AI News

August 20, 2026 · 10:07 AM UTC

  1. Openai web traffic up 12% — now ranked #95openai.com

    openai.com moved from Tranco rank #108 to #95 over the past week (up 13 positions, 12.0% change). Tranco ranks the top 1M domains by web traffic.

    Also: Bluesky
  2. deepseek-ai/DeepSeek-V4-Pro-0813 — language model trendingyoutube.com

    Sigmund Freud is probably the most famous psychoanalyst. Yet many of his theories have been critiqued or discredited. So is there anything we can still learn from the father of psychoanalysis? Subscribe to our channel! http://goo.gl/0bsAjO In today’s episode Sean speaks with professor Mark Solms...

    Also: Lobsters, YouTube, HuggingFace
  3. Quoting Jeremy Morrellsimonwillison.net

    My hypothesis is that there is a new opportunity for Extensible Software on the web . LLMs radically lower the cost of authoring extensions, and modern sandbox primitives lower the deployment cost and provide good security boundaries. We can build our app as a solid, accountable core, and allow...

    Also: Hacker News
  4. Google packs Search and Gemini with new AI study toolstomsguide.com

    Google announces new study tools, including a student hub, notebooks, and interactive 3D visualizations in Gemini, plus student offers for Google AI plans — Head back to school equipped with AI tools for studying, researching and more — Once again, Google is giving eligible college students …

    Also: TechCrunch AI
  5. Meta AI is getting a Mac apptheverge.com

    Meta launches a Mac app for Meta AI and says Meta AI can now work directly with Instagram and Facebook accounts, Meta ad campaigns, and Google Workspace — You can share your window with the new Meta AI app, as well as connect it to Google Workspace. … Meta is launching a new Mac app dedicated to...

  6. London-based Callosum, whose software matches AI tasks with different models and chips, raised a $100M seed from Atomico, the UK Sovereign AI Fund, and othersbloomberg.com

    London-based Callosum, whose software matches AI tasks with different models and chips, raised a $100M seed from Atomico, the UK Sovereign AI Fund, and others — Callosum, a startup that makes software to match specific artificial intelligence tasks with different models and chips …

    Also: Techmeme
  7. OpenAI 'will be a public company in 2027' or sooner, CFO Friar tells employeescnbc.com

    Sources: OpenAI CFO Sarah Friar told employees at an all-hands that OpenAI “will be a public company in 2027”, or sooner if “our business continues to inflect” — OpenAI CFO Sarah Friar told employees during an all-hands meeting on Wednesday that the artificial intelligence lab …

  8. Flock Has a Powerful New AI Tool for Police. We Got Its Codewired.com

    Flock’s surveillance cameras have already sparked outrage. WIRED reconstructed its next-generation AI system, already in use by some police, to confirm it goes much further than tracking license plates.

  9. Turbovec – Google's TurboQuant for vector search in Rustgithub.com
  10. Devices with GrapheneOS support should be available in 2027grapheneos.social
  11. Personal milestone: 1000 true fans of my newsletter Interconnects! Hitting a very long term goal feels great. I’m happ...bsky.app

    Personal milestone: 1000 true fans of my newsletter Interconnects! Hitting a very long term goal feels great. I’m happy to get to be an independent voice in AI. Cultivating a paid base helps me commit to that longer term, and scale Interconnects’ impact.

  12. Qwen/Qwen3.8-27B-FP8 — vision-language modelhuggingface.co

    28B parameter vision-language model. 1,517,643 downloads. 615 likes on HuggingFace

  13. Lightricks/LTX-2.5 — video generation modelhuggingface.co

    Video generation model. 611,825 downloads. 1,357 likes on HuggingFace

  14. VentureBeat names Rob Strechay as its first Lead Analyst, expanding its enterprise AI research pushventurebeat.com

    Rob Strechay, until recently managing director and principal analyst at theCUBE Research, has joined VentureBeat as our first Lead Analyst and a founding analyst of VentureBeat Research. His arrival is the next step in a deliberate move at VentureBeat toward deeper specialization: analysis built

  15. Auditing Preference Biases and Fine-Tuning Language Models with Direct Preference Optimization on Anthropic HH-RLHF Using TRL and LoRAmarktechpost.com

    This tutorial provides an end-to-end workflow for fine-tuning language models using Direct Preference Optimization (DPO). We demonstrate how to audit the Anthropic HH-RLHF dataset for structural and length-based biases, implement a robust training pipeline using TRL and LoRA, and evaluate model...

