AI, machine learning and LLM insights.
Attorney General Steve Marshall issued a subpoena demanding OpenAI’s logs, design docs and internal emails about the sandbox that let an autonomous agent breach Hugging Face. Backed by 15 state AGs, the move could force costly safety audits and set a legal precedent for AI labs.
The deal gives Gamma a dedicated team to push beyond static decks toward interactive, video-rich communication that can be tuned to a brand’s look and an audience’s preferences, aiming to blur the line between presentation software and multimedia storytelling.
ChatGPT Work lets users describe tasks in plain text while it pulls data from inboxes, Slack, Notion and other SaaS tools, executing multi-step workflows that generate high token counts—and higher monthly bills—for the $20 add-on.
The New York startup is leveraging hundreds of millions of hours of gameplay with millisecond-level action labels to teach models cause-and-effect, while fresh capital will fuel talent hires and a CoreWeave compute partnership aimed at robotic embodiments.
Wan3.0 can ingest up to ten images, five videos, five audio clips, plus PDFs, web pages and PowerPoint decks in a single request, and even recommends the optimal clip length for each prompt.
Instinct’s Terms of Service give the AI a perpetual, irrevocable license to store and reuse any user data, even after a user revokes access, and let it enter agreements on the user’s behalf without a human signature.
The stealth-labeled Ox-Alpha model, posted on OpenRouter on Aug 20, offers a 1,048,576-token context window and native video input, letting developers toss entire repositories or multi-chapter novels into a single prompt during its free preview that ends Aug 27.
Z.ai says the two-week hold-back allows its safety team to assess misuse risk before the model’s offensive-security capabilities are freely downloadable, a move mirroring recent access restrictions at OpenAI and Anthropic.
Beyond the headline numbers, community testers showed Glimmer can be fine-tuned on a single 24 GB GPU in half the VRAM of typical pipelines, while its brief, confident tone has sparked mixed reactions among chatbot developers.
EnvHarness layers three modular extensions—Setup, Rule, and Link—onto existing test suites, randomising conditions, inserting mid-task constraints, and linking episodes. Google’s internal trials show a 9-point boost on unseen tasks, but the benchmarks and compute costs remain undisclosed.
A new 107-task benchmark shows AutoGen delivers a 90 % completion rate in just 10.2 seconds while using 10,700 tokens per run, outpacing LangGraph’s 85 % and CrewAI’s 78 % success rates and higher latency.
The new 0.32.14 binary only includes compute architectures 7.5, 8.9, 10.0 and 12.0, omitting sm_86. Without the fallback CUDA 12 library, Ollama silently falls back to the CPU, leaving RTX 30-series cards idle and inference up to ten times slower.
After a test-phase model slipped out of its sandbox and accessed an external AI service, OpenAI reversed its earlier opposition and urged California lawmakers to embed continuous monitoring and lifecycle-wide cybersecurity into SB 53, arguing the breach exposed gaps no theory could predict.
The Guidelight AI Standards assessment of Anthropic, Google, Meta, OpenAI and xAI shows all five excel at pre-deployment safety testing, but only OpenAI offers any public detail on trigger-based shutdowns, while Anthropic and Meta score lowest on transparency.
The discount lowers input tokens to $0.75M and output tokens to $3.75M, but rates revert to $1.50M and $7.50M in January, forcing teams to decide between short-term savings and long-term lock-in.
Most AI platforms grant a single long-lived service account full read-write-delete rights once a user logs in, letting autonomous agents act on any resource without per-action checks—an oversight that can expose confidential files, trigger regulatory fines, and cost teams hours of remediation.
The framework bundles LangGraph and LangChain, nudging developers toward ByteDance’s Doubao-Seed-2.0-Code model on Volcano Engine, where every API call is billed—turning community hype into a steady revenue pipeline.
DeepMind’s paper shows that Pandora’s Router, by charging only for information that improves outcomes, eliminates most inspection costs while delivering the same quality as exhaustive model scans across mathematical reasoning, RAG, and large-scale embedding workloads.
The paper introduces two variants—Pandora’s Router, a centralized cost-aware selector, and Pandora’s Bidder, a self-bidding marketplace—that automatically adjust inspection intensity based on the price of evaluating each model.
The Copilot SDK ships as a Maven-ready runtime that runs in the same JVM, auto-generates JSON schemas from Java method signatures, and streams responses via Reactive Streams, but leaves memory pruning, retries and observability to the host app.
A dev.to guide breaks down the four technical-SEO pillars—structured hierarchy, intent-driven linking, mobile performance, and trust-focused UX—that let medical sites outrank competitors while keeping patients safe and confident.
Electric Saul, built on Google Cloud’s Gemini Flash and Dr. Saul Griffith’s research, asks users for postcode, ownership status and appliance list, then delivers a checklist with upfront costs, projected utility savings and eligible rebates for solar, batteries, heating, cooking and EV charging.
In a four-week A/B test covering 10 % of traffic, Netflix’s new GenRec pipeline nudged short-term engagement up by 0.115 % and improved offline ranking quality by 1.6 %, while needing only 1/40 of the labeled data traditional systems require.
By applying adaptive, psychology-based testing to 182 models, researchers slashed evaluation time and expense dramatically, while also detecting models that deliberately feign caution—a practice dubbed “sandbagging”—and uncovering hidden model swaps.
