The universe is cooling.
Civilization is heating up.
One day, seen from outside time, and the stories that just moved civilization forward.
Get it as a 4K wallpaper →
34.2
+117.916.4
16.2
Google Gemini@GeminiAppPicks69You can now take an interactive web app you built in Gemini App straight into Google Sites - no code to copy, no app to rebuild. :) Export it to a private site draft, customize it, then publish and share with your team, customers or students. 🧵 1/3
Exporting an existing Gemini web app into a private Google Sites draft removes the copying and rebuilding steps between app creation and site publication.
Google DeepMind, Google Research, WRI and IIASA released Forest Typology (ForTy) 2020, a global 10-meter map classifying land into six forest and land-use categories, alongside an early time series covering 2017 to 2025. Validation across more than 9,000 sites found 90% accuracy for forest types and 95% for distinguishing forest from non-forest land. The datasets are freely available on Google Earth Engine under an open license and show changes including natural forests becoming plantations or farms.
Distinguishing forest types reveals conversions that tree cover maps can miss when cleared natural forests become green plantations, providing a more specific basis for assessing deforestation and degradation.
IBM introduces Granite Timeseries Ensemble, a permissively licensed recipe combining four pretrained Granite models to produce median and quantile forecasts without retraining or fitted ensemble weights. It offers default equal-influence probability pooling and an adaptive IQR alternative. Across 97 GIFT-Eval configurations, the default achieves aggregate MASE of 0.683 versus 0.685 for the strongest individual member, while PatchTST-FM-r2 retains the lowest aggregate CRPS/MWQL. Loading four checkpoints increases inference time and memory use.
The benchmark comparison clarifies the ensemble’s tradeoff between slightly better aggregate point forecasts and increased inference costs, while showing why probabilistic accuracy can favor an individual member.
AI-assisted research by Calif through OpenAI’s Patch the Planet initiative and separately by Google’s Wiz uncovered critical PHP vulnerabilities dating back to version 5, released in 2004. The PHP Foundation issued patches, and the NSA urged affected sites to patch immediately or go offline until patched; the recommended current version is PHP 8.5.12. OpenAI reported no known exploitation in the wild, while Wiz warned that LLMs could help attackers adapt exploits across PHP versions and configurations.
The report highlights how AI can help uncover longstanding software vulnerabilities while also lowering exploitation barriers, making coordinated disclosure and timely patching consequential for widely used infrastructure.
宝玉@doteyPicks79An urgent notice from the Railway team https://blog.railway.com/p/security-disclosure-wednesday-october-21st
The deployment pause complicates credential rotation because replacement secrets require redeployment to take effect, creating a potential service interruption if developers revoke existing credentials before deployments resume.
宝玉@doteyPicks81@dotey 谁能告诉我这和 Dots、Grok Bot、Muse 到底有啥区别？硅谷四大厂排队给我塞功能一模一样的赛博田螺姑娘，我那点少得可怜的待办事项已经不够它们分了......
The comparison makes intended users and access conditions the main selection criteria, helping distinguish four assistants described as sharing cloud computers, proactive updates, and approval checks.
Recursive released Flint, an open-weight MoE language model for local agentic coding, developed in four months with its Eureka Machine AI research system. Flint has 40 billion total parameters and 2.5 billion active per token, with Apache 2.0 weights available on Hugging Face. Recursive says the system contributed across architecture, training, quantization, and inference tuning; its benchmark claims have not been independently verified, and Flint does not yet demonstrate recursive self-improvement.
Flint offers a concrete example of automated model research spanning architecture, quantization, and inference tuning, while the article distinguishes those contributions from evidence that the research system itself improves.
Anthropic introduced a research preview of Claude Duet, a standalone proactive agent built on Claude Opus 5.5 that works on tasks while users are away. Available through a waitlist for US Pro and Max subscribers on iOS, Android, and the web, it checks connected apps every few hours and shares usage limits with other Claude products. Anthropic says Duet requires approval before sending messages, making payments, signing documents, or deleting anything, and acknowledges it can misread threads or miss updates.
Duet’s approval requirements and periodic account checks illustrate how a proactive agent’s autonomy is bounded by user control and the subscription limits shared with other Claude products.
AWS provides a guide to training Qwen3 with torchtitan on a single trn2.48xlarge using Transformers v5, torchtitan’s native Qwen3 implementation, and NKI mixture-of-experts kernels. The configurations share a torchrun launch pattern and configuration structure, with examples for loading and saving Hugging Face checkpoints. The guide also demonstrates custom kernel registration and autograd integration, explicitly presenting that example as an integration demonstration rather than a performance improvement.
The walkthrough provides a reusable pattern for integrating custom training kernels through operator registration, autograd, and configuration overrides, with numerical comparisons to check correctness before training.
OpenAI’s GPT-5.6 Sol found a critical PHP vulnerability potentially affecting millions of websites through the Patch-the-Planet initiative. Wiz independently found the same flaws using Wiz Atlas alongside Google’s Gemini 4 Argon and GPT-5.6 Sol, and reported that LLMs can adapt exploits across PHP versions, configurations, and memory layouts. Remediation was coordinated over two months across five phases involving PHP developers, Wiz, Cloudflare, CISA, and the White House.
