이번주 AI 동향: 수리 난제 성과·실시간 음성 스택·취약점 급증

An internal version of our next major model produced 10 new results on long-standing open problems in mathematics and theoretical computer science, using roughly $2,000 worth of tokens at GPT-5.6 Sol API rates.

The results span sphere packing, coding theory, group theory, quantum complexity, lattice cryptography, extremal combinatorics, and more. Among them: establishing the existence of non-sofic groups and exponential improvements to bounds on high dimensional sphere packing.

GPT-Live can listen while it speaks. To make that feel natural at ChatGPT scale, we rebuilt the voice stack from client to model. This new architecture keeps audio flowing continuously, so deeper reasoning and tool use don't interrupt the conversation.

Audio moves through a dedicated fast path, while deeper reasoning and tool use happen asynchronously. We also reduced voice-session startup from six network round trips to one.

Serious cyber vulnerability disclosures keep climbing. In July, 21 major tech organizations published ~2,500 high- and critical-severity CVEs — about 5× the monthly record before Anthropic revealed Claude Mythos Preview could autonomously find software vulnerabilities.

Bloomberg TV

@BloombergTV

. @huggingface CEO @ClementDelangue says the recent OpenAI-linked cyberattack highlights the growing risks posed by rapidly-evolving AI models. Speaking to @edludlow , Delangue says that open-source AI models should be a part of the solution. Watch the full interview here https:// bloom.bg/4wE97K0

Cursor can now read, write, and act across your Google Workspace. New plugins give agents direct access to Gmail, Google Drive, Calendar, Docs, and Sheets.

Install them from the Customize page in Cursor. Learn more:

H3 just found a much bigger canvas Now live on @Picsart bringing open weights, native 2K, and omni-reference creation to millions of creators around the world. #MiniMaxH3 #OpenWeights

This is how much you can do with a single RTX 5090 NOW with MiniMax H3. WE HAVE CROSSED A LINE.

Finally a good paper testing whether agent memory needs an LLM at all. Production memory stacks spend extra model calls on summarizing interactions, writing records, and reranking retrievals. Every one of those calls costs tokens and latency, and the summaries quietly discard the evidence you later need. Zero-Mem removes generation from every memory step. Only the final question-answering reader ever invokes an LLM. It keeps original interaction traces as the record and indexes them two ways. An entity-context graph exposes connections across sessions. A temporal hierarchy preserves conversational locality and session state. For each query it weighs both views, retrieves from both, and follows their structure to recover supporting relations or surrounding context. Deterministic calibration then discards conflicting evidence before the reader answers. At matched reader and context budget, memory-operation time cost drops 57.6% against the fastest compared baseline, with competitive accuracy on long-memory and long-context QA. Paper: https:// arxiv.org/abs/2607.29377 Track more trending AI papers in our academy: https:// academy.dair.ai

This is a wild result. Locus, the automated research system from @intology , post-trained Qwen3 base models that beat the official human-tuned Qwen3 1.7B Instruct release. SoTA on PostTrainBench!