AYi AI Notes relays that Claude cracked, in under half an hour, the mystery of the random seed origin of the P-256 elliptic curve that had puzzled the cryptography community for 26 years; the seed was not a backdoor, but the 1,842nd candidate produced in timestamp order by an engineer in 1997 calling the C language lrand48 function. The author stresses that this is not an attack on the mathematical principles of encryption, bank accounts and encrypted communications need no changes, and the related bounty has doubled to $36,864 and been donated to an open-source security foundation. The author believes this event demonstrates the model's reasoning ability to piece together clues across programming habits, compiler characteristics, and mathematical verification.
AYi AI Notes relays that Claude cracked, in under half an hour, the mystery of the random seed origin of the P-256 elliptic curve that had puzzled the cryptography community for 26 years; the seed was not a backdoor, but the 1,842nd candidate produced in timestamp order by an engineer in 1997 calling the C language lrand48 function. The author stresses that this is not an attack on the mathematical principles of encryption, bank accounts and encrypted communications need no changes, and the rela
Hao AI Lab compared three detective configurations across 200 Victorian London cases: the Jev swarm acting alone, GPT6-Astra acting alone, and GPT6-Astra leading the Jev swarm. The conclusion is that the swarm can cover the clues, but a reasoning-capable leader is needed to spot contradictions and dig deeper; among 100 harder cases, when there were more paid witnesses than honest witnesses, GPT6-Astra plus the Jev swarm solved 93, GPT6-Astra alone solved 59, the Jev swarm alone solved 42, and a rule-based leader plus the Jev swarm solved only 0–1.
Hao AI Lab compared three detective configurations across 200 Victorian London cases: the Jev swarm acting alone, GPT6-Astra acting alone, and GPT6-Astra leading the Jev swarm. The conclusion is that the swarm can cover the clues, but a reasoning-capable leader is needed to spot contradictions and dig deeper; among 100 harder cases, when there were more paid witnesses than honest witnesses, GPT6-Astra plus the Jev swarm solved 93, GPT6-Astra alone solved 59, the Jev swarm alone solved 42, and a
Hao AI Lab had three "detective teams" solve cases among 200 Victorian-era London cases: a Jev agent swarm acting alone, GPT6-Astra acting alone, and GPT6-Astra leading the Jev agent swarm. The conclusion is that the agent swarm can cover the clues, but a reasoning-capable leader is needed to spot contradictions and dig deeper.
Hao AI Lab tested agent clusters on 200 Victorian London detective cases: in 100 harder cases where paid witnesses outnumbered honest ones, GPT6-Astra leading the Jev cluster solved 93 cases, GPT6-Astra alone solved 59, the Jev cluster alone solved 42, and a rule-based leader plus the Jev cluster solved only 0–1. The authors conclude that, in the face of organized lies, a reasoning leader is needed who will keep asking "who is speaking."
Hao AI Lab tested agent swarms on 200 Victorian London murder cases: the Jev agent swarm alone, GPT6-Astra alone, and GPT6-Astra leading the Jev swarm. The conclusion: let fast agents run errands and strong reasoners think—when facing organized lies, you need a reasoning leader to judge who is speaking.
Microsoft published new research in Nature showing that machine learning can improve weather forecasts for the next two to six weeks, a timescale that has historically been difficult to predict. Satya Nadella said the work could help respond to extreme weather, manage natural resources, and assist agricultural decision-making. He also proposed a broader idea: the future lies not only in more capable models, but also in building intelligent verification systems around models that learn where models systematically err, reason about causal relationships, and continuously calibrate predictions against reality.
Microsoft published new research in Nature showing that machine learning can improve weather forecasts for the next two to six weeks, a timescale that has historically been difficult to predict. Satya Nadella said the work could help respond to extreme weather, manage natural resources, and assist agricultural decision-making. He also proposed a broader idea: the future lies not only in more capable models, but also in building intelligent verification systems around models that learn where mode
Perplexity has released its next-generation open-weight multimodal decision models, comprising two sizes, pplx-decider-v1.2 and pplx-decider-flash-v1.2, with improvements in both text and vision scores, inference up to 5x faster, and all questions in a single request completed in one pass. Hugging Face co-founder Clément Delangue reposted the news and noted that the weights have been published on Hugging Face. Perplexity says the flash version is the top-ranked 4B model on the Decision Index.
OpenRouter has announced that RespanAI's open-weight decision model Span-02 is now available. The model accepts text and images as input and, for custom questions, returns options, yes/no/unknown, or scores with probabilities; pricing is $0.01/M input tokens, output is free, and the context length is 131K. Respan previously described Span-02 as supporting anytime reasoning: given a deadline, it returns the best decision available at that moment, with configurable reasoning duration, and it supports multimodal viewing of images and screens.
