π΄ High Significance
Model Releases
π΄ π¬ Why are MoE models so belittled? β score 79
Sources: reddit/r/LocalLLaMA
E.g "Qwen 3.5 122B is just 10B active, so it's no where close to the dense 27B model" That is the main sentiment around here and it puzzles me. If a 122B is just worth 10B, then why does model providers bother creating an MoE model when they could've just released a dense 10B model? Heck the 10B d
Developer Tools
π΄ π¬ The U.S. tech industry is increasingly anxious about the rising power and competitive price of open-source AI models from China β and whether the Trump administration will respond with yet another executive order | Politico β score 96
Sources: reddit/r/LocalLLaMA
Politico: Wall Streetβs new obsession: Which CEOs have Trumpβs ear?: [https://www.politico.com/newsletters/politico-influence/2026/07/10/wall-streets-new-obsession-which-ceos-have-trumps-ear-00993324](https://www.politico.com/newsletters/politico-influence/2026/07/10/wall-streets-new-obsession-which
π΄ π¬ Public Library Find [D] β score 94
Sources: reddit/r/MachineLearning
Pleasantly surprised to find OβReilly books on ML at a public library
π΄ π Shubhamsaboo/awesome-llm-apps β 100+ AI Agent & RAG apps you can actually run β clone, customize, ship. β score 93
Sources: github_trending
100+ AI Agent & RAG apps you can actually run β clone, customize, ship.
π΄ π¬ I created a super harmful model ! :D (by tweaking it's J-Space!!!) β score 88
Sources: reddit/r/LocalLLaMA
Soooo! Since Anthropic share their Jacobian-Lens a few days ago I went on and made a tool based on it which adds the possibilitΓ© to export a model which will have the same behavior after tweaking it's J-Space. This means manually alter the behavior an
π΄ π FoundationAgents/OpenManus β No fortress, purely open ground. OpenManus is Coming. β score 86
Sources: github_trending
No fortress, purely open ground. OpenManus is Coming.
Omitted 3 additional developer tools items from the main section; see raw data and source-specific sections below.
Other Signals
π΄ 𧑠Reverse centaurs are the answer to the AI paradox (2025) β score 88
Sources: hackernews
π‘ Notable
Model Releases
π‘ βοΈ WHY IS CHATGPT FOR MAC SO GOOD? (5 MINUTE READ) β score 65
Sources: newsletter/tldr
π‘ βοΈ META IS QUIETLY LAUNCHING POCKET, AN APP FOR VIBE-CODING AND SCROLLING SMALL 'GIZMOS' (1 MINUTE READ) β score 65
Sources: newsletter/tldr
π‘ βοΈ EXPANDING MANAGED AGENTS IN GEMINI API: BACKGROUND TASKS, REMOTE MCP AND MORE (2 MINUTE READ) β score 65
Sources: newsletter/tldr
π‘ βοΈ USE CLAUDE COWORK ON WEB, DESKTOP, AND MOBILE (3 MINUTE READ) β score 65
Sources: newsletter/tldr
π‘ βοΈ GPT-5.6 SOL, ALONG WITH TERRA AND LUNA, WILL LAUNCH PUBLICLY THIS THURSDAY (1 MINUTE READ) β score 65
Sources: newsletter/tldr
Omitted 8 additional model releases items from the main section; see raw data and source-specific sections below.
Developer Tools
π‘ βοΈ HOW TO USE AI AGENTS BETTER THAN 99% OF PEOPLE (32 MINUTE READ) β score 65
Sources: newsletter/tldr
π‘ βοΈ AGENTIC AI ADOPTION IS ON FIRE AT UBER (2 MINUTE READ) β score 65
Sources: newsletter/tldr
π‘ βοΈ THE AGENT-ERA CAREER (7 MINUTE READ) β score 65
Sources: newsletter/tldr
π‘ βοΈ GITLOST: HOW WE TRICKED GITHUB'S AI AGENT INTO LEAKING PRIVATE REPOS (3 MINUTE READ) β score 65
Sources: newsletter/tldr
π‘ βοΈ ROGUE AGENT: HOW A SINGLE CODE BLOCK COULD HIJACK YOUR AI CONVERSATIONS IN GOOGLE'S DIALOGFLOW (6 MINUTE READ) β score 65
Sources: newsletter/tldr
Omitted 5 additional developer tools items from the main section; see raw data and source-specific sections below.
