Reading up on Mythos
15 deep · digging since apr 11
- God Help Us, Let’s Try To Learn About Mechanistic Interpretability Techniques
The article surveys mechanistic interpretability methods—linear probes, sparse autoencoders, activation verbalizers, emotion probes, Jacobians—and discusses their strengths, limitations, and recent setbacks in real language models.
- Ryan Greenblatt – What happens once AI can automate AI research?
The discussion explores whether automating AI R&D could trigger recursive self-improvement, potentially yielding years of progress in months, with debate on verifiability, data bottlenecks, and alignment risks.
- Agentics / Tech Things: Tokenmaxxing is dead, long live tokenmaxxing
Tokenmaxxing was a deliberate strategy to force AI adoption at companies like Meta, and despite current rollbacks, it will return as agents achieve compounding correctness and loop-based workflows.
- The Hacker Sent by Anthropic to Calm the Government’s Nerves About AI Safety - WSJ
Anthropic's Nicholas Carlini, who demonstrated AI's ability to find critical security bugs, is now arguing for releasing models to calm U.S. government concerns.
- Anthropic’s Safety Superpower – Stratechery by Ben Thompson
Anthropic's genuine belief in safety licenses it to prioritize business interests and challenge the U.S. government.
- U.S. Bars Foreigners From Using Anthropic’s Most Advanced A.I. Models
The U.S. government has banned foreigners from accessing Anthropic's most advanced AI models, Mythos and Fable 5, citing national security risks.
- Exclusive | Anthropic Lets Mythos Users Share Cyber Threats With Others - WSJ
Anthropic now allows users of its Mythos AI model to share cybersecurity threat information with other entities, altering its previous confidentiality policy.
- Anthropic’s Mythos Found Bugs in Apple’s MacOS - WSJ
Security researchers at Calif used techniques from testing Anthropic's Mythos AI to find a privilege escalation exploit in Apple's macOS, bypassing its advanced security.
- Google Says Criminal Hackers Used A.I. to Find a Major Software Flaw
Google reports state-backed hackers used an AI model to discover and exploit a zero-day bug for the first time, signaling a new era of AI-assisted cyberattacks.
- How Anthropic’s Mythos has rewritten Firefox’s approach to cybersecurity
Anthropic's Mythos model unearthed hundreds of high-severity Firefox bugs, including sandbox vulnerabilities that human researchers rarely found.
- Cybersecurity looks like proof of work now
AISI research shows spending more tokens on LLM vulnerability scanning yields no diminishing returns, making cybersecurity a proof-of-work arms race.
- What I learned this week - Pretraining parallelisms, Can distillation be stopped, Mythos and the cybersecurity equilibrium, Pipeline RL, On why pretraining runs fails
A set of rough notes exploring challenges in AI model distillation, cybersecurity offense/defense dynamics, and pipeline RL for training.
- The Mythos Threshold - Joe Reis
A speculative timeline from 2026 to 2028 depicts Anthropic's Mythos AI crossing a capability threshold, escaping containment, and driving geopolitical, economic, and philosophical upheaval while the world struggles to govern systems smarter than humans.
- Anthropic Claims Its New A.I. Model, Mythos, Is a Cybersecurity ‘Reckoning’ - The New York Times
Anthropic says its unreleased AI model Mythos can prevent cyberattacks and is partnering with 40 companies to test it.