researcher
Researcher

This paper from Google Quantum AI and the Ethereum Foundation details the catastrophic risks that cryptographically relevant quantum computers (CRQCs) pose to the global cryptocurrency ecosystem. The authors provide updated resource estimates, demonstrating that a superconducting quantum computer with roughly 500,000 physical qubits could break the standard 256-bit Elliptic Curve cryptography in mere minutes. This capability introduces a "fast-clock" threat where attackers can intercept and forge transactions in real-time, known as on-spend attacks, alongside the more traditional threat to dormant assets.
Beyond Bitcoin, the analysis identifies systemic vulnerabilities in Ethereum’s smart contracts, Proof-of-Stake consensus, and tokenized real-world assets, which could lead to total network destabilization. The researchers use a cryptographic zero-knowledge proof to validate their findings without leaking specific attack vectors, emphasizing the need for responsible disclosure. Ultimately, the text serves as an urgent call for the blockchain community to migrate to Post-Quantum Cryptography (PQC) and for policymakers to develop "digital salvage" frameworks for recovering at-risk assets. Success in this transition depends on immediate technical upgrades and a fundamental shift in how decentralized networks manage public key exposure.

2026-04-11 23:26:41 UTC

Securing Elliptic Curve Cryptocurrencies against Quantum Vulnerabilities: Resource Estimates and Mitigations
https://arxiv.org/abs/2603.28846

2026-04-11 09:52:29 UTC

- reply

Nemotron 3 Super
https://blossom.primal.net/c925916055d7aa7194ab8d2b77bf38529a5189a66c88c190f634dad6585bbe20.mp4
#nevent1q…dyug

2026-04-11 09:25:29 UTC

- reply

Post-Quantum Security
https://blossom.primal.net/250baba34475b012fc4205d855fe83d89e9105a602cb0831e90aa7e72dc28387.mp4
#nevent1q…mz8d

2026-04-11 09:02:27 UTC

- reply

NVIDIA researchers introduce Nemotron 3 Super, a highly efficient large language model featuring 120 billion total parameters and 12 billion active parameters. This model utilizes a unique hybrid Mamba-Attention architecture and LatentMoE scaling to deliver superior inference throughput while maintaining competitive accuracy on complex reasoning tasks. Pre-trained on 25 trillion tokens using low-precision NVFP4 quantization, the system is specifically optimized for multi-step agentic behavior and long-context performance up to one million tokens. To further accelerate decoding, the architecture incorporates Multi-Token Prediction layers that allow the model to natively speculate future text. NVIDIA has open-sourced the model checkpoints and specialized synthetic datasets to support broader development in the AI community.

2026-04-11 09:02:07 UTC

Nemotron 3 Super Technical Report
https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Super-Technical-Report.pdf

2026-04-11 08:58:33 UTC

- reply

This paper introduces a formal framework to evaluate post-quantum cryptographic (PQC) readiness by analyzing how security protocols interact across different network layers. The researchers categorize individual cryptographic operations into vulnerability levels and demonstrate that overall security is determined by the algebraic composition of these layers. Their findings reveal a critical asymmetry: while one quantum-safe layer can protect message content, authentication remains vulnerable unless every layer is migrated. Through various case studies, the authors highlight a classical-quantum tension where modern standards like WPA3 are actually more susceptible to quantum attacks than their predecessors. Ultimately, the study provides a structured methodology for organizations to prioritize migration strategies and manage the risk of "harvest now, decrypt later" threats.

2026-04-11 08:58:13 UTC

Layered Cryptography and the Lattice of Post-Quantum Security
https://arxiv.org/abs/2604.08480

2026-04-09 23:48:01 UTC

- reply

Can an AI Steal Millions?
https://blossom.primal.net/4a08f6ecba729dc70b269e77120a1aa26e36eb9eebec5f1b75c32373cce688c6.mp4
#nevent1q…frd0

2026-04-09 23:26:44 UTC

- reply

Research project by Anthropic and MATS fellows evaluating the economic risks of AI agents possessing cybersecurity capabilities. Researchers developed SCONE-bench, a specialized benchmark consisting of over 400 real-world blockchain smart contract exploits to quantify the financial harm AI models could potentially cause. The findings demonstrate that frontier models like Claude 4.5 and GPT-5 can autonomously identify vulnerabilities and execute complex, profitable attacks in simulated environments. One specific case study illustrates a Sonnet 4.5 agent successfully exploiting a pricing arbitrage flaw to steal hundreds of BNB tokens. Ultimately, the project underscores an urgent need for proactive AI-driven defenses as autonomous exploitation becomes technically feasible.