Infrastructure & Compute
π‘ βοΈ FACING US EXPORT CONTROLS, CHINA'S DEEPSEEK PLANS TO MAKE ITS OWN CHIPS (2 MINUTE READ) β score 65
Sources: newsletter/tldr
π‘ π¬ Performance comparison on full compute performance (Anima) and LLM prompt processing of 5090 (600,475 and 400W) vs 6000 PRO MaxQ shunt modded and water cooled (at 300, 400, 475 and 600W), and 6000 PRO WS/SE (600W). β score 46
Sources: reddit/r/LocalLLaMA
Hello guys, hoping you're doing fine! I'm continuing after this post some time ago, comparing stock MaxQ performance and such on Anima here. This time, I shunt modded the 6000 PRO MaxQ, to
Business & Funding
π‘ π¬ Withdraw from ACL ARR and resubmit to a workshop? [D] β score 56
Sources: reddit/r/MachineLearning
Hey guys, I received mediocre scores for my EMNLP paper during the May ACL ARR cycle: 2.5/3, 3/4, 2.5/4. The paper is in the Interpretability track. The reviewers had no larger issue with the methodology or the paper in general, but it seemed like they didn't fully get the so what of my paper. I've
Other Signals
π‘ βοΈ THE PEOPLE WHO WILL THRIVE IN THE AI AGE (22 MINUTE READ) β score 65
Sources: newsletter/tldr
π‘ βοΈ CHINA MAY RESTRICT ACCESS TO ITS MOST POWERFUL AI MODELS (4 MINUTE READ) β score 65
Sources: newsletter/tldr
π‘ βοΈ MICROSOFT REPLACES OPENAI, ANTHROPIC WITH OWN AI IN SOME APPS (2 MINUTE READ) β score 65
Sources: newsletter/tldr
π‘ βοΈ WILL SOMEONE FINALLY BLINK IN THE AI SPENDING WAR? (8 MINUTE READ) β score 65
Sources: newsletter/tldr
π‘ βοΈ THEORY OF CONSTRAINTS, AI, AND CODE REVIEW (10 MINUTE READ) β score 65
Sources: newsletter/tldr
Omitted 14 additional other signals items from the main section; see raw data and source-specific sections below.
π’ Incremental
Model Releases
π’ π¬ VultronRetriever family of models released on HuggingFace![R] β score 6
Sources: reddit/r/MachineLearning
Thrilled to announce the VultronRetriever family of models, which were revealed during Raise Summit Paris and demonstrated running Q&A and embedding documents on the iPhone, fully offline! π± Some highlights from the VultronRetriever model family: π₯ Each model ranks #1 in its respective class on
Developer Tools
π’ π¬ How does *ACL conferences acceptance work [D] β score 38
Sources: reddit/r/MachineLearning
Even after getting ARR reviews and a meta review, how is the acceptance decided at the *ACL venues, because I have seen meta review 3.5 getting to findings and 3 getting to main or even getting rejected. Then what is the purpose of the overall score and recommendation? What do the conferences see w
π’ π cognizant-ai-lab/neuro-san-studio β A playground for neuro-san β score 37
Sources: github_trending
A playground for neuro-san
π’ π¬ Can you build AI agents if you understand the concepts but can't code from scratch? β score 33
Sources: reddit/r/AIAgents
I've completed a beginner-to-intermediate Python course covering variables, functions, OOP, APIs, Pandas, Git, environments, and project structure. I also studied AI agents, LLM evaluations, prompt chaining, memory, tools, routing, and building simple AI agents. The problem is: I understand what
π’ π¬ Predicting human preference for generated image pairs using HPSv3 [P] β score 18
Sources: reddit/r/MachineLearning
Hey! I'm looking for ways to predict human preference for a project I'm building. (imagebench.ai) I've tryed HPSv3, https://github.com/MizzenAI/HPSv3 and made post about it here: [https://imagebench.ai/blog/does-the-score-match-your-eye](https://imagebench.ai
π’ π¬ Is deploying and scaling ai agents one of the most frustrating problem? β score 17
Sources: reddit/r/AIAgents
Everyone talks about hallucinations, state management but forgets this basics ,I wonder whether scaling ai agents for production is easy? Even though many platforms claim it is incredibly frustrating to setup complex things to just get things tested. Is it a real problem for all or just me experienc
Omitted 1 additional developer tools items from the main section; see raw data and source-specific sections below.
