📂 Category
Research & Benchmarks News Archive - Page 7 of 10
932 articles in this category • Page 7 of 10
- 601. DeepMind spinoff’s AI‑designed drugs enter human trials after AlphaFold 3
- 602. COALA paper defines agent memory types: procedural rules and semantic facts
- 603. Google DeepMind's Decoupled DiLoCo hits 88% goodput despite hardware failures
- 604. Agent observability powers production evaluation through trace analysis
- 605. Xiaomi launches MiMo‑V2.5‑Pro and V2.5, matching benchmarks at lower token cost
- 606. Designing Production-Grade CAMEL Multi-Agent Systems: Start with Docs and GitHub
- 607. Multi-agent AI systems incur higher token costs than single agents in practice
- 608. Reinforcement learning trains AI like OpenAI's o1 to admit uncertainty
- 609. LangSmith adds reusable LLM-as-judge and rule-based code evaluator templates
- 610. AI made up over a third of new sites by 2025; Pope warning flagged as AI
- 611. Sergey Brin pushes DeepMind to match Claude, unveils agent skills catalog
- 612. Fortnite adds AI‑powered NPCs for unscripted player conversations
- 613. TabPFN hits 98.8% accuracy in 0.47 s, beating Random Forest and CatBoost
- 614. NVIDIA PhysicsNeMo Tutorial Maps k(x,y) to u(x,y) for Darcy Flow
- 615. OpenAI unveils GPT‑Rosalind, AI model to speed drug discovery and genomics
- 616. Standard LLM guidelines focus on training costs, overlook inference budget
- 617. GPT‑Rosalind life‑sciences plugin for Codex launches on GitHub
- 618. OpenAI launches GPT-Rosalind, hits top score on BixBench benchmark
- 619. Frontier AI models fail one in three production runs, audits grow harder
- 620. Meta researchers unveil hyperagents for self‑improving AI in non‑coding tasks
- 621. Claude outperforms humans on alignment task, but results disappear in production
- 622. Google DeepMind unveils Gemini Robotics‑ER 1.6, beats prior model in tool count
- 623. UK tests Mythos AI, noting its ability to chain multistep attacks
- 624. AI Forum Launches Professional Certificate and USD 120M Fund for AI Fluency
- 625. Databricks finds multi-step agents beat single-turn RAG by 21% to 38% on STaRK
- 626. Stanford AI Index 2026: 53% adopt generative AI in 3 years, education lags
- 627. NVIDIA, UMD release AF-Next audio model, beats Phi-4-mm by 12 points on Arabic
- 628. Developers Claim Measured Drop in Claude's Performance, Sparking Nerf Debate
- 629. Seven AI agents in finance lift cash flow >3% monthly, boost productivity 50%
- 630. Meta AI and KAUST Propose Neural Computers Merging Compute, Memory, I/O
- 631. Prediction drift can mask security model decay despite stable accuracy
- 632. Researchers say OpenAI's Sora and Google's Veo aren't true world models
- 633. TriAttention KV Cache Compression Matches Full Attention, 2.5× Faster
- 634. Knowledge Distillation Keeps Student Model Capacity to Match Ensemble Boundaries
- 635. Google AI's PaperOrchestra boosts manuscript success, 79‑81% win rate
- 636. OSGym runs 1,000+ OS replicas at USD 0.23/day with decentralized state management
- 637. Stanford study finds AI agent handoffs lose information, affecting compute cost
- 638. Meta Superintelligence Labs launches Muse Spark, its first multimodal AI model
- 639. Better Harness updates add usage examples, chaining guide, and tool clarifications
- 640. Study finds ‘bot’ term used 16,232 times in 2.8M Telegram messages
- 641. Google AI Overviews answers 91% of test questions correctly after Gemini 3 update
- 642. MaxToki AI boosts context to 16,384 tokens with RoPE scaling
- 643. Meta staff inflate AI token counts on internal leaderboard, wasting resources
- 644. MassMutual, Mass General Brigham turn AI pilot sprawl into production
- 645. OpenAI safety staff exit as Altman dismisses Pentagon contract concerns
- 646. OpenAI urges firms to fund pensions, health, childcare as AI cuts costs
- 647. Study shows sycophantic AI chatbots can outwit ideal rational users
- 648. Americans use AI more than ever but trust it less, Quinnipiac poll shows
- 649. Study maps developer frustration with AI slop as tragedy of the commons