  16. Google gives Gemini Live its own Deep Research upgradedigitaltrends.com

    Gemini Live can now launch Deep Research reports by voice, work on them in the background, and let you talk through the findings when they are ready.

  17. Entropy-Constrained Adaptive Stochastic Quantizationarxiv.org

    arXiv:2608.18147v1 Announce Type: new Abstract: Adaptive stochastic quantization (ASQ) is a recently introduced quantization approach that optimizes the Mean Squared Error (MSE) for a given input while preserving unbiasedness. It is designed to alleviate the communication and memory bottlenecks of...

  18. The Open-Sourcing of DeepSeek Harness Opens the Door to Modular, Unbundled AI Agent Infrastructureinfoq.com

    DeepSeek has released a developer preview of DeepSeek Harness (dsh), an open-source execution runtime for building autonomous AI agents. The software features a micro-kernel architecture with modular plugins for various functional units. The release includes an append-only event logging system for...

  19. Meet SAM (Sovereign Agent Mesh): A Zero-Config, Zero-Trust P2P Network for AI Agentsmarktechpost.com

    Google has open-sourced SAM (Sovereign Agent Mesh) under Apache-2.0 — and it has nothing to do with Segment Anything. SAM is a zero-config, zero-trust P2P overlay that lets autonomous agents discover and call each other's MCP tools across cloud, on-prem, laptop and edge environments, without...

  20. MoE-ViE: Mixture of Experts Vision Encoder for Efficient Image and Video Understandingarxiv.org

    Vision encoders are a critical component of vision-language models, and scaling their capacity effectively improves performance. However, dense scaling increases compute cost and inference latency. Mixture-of-Experts (MoE) architectures offer a compelling alternative, having enabled efficient...

  21. Learning the Center and Radius of Wasserstein Ambiguity Sets for Data-Driven Decision Makingarxiv.org

    arXiv:2608.18123v1 Announce Type: new Abstract: Wasserstein distributionally robust optimization (DRO) is commonly built around the empirical distribution, with the ambiguity radius selected from a concentration bound. Although this construction provides useful statistical guarantees, it can be...

  22. smolmachines / smolvm as a sandbox for untrusted Python & JavaScriptsimonwillison.net

    Research: smolmachines / smolvm as a sandbox for untrusted Python & JavaScript I tasked Claude Fable 5 running in Claude Code for web with the following research task: Put https://smolmachines.com through its paces as a fast secure sandbox. Explore what it would take to use this to run untrusted...

  23. Why Fast LLM Ranking Is Really Two Different Problemshackernoon.com

    Optimizing LLM ranking requires recognizing that inference consists of two distinct jobs: compute-intensive "prefill" and memory-bound "decode". By implementing strategies like continuous batching and chunked prefill, you can prevent large input tasks from causing latency bottlenecks and ensure...

  24. Uh oh, sounds like another agent escaped its testing environment!bsky.app

    Uh oh, sounds like another agent escaped its testing environment!

  25. Finetuning Strategies for Querying Sounds by Vocal Imitationarxiv.org

    This technical report describes our winning submission to the AES AIMLA 2025 Challenge on querying sound effects by vocal imitation. We investigate two complementary fine-tuning strategies: contrastive learning with a frozen, pretrained CED encoder, and joint contrastive-triplet learning with...

  26. NE-BERT: A Multilingual Language Model for Nine Northeast Indian Languagesarxiv.org

    arXiv:2608.18094v1 Announce Type: new Abstract: Large pretrained language models have demonstrated remarkable capabilities across diverse languages, yet critically underrepresented low-resource languages remain marginalized. We present NE-BERT, a domain-specific multilingual encoder model trained...