Infrastructure & Compute
π’ 𧑠Show HN: Reame β a CPU inference server that gets faster as it runs β score 38
Sources: hackernews
π’ π¬ [not mine] costs β score 12
Sources: reddit/r/LocalLLaMA
Saw this on a random social media post. I believe what we are seeing here is two sxm2 gpus, right? V100 or a100? Assuming two v100 with 32gb each, can you get nvlink working with these weird aftermarket base boards? Costs, experience, regrets? Not trying to build a similar case, but i am curious to
Other Signals
π’ π¬ CTX: How far can you reasonably go with Qwen 3.6 27B? β score 38
Sources: reddit/r/LocalLLaMA
How far can i stretch the context window with Qwen 3.6 27B (using Q8_0) before it gets too unreliable? I am at 100k right now and i am not quite statisfied. Other than not quantizing KV cache, is there anything else that can be done to make the model more stable over longer CTX?
π’ π¬ I feel like I'm not using my hardware efficiently β score 29
Sources: reddit/r/LocalLLaMA
Hi there, got a 7950x,128GB DDR5, RTX 4090 and RTX 3090TI. I'm currently running Qwen3.6 27B Q8 with 262k Context at Q8 with llama.cpp. It's not touching the DDR5 RAM at all but at the same time I couldn't get 122B A10B or the likes to run. Is my FOMO justified or isn't there anything better than th
π’ π¬ Is it possible to run Qwen 122B in 64GB ram + 24gb vram ? If so, how ? β score 21
Sources: reddit/r/LocalLLaMA
Which settings would suffice to work with it ?
π’ 𧑠Mesh LLM: distributed AI computing on iroh β score 12
Sources: hackernews
π’ π¬ Here's your recipes for models for the most popular hardware β score 4
Sources: reddit/r/LocalLLaMA
I occasionally comment and people ask how do I run something, how fast the model x on hardware y. There you go. Added a separate website page for recipes. Filter by hardware you have and see all the recipes. No recipe you created or you run? Submit it. Running some recipe from the list? Press I run
π Trending Repos
| Repo | Description | Stars Today | Language |
|---|---|---|---|
| Shubhamsaboo/awesome-llm-apps | 100+ AI Agent & RAG apps you can actually run β clone, customize, ship. | 519 | python |
| FoundationAgents/OpenManus | No fortress, purely open ground. OpenManus is Coming. | 217 | python |
| ATH-MaaS/Pixelle-Video | π AI ε ¨θͺε¨ηθ§ι’εΌζ | AI Fully Automated Short Video Engine | 157 | python |
| volcengine/OpenViking | Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills. | 32 | python |
| cognizant-ai-lab/neuro-san-studio | A playground for neuro-san | 29 | python |
| Skyvern-AI/skyvern | Automate browser based workflows with AI | 11 | python |
π New Papers
| Title | Category | Hotness | Link |
|---|---|---|---|
| AI-integrated models for assessing agricultural resilience | cs.AI | 0 | Open |
| Adversarial Social Epistemology for Assemblies of Humans and Large Language Models | cs.AI | 0 | Open |
| Aligning Clinical Needs and AI Capabilities: A Survey on LLMs for Medical Reasoning | cs.AI | 0 | Open |
| Alignment Plausibility: A New Standard for Assuring AI in Healthcare | cs.AI | 0 | Open |
| Idiobionics: The Unification of Privacy and Intelligent Robotic Prostheses | cs.AI | 0 | Open |
| Infinity-Parser2 Technical Report | cs.AI | 0 | Open |
| VectorizationLLM: Smart Vectorization Based AI Assistant | cs.AI | 0 | Open |
| A Graph Neural Network Model for Real-Time Gesture Recognition Based on sEMG Signals | cs.AI | 0 | Open |