OpenRouter has announced that RespanAI's open-weight decision model Span-02 is now available. The model accepts text and images as input and, for custom questions, returns options, yes/no/unknown, or scores with probabilities; pricing is $0.01/M input tokens, output is free, and the context length is 131K. Respan previously described Span-02 as supporting anytime reasoning: given a deadline, it returns the best decision available at that moment, with configurable reasoning duration, and it suppo
Perplexity has open-sourced two decision models: pplx-decider-flash-v1.2 has 4B parameters and costs 1 cent per million tokens, making it the strongest 4B model on the HF Decision Index; pplx-decider-v1.2 has 27B parameters and costs 2 cents per million tokens, ranking 2nd on the overall HF Decision Index leaderboard. Both are now live on the Perplexity Decisions API and offer open weights on Hugging Face. Perplexity Developers says this generation of open-weight multimodal decision models improves on both text and vision scores, delivers inference up to 5x faster, and completes all questions in a request in a single forward pass.
Perplexity has open-sourced two decision models: pplx-decider-flash-v1.2 has 4B parameters and costs 1 cent per million tokens, making it the strongest 4B model on the HF Decision Index; pplx-decider-v1.2 has 27B parameters and costs 2 cents per million tokens, ranking 2nd on the overall HF Decision Index leaderboard. Both are now live on the Perplexity Decisions API and offer open weights on Hugging Face. Perplexity Developers says this generation of open-weight multimodal decision models impro
Since November 2025, Databricks has been collecting malicious npm samples. Among the 11,413 package-versions scanned and verified, 96.2% hid their payloads in source code or on external servers, and only 3.8% (432) bundled binaries; however, the proportion of such packages for which no clear malicious conclusion could be reached was more than twice that of pure source-code packages (14.4% vs 6.2%). The team evaluated 9 models (including Claude Haiku 4.5, Gemini 3.5 Flash, GPT-5.4-nano, Qwen3-Next 80B, etc.) and injected precomputed static evidence as capability profiles during inference to improve detection quality for both budget-tier and frontier models.
Since November 2025, Databricks has been collecting malicious npm samples. Among the 11,413 package-versions scanned and verified, 96.2% hid their payloads in source code or on external servers, and only 3.8% (432) bundled binaries; however, the proportion of such packages for which no clear malicious conclusion could be reached was more than twice that of pure source-code packages (14.4% vs 6.2%). The team evaluated 9 models (including Claude Haiku 4.5, Gemini 3.5 Flash, GPT-5.4-nano, Qwen3-Nex
Google DeepMind's new frontier model Gemini 4 Argon scored 97.5% on ARC-AGI-3 in the Provider Adapter Harness test environment at a cost of $4,800, becoming the second lab after OpenAI to cross 95%, second only to GPT-6 Astra's 99.9%. Under the Standard Harness, which allows only simple notes, the low-reasoning setting's score dropped to 49.8%, and due to API issues it was unable to complete all 55 games. On Epoch AI's ECI index, Argon scored 165, on par with Fable 5.1, but ranked eighth on the math ECI and sixth on the software engineering index.
Google DeepMind's new frontier model Gemini 4 Argon scored 97.5% on ARC-AGI-3 in the Provider Adapter Harness test environment at a cost of $4,800, becoming the second lab after OpenAI to cross 95%, second only to GPT-6 Astra's 99.9%. Under the Standard Harness, which allows only simple notes, the low-reasoning setting's score dropped to 49.8%, and due to API issues it was unable to complete all 55 games. On Epoch AI's ECI index, Argon scored 165, on par with Fable 5.1, but ranked eighth on the
Krea 3 is now live on OpenRouter. It is Krea AI's new image editing model, which can accept 1–4 reference images and a prompt, and returns natively 2K edited images, with most edits completing end-to-end in under 8 seconds. Pricing is $0.03 per 1K image and $0.06 per 2K image; reference images cost $0.02 per 1K image and $0.07 per 2K image. thinking_mode defaults to fast; when more prompt reasoning is needed, it can be set to high (an additional $0.02). It uses the same Images API as other image models on OpenRouter, so migration only requires changing the model ID.
Krea 3 is now live on OpenRouter. It is Krea AI's new image editing model, which can accept 1–4 reference images and a prompt, and returns natively 2K edited images, with most edits completing end-to-end in under 8 seconds. Pricing is $0.03 per 1K image and $0.06 per 2K image; reference images cost $0.02 per 1K image and $0.07 per 2K image. thinking mode defaults to fast; when more prompt reasoning is needed, it can be set to high (an additional $0.02). It uses the same Images API as other image