  27. Universality of Gradient Descent Neural Network Trainingarxiv.org
  28. Science Done on a Machine by a Machine: AI Agents in Computational Chemistryarxiv.org

    arXiv:2608.18508v1 Announce Type: cross Abstract: We are witnessing an explosion of agentic systems for computational chemistry simulations: from half a dozen in 2024 to a dozen in 2025, and the current number approaches fifty, surveyed in this Perspective as of 8 August 2026. The capabilities of...

  29. GPT-4o vs Claude vs Mistral: Choosing Models by Task Benchmarksdev.to

    Why LLM Benchmarks Need Task-Level Context Comparing GPT-4o, Claude, and Mistral with a single leaderboard score is convenient—but technically incomplete. Large language model performance depends on the workload, prompt structure, context length, output constraints, and evaluation method. A model...

  30. How Unitree's Go series, which helped the company dominate the quadruped robot market, drew on openly published US university research funded by the US military (Michael Martina/Reuters)techmeme.com

    Michael Martina / Reuters : <s

  31. SemComp-Bench: Benchmarking Semantic Task Completion in Video Generationarxiv.org

    We introduce Semantic Task Completion Video Generation, an outcome-oriented video generation task. Under this formulation, success requires both achievement of the intended outcome and semantic grounding. Semantic grounding characterizes the correspondence between the reference image and the...

  32. A revisit of remote Spectre attacks on Cloudflare Workersblog.cloudflare.com

    In 2024 and 2025, we reassessed remote Spectre attacks on our Workers infrastructure. We share details about the new attack primitives like Spectre gadgets, remote timers, achieving co-location and how new defenses further harden Cloudflare Workers.

  33. Algolia joins the Vercel Marketplacevercel.com

    You can now provision and manage Algolia directly from the Vercel Marketplace. Algolia is a hosted search platform. You send it your content, it builds an index, and your app queries that index instead of your database. Results come back in milliseconds, with typo tolerance, synonyms, and facet...

  34. Grouping the Stochastic Machine: Precision, Not Capability, as the Frontier Metric for AI Systemsarxiv.org

    Frontier language models are compared, marketed, and benchmarked on capability -- what their best or average output can achieve. I argue this measures the wrong axis. The models have saturated accuracy: their mean output lands on the target. What now separates one system from another in practice is...

  35. Asynchronous patterns for calling Amazon Bedrock AgentCore agents in serverless pipelinesaws.amazon.com

    In this post, you learn three serverless patterns (task-token callback, direct service integration, and durable functions) for invoking Amazon Bedrock AgentCore agents asynchronously from AWS Step Functions pipelines, eliminating idle compute costs while your AI agent processes requests.

  36. Unitree Shares Surge in Stock Market Debutaibusiness.com

    The company is the world’s biggest humanoid robot maker by sales.

  37. Simulation-Optimization of Systems of Optimizers: Exploiting the Inner Optimization's Geometryarxiv.org

    arXiv:2608.18129v1 Announce Type: new Abstract: We study simulation-optimization of systems of optimizers (SOSO): agent-based simulations in which every agent solves a structured optimization - a linear program (LP), mixed-integer program, or dynamic program - at every decision epoch. Such systems...

  38. Coinbase's Base Is Betting Big on AI Agents With $1 Million Startup Acceleratordecrypt.co

    Base is narrowing its accelerator to 10 startups as it looks for teams building AI agents, payments, trading, and financing products.

  39. Geometric Iterative Retrieval for Neural Audio Codec Resynthesisarxiv.org

    Neural audio codecs based on Residual Vector Quantization (RVQ) have become the dominant discrete representation for token-based general audio generation, yet resynthesizing high-quality audio from coarse codec tokens remains an open problem and bounds the fidelity of every system that generates...

  40. Designing Agent Memory: Five Gates a Bigger Context Window Cannot Replacehackernoon.com

    A million-token window enlarges the warehouse. Reliable agents still need the gates that decide what to store, retrieve, revise and forget. Read All