| Nigeria Machinery: A Low-Resource Industrial Dataset with a Domain-Grounded Reasoning Layer | cs.AI | 0 | Open |
| Agentic Neural Architecture Search | cs.AI | 0 | Open |
| Concretized Proposition Prompting Resolves Composition-Knowledge Dichotomy in Large Language Models | cs.AI | 0 | Open |
| A safety-oriented hypothetico-deductive framework for AI-assisted differential diagnosis | cs.AI | 0 | Open |
| When LLMs Agree, Are They Right? Auditing Self-Consistency and Cross-Model Agreement as Confidence Signals | cs.AI | 0 | Open |
| PARA-PV: Physics-Aware Retrieval-Augmented PV Prediction Based on Frozen Foundation Model and Distribution Shift Correction | cs.AI | 0 | Open |
| Answer Set Programming Energised! End-to-End Neurosymbolic Reasoning and Learning with ASP and Energy Based Models | cs.AI | 0 | Open |
π¦ Twitter/X Highlights
| Account | Tweet Summary |
|---|---|
| sama | there are a lot of benchmarks that suggest 5.6 sol is the best model in the world right now, but the most reliable way to tell is that elon is obsessed with me again Post |
Newsletter
- tldr: THE PEOPLE WHO WILL THRIVE IN THE AI AGE (22 MINUTE READ)
- tldr: CHINA MAY RESTRICT ACCESS TO ITS MOST POWERFUL AI MODELS (4 MINUTE READ)
- tldr: MICROSOFT REPLACES OPENAI, ANTHROPIC WITH OWN AI IN SOME APPS (2 MINUTE READ)
- tldr: HOW TO USE AI AGENTS BETTER THAN 99% OF PEOPLE (32 MINUTE READ)
- tldr: WILL SOMEONE FINALLY BLINK IN THE AI SPENDING WAR? (8 MINUTE READ)
- tldr: WHY IS CHATGPT FOR MAC SO GOOD? (5 MINUTE READ)
- tldr: THEORY OF CONSTRAINTS, AI, AND CODE REVIEW (10 MINUTE READ)
- tldr: META IS QUIETLY LAUNCHING POCKET, AN APP FOR VIBE-CODING AND SCROLLING SMALL 'GIZMOS' (1 MINUTE READ)
- tldr: WHAT IS PRODUCT CRAFT IN THE AGE OF AI AND DESIGN SYSTEMS? (5 MINUTE READ)
- tldr: AI-POWERED SKETCH TO REALITY (WEBSITE)
- tldr: AGENTIC AI ADOPTION IS ON FIRE AT UBER (2 MINUTE READ)
- tldr: EXPANDING MANAGED AGENTS IN GEMINI API: BACKGROUND TASKS, REMOTE MCP AND MORE (2 MINUTE READ)
- tldr: USE CLAUDE COWORK ON WEB, DESKTOP, AND MOBILE (3 MINUTE READ)
- tldr: AI GIANTS ARE HANDING OUT TONS OF FREE COMPUTING POWER TO GRAB STARTUP SHARE (8 MINUTE READ)
- tldr: AI-NATIVE STARTUPS HIRE FEWER JUNIORS AND MORE ELITES, HARVARD STUDY FINDS (3 MINUTE READ)
- tldr: THE AGENT-ERA CAREER (7 MINUTE READ)
- tldr: GITLOST: HOW WE TRICKED GITHUB'S AI AGENT INTO LEAKING PRIVATE REPOS (3 MINUTE READ)
- tldr: ROGUE AGENT: HOW A SINGLE CODE BLOCK COULD HIJACK YOUR AI CONVERSATIONS IN GOOGLE'S DIALOGFLOW (6 MINUTE READ)
- tldr: WRITER AI FLAW COULD LET AGENT PREVIEWS LEAK SESSION TOKENS ACROSS TENANTS (3 MINUTE READ)
- tldr: THE βFIRST' AI-RUN RANSOMWARE ATTACK STILL NEEDED A HUMAN (3 MINUTE READ)
- tldr: GPT-5.6 SOL, ALONG WITH TERRA AND LUNA, WILL LAUNCH PUBLICLY THIS THURSDAY (1 MINUTE READ)
- tldr: FACING US EXPORT CONTROLS, CHINA'S DEEPSEEK PLANS TO MAKE ITS OWN CHIPS (2 MINUTE READ)
- tldr: MINIMAX M3: HOW SPARSE ATTENTION MAKES LONG-HORIZON AGENTS PRACTICAL (11 MINUTE READ)
- rundown-ai: Muse Image - Metaβs new in-house AI image generator
- rundown-ai: Willow Frontier Mini - Willow's free, unlimited AI dictation model
- rundown-ai: Gemini Managed Agents - Agent API with background tasks, remote tools
- rundown-ai: Anthropic said that Fable 5 will now be available in subscription plans until July 12, after initially planning to move it off plans and onto usage credits on Wednesday.