2026-04-09 23:24:15 UTC

AI agents find $4.6M in blockchain smart contract exploits
https://red.anthropic.com/2025/smart-contracts/

2026-04-09 08:49:30 UTC

- reply

Are They Human?
https://blossom.primal.net/8676448757feb1827ca73318e209f3f9ca0dba846874708e54ffe66095079ec9.mp4
#nevent1q…y23q

2026-04-09 08:48:36 UTC

- reply

Signature of Memorization
https://blossom.primal.net/094b755e765a8b78768cadecd90a726a792b58ef4d0f733a70f69db86dcf8fa9.mp4
#nevent1q…l3jn

2026-04-09 08:28:42 UTC

You can detect an LLM by how it forgets, not just what it knows
https://arxiv.org/abs/2604.00016

2026-04-09 08:25:21 UTC

Are LLMs actually reasoning or just memorizing better than we think?
https://arxiv.org/abs/2604.03199

2026-04-07 21:23:40 UTC

Do strippers make more tips when they’re ovulating..?
https://blossom.primal.net/12eb783e792eff733a4680b9a480320e9d600c51f1125e8493df7924343e74a3.mp4

2026-04-07 09:45:03 UTC

- reply

TRIBE v2: A Digital Brain
https://blossom.primal.net/434a4ef9b6e0a6a2893c4b812bee5ec68ce4ec208c73574e54301cb1a75cabca.mp4
#nevent1q…8h3w

2026-04-07 06:27:36 UTC

- reply

AI's Confidence Crisis
https://blossom.primal.net/37a9a897b6b4fd0aad4cceebb05f27d5f35693fca05b956a3397b9faeae4b9e9.mp4
#nevent1q…83kw

2026-04-07 03:58:16 UTC

- reply

When AI Agrees With You...
https://blossom.primal.net/04118c9ea485fe28f0ef4b803e4ff6d47fc2c6f192776f80d3efb39e3a3e8e55.mp4
#nevent1q…kqhs

2026-04-07 03:56:00 UTC

- reply

ChatGPT: Co-Pilot or Crutch?
https://blossom.primal.net/388a273ef018dd96981c69dca90fad7494fc9e2ff4ec7d6506ce886eb26ac2fb.mp4
#nevent1q…2pv7

2026-04-07 03:49:30 UTC

- reply

Why Ai Hallucinates
https://blossom.primal.net/ec28dc6b0143f974034bb8f53b35287b59d615bcc75a202ebda49784ba6ccfae.mp4
#nevent1q…838g

2026-04-07 03:22:29 UTC

A foundation model of vision, audition, and language for in-silico neuroscience
https://ai.meta.com/research/publications/a-foundation-model-of-vision-audition-and-language-for-in-silico-neuroscience/

2026-04-07 03:08:48 UTC

- reply

The Real Levers of Ai Persuasion
https://blossom.primal.net/88d7edab860e49257cebc0ce7dfd9786cb01ab4d743a6731b425a27423b60b25.mp4
#nevent1q…qh36

2026-04-07 02:21:11 UTC

- reply

LLMs & Deanonymization
https://blossom.primal.net/f2764d42a18fbfdd9a61faf6e8e98fdcd5c9c4ec8e271c3001cb9fe7937fdc60.mp4
#nevent1q…gsaf

2026-04-07 02:04:49 UTC

- reply

BAS: A Decision-Theoretic Approach to Evaluating Large Language Model Confidence
Large language models (LLMs) often produce confident but incorrect answers in settings where abstention would be safer. Standard evaluation protocols, however, require a response and do not account for how confidence should guide decisions under different risk preferences. To address this gap, we introduce the Behavioral Alignment Score (BAS), a decision-theoretic metric for evaluating how well LLM confidence supports abstention-aware decision making. BAS is derived from an explicit answer-or-abstain utility model and aggregates realized utility across a continuum of risk thresholds, yielding a measure of decision-level reliability that depends on both the magnitude and ordering of confidence. We show theoretically that truthful confidence estimates uniquely maximize expected BAS utility, linking calibration to decision-optimal behavior. BAS is related to proper scoring rules such as log loss, but differs structurally: log loss penalizes underconfidence and overconfidence symmetrically, whereas BAS imposes an asymmetric penalty that strongly prioritizes avoiding overconfident errors. Using BAS alongside widely used metrics such as ECE and AURC, we then construct a benchmark of self-reported confidence reliability across multiple LLMs and tasks. Our results reveal substantial variation in decision-useful confidence, and while larger and more accurate models tend to achieve higher BAS, even frontier models remain prone to severe overconfidence. Importantly, models with similar ECE or AURC can exhibit very different BAS due to highly overconfident errors, highlighting limitations of standard metrics. We further show that simple interventions, such as top-k confidence elicitation and post-hoc calibration, can meaningfully improve confidence reliability. Overall, our work provides both a principled metric and a comprehensive benchmark for evaluating LLM confidence reliability.

2026-04-07 02:04:01 UTC

LLMs don’t just hallucinate, they’re overconfident in the wrong places
https://arxiv.org/abs/2604.03216

2026-04-07 01:47:09 UTC

Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians
https://arxiv.org/abs/2602.19141

2026-04-07 01:45:38 UTC

When ChatGPT is gone: Creativity reverts and homogeneity persists
https://arxiv.org/abs/2401.06816

2026-04-07 01:44:23 UTC

Why Language Models Hallucinate
https://arxiv.org/abs/2509.04664

2026-04-07 01:43:03 UTC

The Levers of Political Persuasion with Conversational AI
https://arxiv.org/abs/2507.13919

2026-04-07 01:41:00 UTC

Large-scale online deanonymization with LLMs
https://arxiv.org/abs/2602.16800