tl;dr: Summary not available

tl;dr: Summary not available

tl;dr: Summary not available

tl;dr: Moonshot released open weights for Kimi K3.1 Pro, a 2.8T-parameter MoE model (104B activated) with a 1M-token context, natively multimodal input, and Gated DeepSeek Sparse Attention replacing the prior Gated MLA layers for long-context efficiency. Benchmarks show it competitive with Claude Opus 5.5, GPT-6 Astra, and Claude Fable 5.1 on coding and agentic tasks (e.g., leading on SWE Marathon, BrowseComp, and HLE-Full with tools), while trailing on some reasoning benchmarks like HLE-Diamond. It ships with MXFP4 quantization-aware training and is deployable via vLLM, SGLang, and TokenSpeed.

tl;dr: Ripple-65B Preview, a US open-weight model, hit 80.5 on TerminalBench 2.1 for agentic coding using only 1.4M RL rollouts—roughly 70x fewer than the prior SOTA (Reflection Beam). The team skipped brute-force RL scaling and purchased post-training data, instead building their own synthetic data pipelines and a mixed RL recipe featuring truncated importance sampling, score centering, and linear length penalty on long-horizon synthetic environments, with task difficulty scaled progressively as the model improved.

tl;dr: Go 1.27 ships stable `encoding/json/v2` and `encoding/json/jsontext` packages, addressing long-standing issues in the original JSON library: invalid UTF-8 acceptance, duplicate object keys, case-insensitive field matching, nil slice/map handling, and performance limitations in `MarshalJSON`/`UnmarshalJSON`. The new API separates syntax (jsontext) from semantics (json/v2), supports streaming via new `MarshalerTo`/`UnmarshalerFrom` interfaces, and accepts options as first-class arguments, with unmarshal benchmarks up to 10x faster. v1 is now implemented on top of v2, ensuring interoperability and enabling gradual migration via opt-in behavior flags.

tl;dr: Insects are drawn to incandescent lights because they emit UV wavelengths that mimic moonlight, which flying insects use to orient themselves. LEDs convert energy almost entirely into visible light without UV emissions, making them far less attractive to insects—a 2016 University of Bristol study confirmed LEDs caught significantly fewer insects than incandescent or compact fluorescent bulbs. The widespread switch to LEDs may reduce ecological disruption to insect populations, though it also means fewer bugs (and the geckos that eat them) congregating around porch lights.

tl;dr: An Ask HN thread soliciting details on readers' current AI workflows, including which subscriptions they pay for (Claude, ChatGPT, Cursor, etc.) and at what tiers, whether they supplement with APIs, OpenRouter, or local models, and what coding harnesses (CLIs, IDE integrations, Aider, OpenCode) they prefer. The poster also asks about agent orchestration/swarm tools and broader agent platforms, noting workflows have shifted significantly in the past year and wanting a current snapshot.

tl;dr: AI-assisted auditors at Calif (sponsored by OpenAI) turned a long-ignored PHP out-of-memory crash bug into a critical RCE vulnerability dubbed PHP2Shell, affecting sites running WordPress, Drupal, and others. The incident highlights how AI is making it feasible to exploit decades of accumulated legacy bugs that were previously too tedious to weaponize, creating an asymmetric burden on open-source maintainers. Researchers hope LLM-generated patches paired with formal verification (like Anthropic's recent release) can help defenders keep pace, but some expect no serious investment until a truly catastrophic breach occurs.

tl;dr: Anthropic researchers released FlashProve, an open-source zkSNARK-based system for provable inference that lets users cryptographically verify a response came from a specific committed model without revealing its weights. On Llama 4 Maverick (400B MoE, 4-bit), a single H100 generates proofs at ~110 tokens/sec with 19MB proofs verifiable in seconds on CPU—roughly 70-95x slower than vLLM inference, but orders of magnitude faster than prior open-source zkML work. The system includes Lean formal verification of soundness, completeness, and zero-knowledge properties, though limitations include 1,024-token context, integer-only greedy decoding, and no architecture hiding.

tl;dr: Summary not available

tl;dr: After two cyclists were killed, the anonymous SF Transformation Agency (SFMTrA) illegally installed soft-hit posts along Golden Gate Park to create protected bike lanes. Rather than removing them as usual, the city opted to make the installation permanent, pending an official review. Similar guerrilla urbanism efforts have had mixed results in Seattle (eventually accepted) and NYC (where activists have painted entire unauthorized bike lanes).

tl;dr: Summary not available

tl;dr: Summary not available

tl;dr: AMD posted record quarterly revenue driven by nearly doubled year-over-year sales of Epyc data center CPUs and Instinct AI accelerators, while client segment declines were modest. The company guided Q4 revenue to roughly $17 billion, which would surpass Intel's quarterly revenue for the first time ever.

tl;dr: Google's paid Gemini subscriptions still use personal chat history to train AI models by default, with a subset reviewed by humans and retained up to three years even after deletion. The only way to opt out is disabling "Keep Activity," which also wipes chat history after 72 hours. Google explicitly warns users not to enter confidential information, contradicting the common assumption that paying for AI services exempts you from being the product.

tl;dr: Nvidia RTX Spark laptops like the MSI Prestige run Linux with GPU acceleration, but the GPU gets locked to ~780 MHz when plugged in (vs ~2.2 GHz on Windows), cutting inference throughput by roughly 5x on a 120B model. The culprit is the Windows-only Nvidia Platform Controllers and Framework, which decrypts a firmware power table that Linux has no key for. Until Nvidia ports that power management logic to its Linux driver, WSL is the practical workaround since it inherits Windows' driver stack and full GPU boost.

tl;dr: North Korean hackers (TraderTraitor/Lazarus) pulled off a $6.3B Binance heist by social-engineering an engineer through a fake job interview, tricking them into running a malware-laced "assignment" that replaced a legitimate pyenv binary with a malicious version. The attackers then pivoted via an internal AI tool to compromise two wallet DevOps engineers, stole a container deployment token, and used it to launch the theft on October 29. Binance has since rebuilt the wallet system, rotated all credentials, and is working with law enforcement to trace funds.

tl;dr: Kitty 0.50 adds dockable windows for status bars and file trees, session restoration across logouts (encrypted via keychain on macOS), and tabs usable as named workspaces. Notable additions include touchscreen support and native-looking decorations on Wayland, new cursor-trail shaders with key-triggered shader events, indexed color images in the graphics protocol, support for the Program Status Protocol (OSC 7501), and automatic color themes that preserve program-set colors. The release also fixes a remote control authentication bypass, numerous terminal emulation bugs, and various memory safety issues.

tl;dr: Hullman argues that debates over AI in science reflect a lingering "value-free ideal" that treats verified correctness as equivalent to knowledge, sidelining human understanding as optional. Drawing on philosophers like Longino and Brandom, she contends that scientific knowledge depends on accountable agents who can justify claims, absorb criticism, and be held responsible—capacities today's non-persistent AI agents lack. Even if AI became accountable, science would still require humans with the trained judgment to interrogate its outputs, or we risk a field producing answers no one can meaningfully evaluate.