- 650. Google study: AI benchmarks ignore human disagreement; under 10 raters fail
- 651. Alibaba's Qwen team adds method that lengthens AI answers, prompting reasoning
- 652. Open models cross threshold; frontier models show per‑category correctness
- 653. Batch Mode VC-6 and NVIDIA Nsight Speed Up Vision AI Pipelines
- 654. CaP-Agent0 Beats Human Code on 4 of 7 Robot Tasks Using Low‑Level Blocks
- 655. Nvidia breaks MLPerf records with 288 GPUs as AMD, Intel pursue other goals
- 656. NVIDIA's 288-GPU Blackwell Ultra Sets New MLPerf Inference Throughput Record
- 657. DeepMind study finds six traps that let a few poisoned docs hijack AI agents
- 658. AI productivity gap: top agent beats baseline in 1 of 15 runs, 26.5% subtasks
- 659. Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
- 660. AI sycophancy cuts apologies, raises double‑downs; lifts moral trust
- 661. AI models fabricate image descriptions; benchmarks miss the shortcuts
- 662. Cohere's open-weight ASR model reaches 5.4% WER, ready for production use
- 663. Free API that evolved from slow web search to top AI tool, beyond scraping
- 664. Meta unveils open-source brain AI, adds Scrunch site audit and Suno v5.5
- 665. AI assurance experts meet to build infrastructure for safe, high‑quality systems
- 666. Study finds overly flattering AI advice can impair users' judgment
- 667. xMemory reduces token usage and context bloat versus MemGPT's raw logging
- 668. Mozilla dev launches cq, a Stack Overflow‑style hub for agents
- 669. Liquid‑cooled AI systems make storage an active cooling and GPU partner
- 670. 10 X Accounts for LLM Updates, Including the ‘Largest AI Newsletter’
- 671. Teens await sentencing for AI‑generated nude images as parents sue school
- 672. Developers say AI‑generated games feel unlike human‑made; audiences don't connect
- 673. Hachette withdraws Shy Girl horror novel amid AI usage concerns
- 674. Scale AI's Voice Showdown ranks Qwen ahead of top models, highlights failures
- 675. SynthID uses steganography to embed hidden watermarks in data
- 676. Google Search experiments with AI-generated headlines, may expand rollout
- 677. Growing cultural disconnect as companies race to deploy AI rapidly
- 678. Deep AI adopters reshape workflow, borrowing product‑manager tactics
- 679. NVIDIA DGX Spark expands node support to four, doubling memory capacity
- 680. Google's MusicFX DJ Enables Real-Time Controllable AI Music Generation
- 681. Paper identifies simple games that defeat AlphaGo and AlphaChess training
- 682. NVIDIA Cosmos Transfer Enables Scalable Synthetic Data for Physical AI
- 683. Trump Administration Signals Possible Additional Sanctions on Anthropic at Hearing
- 684. YouTube extends AI deepfake detection to politicians, journalists
- 685. Karpathy releases open-source Autoresearch, runs hundreds of AI tests nightly
- 686. AI spots trends but misses significance, keeping humans essential
- 687. Large CUDA Tiles Reduce Flash Attention TFLOPS by 18‑43% Across Sequences
- 688. KV cache compaction cuts LLM memory 50×, chunked processing long contexts
- 689. AI system flags probable matches, narrows anonymous accounts to shortlist
- 690. Seven tech giants sign Trump pledge to curb data‑center power cost spikes
- 691. Microsoft's Phi-4 Reasoning Vision 15B offers low‑latency, compact AI
- 692. LangSmith CLI adds three portable skills for coding agents in the repo
- 693. Secret meeting sees 94% approve even least‑popular AI resistance stance
- 694. AI data centers move to Arctic edge, boosting Nordic rural economies
- 695. Microsoft's OPCD cuts system prompts while preserving AI performance
- 696. Wall Street shows persistent AI anxiety, sparking frequent mini‑panics
- 697. Riley Walz, the ‘Jester of Silicon Valley,’ joins OpenAI’s OAI Labs team
- 698. AI enables scientists to integrate multiple cell measurements
- 699. Researchers argue building conscious AI could foster empathy, despite doubts
- 700. AI Researchers Resign, Bots Hire Humans, Anthropic Targeted, Evie Party