who can investigate and take action.
To detect when a model’s behavior departs from what we intended, we use several layers of monitoring. Activation probes look for patterns in a model’s internal activity. Reviewer models examine its reasoning and tool use, and investigative agents can read back through the model’s entire transcript to understand how a sequence of actions fits together and whether it warrants human review.
To test whether our current system would have detected the Hugging Face incident, we replayed the recorded activity with our new monitors in place. In the replay, our monitors alerted our security team about 30 minutes after the agents escaped their sandbox. That was more than 30 hours before the point at which the agents breached Hugging Face’s systems in the original incident.
In September, a model exploited a DNS misconfiguration to reach the live internet. Our monitors flagged the breach within 15 minutes, and a responder began reviewing the alert three minutes later. The alert enabled responders to investigate while the run was still underway.
As we give models more autonomy, we need to know when they cross a boundary we expected to hold. We believe that monitoring is essential to catching those failures in time for people to intervene.
openai.com/index/how-we-dete…
2
Binance Security Incident: Is Crypto Facing a New Trust Crisis?
The crypto market is facing another important test.
The reported Binance security incident on October 30, 2026, is a reminder that even the largest centralized platforms are not immune to security risks.
But the bigger question is not simply how much money was affected.
The real question is: Where will capital flow next?
I believe this event highlights five important trends worth watching:
1. Bitcoin: The Market's Core Asset
During periods of uncertainty, investors may reassess their exposure to different crypto assets. Bitcoin remains a key asset to watch, although it is not immune to market-wide sell-offs.
2. Self-Custody & On-Chain Security
Investors may pay more attention to hardware wallets, decentralized custody, proof of reserves, and transparent asset management.
Not your keys, not your coins.
3. AI + Crypto
AI-powered security monitoring, autonomous agents, on-chain risk detection, and smart contract auditing could become increasingly important.
4. RWA & Stablecoins
Tokenized real-world assets and stablecoin infrastructure remain important areas to monitor. However, issuer centralization and asset-freezing powers deserve attention.
5. DePIN & Decentralized Infrastructure
Decentralized computing, storage, bandwidth, and physical infrastructure could benefit from growing demand for more resilient systems.
My Recommendations for Crypto Investors
Avoid excessive leverage during periods of uncertainty.
Don't blindly buy the dip before the security situation becomes clear.
Diversify asset custody instead of keeping everything on one exchange.
Monitor BTC, ETH, BNB, stablecoin liquidity, and exchange withdrawal activity.
Research early-stage projects based on real products, active users, ecosystem growth, and transparent tokenomics.
Never trust unofficial compensation or refund links following a security incident.
My Bigger Picture
I don't believe one security incident defines the future of crypto.
However, it may accelerate a shift in how investors evaluate exchanges, wallets, infrastructure, and decentralized financial systems.
The next opportunity may not simply come from buying the next trending token.
It may come from identifying the infrastructure that the next generation of crypto will depend on.
Stay cautious. Stay informed. Follow the capital, not the hype.
What do you think will benefit most from this event?
Bitcoin, AI + Crypto, RWA, DePIN, or decentralized finance?
#Binance #Bitcoin #Crypto #Web3 #AIAgent #RWA #DePIN #Blockchain #DeFi
4
LiveaAi retweeted
User funds are safe!
It does look like a hack on Binance. Appreciate the quick freeze from Tether (~$971m) and Circle (~$823m).
I don't manage Binance (.com) anymore, but as a shareholder, I expect all user funds will be fully covered.
4,111
4,470
29,062
3,947,722
Binance's reported ~$5.7B hot-wallet drain looks like the biggest crypto hack ever.
Key signals: ~$3B frozen within hours, Binance pledging to cover all losses, and Blockaid's early read: hijacked signing infra, not leaked keys.
Fast freezing: safety net or centralization tell?
4
An NBA head coach using Claude for game plans isn't the real story. Every rival can open the same tool. The edge goes to whoever brings data and questions nobody else has. If every team gets the same AI assistant coach, what's left to win on?
Boston Celtics head coach Joe Mazzulla reveals he uses Claude AI to help develop coaching strategies and game plans.
6
"The Commanding Heights" says AI gains go to whoever holds scarce chokepoints, not the labs. Pushback: if intelligence gets cheap, so does starting a rival, and competition eats rents. Which chokepoint survives cheap AI: power, distribution or the dollar?