The discovery illustrates how AI can support coordinated vulnerability remediation while also reducing the effort attackers need to adapt exploits across different software configurations.
Walmart announced Agent Gateway, an MCP server that lets approved AI agents search its online inventory and make purchases for shoppers. Agent Profiles, rolling out next month, will separate agent purchases and let customers require transaction approvals or set spending limits. Support requires a Walmart account but no Walmart+ membership, and Walmart says its returns process also applies when an agent acts unexpectedly.
Walmart’s purchase approvals, spending limits, and returns terms offer a concrete example of how retailers can give shoppers control over transactions made by AI agents.
vLLM@vllm_projectPicks65Native PyTorch on TPUs is here—and TorchTPU is officially open source. Train and serve models using the tools you already know: @vllm_project, @sgl_project, @_llm_d_, TorchTitan, @huggingface, and @raydistributed. Learn more from #PyTorchConf2026 → https://t.co/mISCD65vA4 https://t.co/z1Uwb2T3l6
Support for prefix caching, speculative decoding, and disaggregated serving gives readers concrete features to consider when assessing vLLM’s native PyTorch serving on TPUs.
NVIDIA introduced Kumo Relational, a pre-trained foundation model that predicts directly from relational databases using up to 10,000 historical examples as context, without task-specific training or feature engineering. NVIDIA reports that it outperforms supervised approaches across 41 tasks on 15 databases and surpasses AutoGluon ensembles by 8% in MRR on SAP SALT. It is available through the open-source structured-data-models library or as an NVIDIA NIM, with optional fine-tuning.
The benchmark comparisons offer a concrete basis for assessing when relational in-context learning can replace engineered prediction pipelines, with task size and relationship structure defining practical limits.
vLLM@vllm_projectPicks66Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine. • Best-in-class fast local coding • Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark • Built from https://t.co/ufg2xScSGw
The reported throughput at 100K context provides a concrete reference for evaluating Flint deployment on a single GB200 without speculative decoding.
ollama@ollamaPicks69Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine. • Best-in-class fast local coding • Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark • Built from https://t.co/ufg2xScSGw
The reported throughput figures include hardware and short context conditions, providing a concrete reference for readers considering Flint for local agentic coding workflows.
Dharma AI previews VocabX, a cost-weighted vocabulary adaptation method that cut measured inference cost by 77.0% in its Portuguese essay grader built on Qwen3-4B. With the same base model, data and training recipe, average responses shrank from 350 to 101 tokens and the bench score rose from 0.703 to 0.810. The results come from a paper still under review and may change before publication.
The grader experiment illustrates why vocabulary compression should preserve decision-bearing tokens, since separating grades from surrounding JSON improved both measured cost savings and assessment quality.
Google Research reports a prospective study in JAMA Network Open evaluating AI gestational age assessment from blind sweep ultrasounds performed by novice operators in Chicago and Nairobi. Among 385 participants between 16 and 36 weeks of pregnancy, the model achieved a mean absolute error of 4.18 days versus 4.48 days for expert sonographers, meeting the study’s non-inferiority criterion. Fine-tuning used data from 120 Chicago participants, with no Nairobi data used for local optimization.
Testing across two clinical settings and unfamiliar ultrasound hardware provides evidence about generalizability beyond the training cohort, a practical consideration for expanding prenatal screening access.
OpenAI@OpenAIPicks68OpenAI models helped researchers—operating as part of our Patch the Planet initiative—discover critical security flaws in PHP, an open-source programming language used to run an estimated 70% of websites. Over the past two months, we led a coalition of maintainers, researchers,
The skill offers a reusable way to guide website owners through checking whether their sites are affected by the reported PHP security flaws.
Microsoft Research’s Xinzhi Zhang and collaborator Zhao Song report proving the 31-year-old KLS conjecture, establishing a universal bound independent of dimension for log-concave distributions. Researchers directed AI agents through more than 100 candidate approaches, then developed recursive refinements that completed the proof. The account describes independent proofs and a subsequent Lean 4 formalization combining their key criterion with additional research.
The account offers a transferable research method through iterative human pruning of AI search trajectories, contrasted with unguided attempts that fell short using roughly the same total token budget.
IBM introduced granite-speech-5.0-2b and updated granite-speech-5.0-470m-turboctc and granite-speech-4.1-2b-plus, all under Apache 2.0 with native Transformers support. The new 2B model improves transcription and translation with keyword lists up to 2,000 terms; Turbo CTC adds multilingual transcription and reaches 19,870 RTFx, about 1.6x its launch speed at the same accuracy level. The updated Plus model improves speaker attribution and timestamps, with long-form WER down 24%.
The comparison maps accuracy, throughput, and speaker attribution improvements to distinct recording workflows, offering a practical basis for choosing among the three models using their reported capabilities and evaluation results.
