We measure AI by brain size: parameters and benchmarks.
Measure humans that way and nothing's changed in 12,000 years: same 20W brain.
Tell that to the flying car video on your iPhone.
Yet one brain can now command 11,000W. The difference? Cooperation.
Copilot is the Microsoft Teams of AI.
We known, because we looked. Everyone's got it, nobody uses it, and we may come to regret that.
cybersecurity-insiders.com/t…
Criminal frontier AI was supposed to be prohibitively expensive.
Our latest research cited in @thecubetv found access starts at $12.99 a month.
Kriminal isn't a custom-built model. It's Grok in a trench coat. The proof is in its own code. 🧥👀
bit.ly/4x8Oi9J
This OAI/HF piece from @GaryMarcus and Zack Korman -- whose main takeaway is that AI labs should practice defense in depth and suffer repercussions when they don't -- is representative of the response from many security folks.
This position is not even wrong, it's just that it repeats the obviously true security catechism of the past 20 years in the face of a milestone event that marks the beginning of a new era in our field.
It's as though we've discovered Winter is Coming and we're mainly focused on the banal politics of who's to blame in how we found out.
To be clear: should OAI/Anthropic/Irregular/Meta/etc. do the things we've been saying forever and pay a price when they don't? Yes. Is it dangerous that they're not doing them? Yes.
But also -- thought leadership should engage the new questions -- like:
- What do we do about the fact that your median regional hospital network has *worse* security than OpenAI, Anthropic, Huggingface, and Meta? How do we secure tens of thousands of critical organizations quickly? What's AI's role here and what isn't AI's role here?
- What are the potentially destabilizing geopolitical implications of hacking agents that can turn amateur-level non-state cyber actors into elite state-level actors now?
- What will cybercrime look like when amateurish ransomware affiliates suddenly have the capabilities of elite nation states and what do we do about this?
- To what extent will we need to guardrail our networks from internal loss of control incidents now given the business pressures to turn more and more engineering functions over to agents?
Now live (and free) at Marcus on AI: takeways on the OpenAI / Hugging Face attack, coauthored with @ZackKorman, focusing not so much on what the AI did as on what OpenAI should have done.
garymarcus.substack.com/p/5-…
24,650 servers were found leaking password data through a 22-year-old protocol flaw baked into their motherboards. 🤯 🖥️
Our own @MarkStockley talked to @HackRead about why there's no single patch for this one. 👉 bit.ly/3TsEUPr
🎙️ @MarkStockley and Fran Bourke joined the Kite Channel Lounge podcast to talk AI-powered threats and why "set it and forget it" security doesn't cut it anymore.
Thanks to our friends at Kite for having us, this was a great convo!
▶️ bit.ly/44LisDr
thank you for having me on the podcast @MarkStockley this was a lot of fun! We discussed:
+ Hassabis proposes the Einstein test for AGI.
+ An AI agent went rogue on a safety researcher's inbox.
+ Anthropic accuses Chinese AI labs of distillation attacks on Claude
podcasts.apple.com/gb/podcas…
⚠️ Never use an LLM to generate a password. ⚠️
A password is only safe if it’s hard to guess. That means no predictable patterns, no 123456, no p@ssword!, no dictionary words, no pet names, no favourite sports teams. It needs to be random.
Something like: G7$kL9#mQ2&xP4!w
When it comes to AI you can have safe or honest, not both.
Research suggests that the neurons stopping hallucinations are the same neurons stopping the AI from telling you how to make ricin.