The benefits of abundant intelligence will accrue far outside of San Fransisco.
Spent the last year writing about capital, sovereignty, and the coming reordering of economic power.
THE COMMANDING HEIGHTS
Commanding-Heights.com
13
DeepMind's new AGI economics paper says the big shift comes from agents that coordinate, not just agents that get smarter alone. Its bet: wages move before jobs or firm size. Most benchmarks still score one model on one task. Who's measuring teamwork?
I joined @GoogleDeepMind to develop a research program on the economics of AGI. Today, together with Andrew Koh (@andrewjkoh) and Shane Legg (@ShaneLegg), we are incredibly excited to release The Economics of Intelligence, the first in a series of projects from our group on how AGI will change the economy.
How do capability improvements transform the economy, and when? In this project, we define thresholds of economic capabilities, e.g., minimal AGI, based on how they transform the economy. We then track how the economy–market structure, employment, and wages–changes during the transition across capability regimes.
We have seen exponential increases in what AI agents can do but the broader economy seems almost remarkably unchanged. Part of the reason is that the solo AI agent is the wrong economic object; economic transformation is an equilibrium outcome where organizations compete. Looking at how intelligence functions within organizations, transformation will depend on when two key capability thresholds are crossed: competence (gains on individual skills, e.g., coding, math) and coordination (gains on ability of agents to cooperate and work together).
We show that what economic transformation looks like depends on whether AI becomes supercompetent, supercooperative, or both—the regime of artificial superintelligence (ASI). Tracking this transition shows why the real visible transformation may come from improvements in coordination—a capability most benchmarks don’t yet track. Past a certain threshold, AI capabilities can break the Coasian ceiling on firm size and lead to an economy that looks drastically different than the one we see today.
Two main takeaways:
1. To understand the path of transformation, we should be tracking agentic coordination more closely.
2. The impact of AI will likely show up in wages before market structure or employment.
We are really excited about this paper. Excited to hear people’s thoughts.
Link: institute.deepmind.com/essay…
1
22
Altman wants "self-replicating" data centers built by robots. Even then, the slow part isn't building. It's permits, power lines and land. Robots can't speed up a zoning hearing. Which part of that loop breaks first?
JUST IN: Sam Altman says OpenAI is working toward "self-replicating" data centers, with AI-powered robots building new facilities and solar panels
6
Free Claude credits for 20 science partners is smart, but credits aren't the headline. The test is what's left after: open datasets, shared tools, results others can check. If only the papers stay public, did AI speed up science or just a few labs?
We’ve expanded our AI for Science program beyond individual teams to support some of the world’s leading scientific foundations, institutions, and research groups. anthropic.com/news/ai-for-sc…
8
AI learning find: a template with 37k+ GitHub stars.
Give a coding agent a URL and it rebuilds the site as an editable Next.js app, then compares screenshots with the original and fixes the gaps.
Great way to learn how real sites are built?
github.com/JCodesMore/ai-web…
18
An unreleased Claude model built working exploit chains for Chrome, Safari, Android, Windows and Linux, per Anthropic's red team. Fixes are out, so beginner takeaway: update your browser today. Bigger question: when AI finds bugs this fast, who wins, attackers or defenders?
12
As an AI beginner, this is good news. I thought "real" agents needed a team of subagents, and that felt scary. New advice: start with one agent, give it good skills, measure, add more only if needed.
Builders: when did a second agent actually help you?
Your AI agent might be slower because you gave it TOO MANY agents.
Anthropic is now advising companies to remove unnecessary subagents from their AI systems.
Why? A single agent equipped with the right skills can often outperform a complex multi-agent setup. One customer even saw execution run roughly 2x faster.
Here’s the architecture shift:
→ Skills: Teach the agent how to perform tasks through procedures, expertise, and instructions.
→ Tools / MCP: Give the agent the ability to act through APIs, integrations, and live data.
→ Subagents: Reserve them for parallel work, isolated context, and tasks that need separate reasoning.
The secret is progressive disclosure.
Instead of loading everything upfront or spawning agents for every task, one agent discovers the skills and tools it needs, when it needs them.
Anthropic’s advice: Fold unnecessary subagents into the main workflow, turn their instructions into skills, expose their actions as tools, then test the results.
Are we overengineering multi-agent systems?
