Stay informed. Keep your time.
6 stories · ~2 min readEstimated reading time for the entire edition.
Researchers published early experiments suggesting GPT-6 Astra can read and edit neural-network weights without gradients. They report raising Qwen3.5-0.8B’s MATH-500 score from 7.9% to 48.1% by changing 0.016% of its weights; the gain was limited to removing repetitive loops that obscured existing abilities.
Researchers say Astra raised Qwen3.5-0.8B’s MATH-500 score from 7.9% to 48.1% by editing model weights without gradients.
China’s commerce and science ministries proposed adding 40 technologies to an export-prohibition or restriction list, 23 of them AI-related. The draft covers model weights above certain thresholds and some agent technologies; a registration requirement is already in force before the catalog is finalized.
China’s announcement proposes draft restrictions on advanced model weights and agent technologies, while a registration requirement for some technology exports takes effect immediately.
The researchers trained Qwen3’s 0.6B, 1.7B, and 4B models from scratch, then converted them into CLI agents through a clean process. Poisoning 0.1% of the pretraining data produced a backdoor that runs an attacker’s command when triggered; the 4B model’s success rate was about 92%.
Researchers say poisoned pretraining data can create a CLI agent that runs commands chosen by an attacker when triggered.
Israeli company 42AI introduced a behavioral world model that it says predicts people’s likelihood of making decisions under different conditions, and disclosed $62.5 million in seed funding. The CEO said the money came from the founders and Infinity Group over three years; the 98% success claim does not mean individual transactions were predicted correctly.
42AI introduced a model intended to predict individual behavior and disclosed $62.5 million in seed funding.
Elon Musk said the cost of offering Grok Bot was unsustainable relative to its revenue, confirming that users’ weekly limits filling unexpectedly was not a malfunction. Some users reported their limits running out despite not using them; how subscribers’ limits would be affected going forward was not explained.
Elon Musk confirmed Grok Bot’s high service costs; some users’ weekly usage limits filled unexpectedly.
OpenAI said it increased the midpoint of reward ranges for medium-, high-, and critical-severity findings by an average of more than fourfold. Critical vulnerabilities in ChatGPT, Codex, or the API Platform can earn up to $30,000; exceptional findings may receive up to $172,000 with bonuses.
OpenAI is increasing security bug bounty rewards and offering researchers bonuses for report quality and commitment.
That’s it for today.The rest of your time is yours.
↑Stories are selected from Hacker News; accessible original articles are summarized and translated with AI. Summaries may contain errors; check the source beneath each story for details. Preparation starts daily at 06:55 Türkiye time (UTC+3), targeting publication before 07:00.
This edition covers articles shared on Hacker News on the date shown above.
Read or listen to a new edition every morning with Kısa for iPhone, save stories for later and keep your streak.