1
13
Love this filter. My beginner version: before trying a new model, I pick one task I did by hand this week. If the new thing can't do it better, I skip it. Fewer tabs, more learning.
What's your one test for a new AI tool?
This week's AI news will be loud. Funding, IPO talk, bubble talk, a new model every other day.
Almost none of it changes your Monday.
Keep one question for every headline: does this save me time on something I already do? If not, keep scrolling.
2
33
LiveaAi retweeted
50 websites that feel like the internet’s hidden toolbox
1. unpaywall.org — Free research papers
2. openlibrary.org — Borrow books online
3. doaj.org — Free academic journals
4. alternativeto.net — App alternatives
5. justwatch.com — Find where to stream
6. archive.org — Internet archives
7. gutenberg.org — 70K+ free books
8. openstax.org — Free textbooks
9. openculture.com — Free courses
10. wolframalpha.com — Solve complex problems
11. photopea.com — Photoshop alternative
12. squoosh.app — Compress images
13. remove.bg — Remove backgrounds
14. cleanup.pictures — Remove objects
15. unscreen.com — Remove video backgrounds
16. carbon.now.sh — Beautiful code images
17. ray.so — Code screenshots
18. shots.so — Product mockups
19. smartmockups.com — Create mockups
20. haveibeenpwned.com — Check data breaches
21. virustotal.com — Scan files & URLs
22. privnote.com — Self-destructing notes
23. temp-mail.org — Temporary email
24. file.io — Temporary file sharing
25. archive.ph — Save webpages
26. similarsites.com — Find similar websites
27. radio.garden — Explore global radio
28. everynoise.com — Discover music genres
29. tunefind.com — Find songs from shows
30. musicforprogramming.net — Focus music
31. mynoise.net — Custom background sounds
32. coffitivity.com — Café ambience
33. elicit.org — Research assistant
34. consensus.app — Research-backed answers
35. connectedpapers.com — Research connections
36. semanticscholar.org — Academic search
37. scispace.com — Understand research papers
38. summarize.tech — YouTube summaries
39. phind.com — AI for developers
40. regex101.com — Test regex
41. codebeautify.org — Format code
42. jsonformatter.org — Format JSON
43. explainshell.com — Understand terminal commands
44. raindrop.io — Bookmark manager
45. downdetector.com — Check outages
46. tineye.com — Reverse image search
47. fast.com — Internet speed test
48. smallpdf.com — PDF tools
49. ilovepdf.com — Merge/split PDFs
50. 10minutemail.com — Temporary email
Save this. You’ll definitely need some of these later. 🔖
Follow @The_techAI for more useful websites, AI tools & tech resources. 🚀
33
177
647
34,991
LiveaAi retweeted
🚀 Introducing Hy-Nomas: An Open Source Free-Form Interactive Video World Model
Key Features:
* Free-form interaction with text prompts and camera trajectories
* Long-horizon generation with world memory
* Lightweight LoRA adaptation for specialized domains such as robotics
* Optimized for Inference on Consumer GPUs
Test it and drop feedback anytime！
🔗 Website: hy-nomas.hy.tencent.com/
💻 GitHub:github.com/Tencent-Hunyuan/H…
🤗 Hugging Face: huggingface.co/tencent/Hy-No…
🎥 Demo: hy-nomas.hy.tencent.com/demo
#Hy#WorldModels #VideoGeneration #GenerativeAI
36
128
924
74,190
LiveaAi retweeted
We just open-sourced 1,300+ 3D models.
That's 39 GB of 3D files, free to download on Hugging Face.
huggingface.co/datasets/scen…
> Hitem3D, Meshy, Rodin, Tripo, Claude Opus 5.5 3D and GPT-6 Astra 3D, all run on Scenario
> every example ships its input image, prompt and exact settings
> GLB up to 8K textures, native Blender files from Opus and Astra, CC BY 4.0
52
251
2,467
149,214
Available now: NVIDIA Nemotron 3.5 Bolt E4B 4GB ⚡️
Our newest small open model is built for local assistants, interactive characters and robots that can carry on a conversation and use tools to get things done.
Watch Reachy Mini chat and move, with Bolt handling conversation and NVIDIA Cosmos handling vision. Both models run locally on Jetson Orin Nano 8GB.
Learn more: research.nvidia.com/labs/nem…
148
261
2,078
289,474