- rundown-ai: Tufa Labs took first place in ARC-AGI-3's $37.5K milestone contest, wrapping a small Qwen model in an open-sourced coding agent to tackle the game-based benchmark.
- rundown-ai: Microsoft is reportedly running Excel and Outlook prompts on its in-house MAI models, with Mustafa Suleyman pushing to "ultimately eliminate" its Anthropic bill.
- rundown-ai: Anthropic also announced that its Claude Cowork platform is being added to mobile and web in beta, giving users the ability to move between devices during tasks.
- rundown-ai: Figma acqui-hired the team behind AI app-builder and agent platform Bud (formerly Orchids).
- rundown-ai: Meta's impressive AI image debut
- rundown-ai: Elon Musk pitched the release on X as "an Opus-class model, but faster, more token-efficient and lower costβ, with an even bigger model coming next month.
- rundown-ai: The Rundown: Serko.ai is a conversational AI travel agent built for modern travellers that handles flights, hotels, and trip disruptions in a single chat β with no forms, tabs, or manual follow-up.
- rundown-ai: The Rundown: ByteDance rolled out Seedream 5.0 Pro, a new image model that claims to go beyond generation to "understand designβ with powerful editing features, and the first release that looks to near the frontier level
- rundown-ai: GPT-Live - OpenAIβs new SOTA, more natural voice model
- rundown-ai: Grok 4.5 - SpaceXAI and Cursorβs powerful, low-cost, efficient new model
- rundown-ai: Seedream 5.0 Pro - ByteDance's image AI built for design work
- rundown-ai: Metaβs new in-house AI image model - first seen 2026-07-10
- rundown-ai: The model slots into the No. 2 spot across Arenaβs text-to-image and editing leaderboards, trailing only OpenAIβs GPT Image 2. - first seen 2026-07-10
- rundown-ai: Beijing considers new restrictions for Chinese AI - first seen 2026-07-10
- rundown-ai: Run better 1-on-1s with employees using AI - first seen 2026-07-10
- rundown-ai: The observability AI teams can't skip - first seen 2026-07-10
- rundown-ai: DoorDash grades the AI reviewing its code - first seen 2026-07-10
- rundown-ai: Read our last AI newsletter: The part of Claudeβs brain nobody built - first seen 2026-07-10
Repeated From Recent Briefings
- wonderwhy-er/DesktopCommanderMCP β This is MCP server for Claude that gives it terminal control, file system search and diff file editing capabilities - first seen 2026-06-10
- Why Can't I Open My Drawer? Mitigating Object-Driven Shortcuts in Zero-Shot Compositional Action Recognition - first seen 2026-07-03
- stablyai/orca β Orca is the ADE for working with a fleet of parallel agents. Run any coding agent with your own subscription. Available on desktop and mobile. - first seen 2026-06-21
- What are the best practical alternatives to Codex and Claude Code for daily coding work - first seen 2026-07-10
- google-labs-code/stitch-skills β A library of Agent Skills designed to work with the Stitch MCP server. Each skill follows the Agent Skills open standard, for compatibility with coding agents such as Antigravity, Gemini CLI, Claude Code, Cursor. - first seen 2026-07-03
- davila7/claude-code-templates β CLI tool for configuring and monitoring Claude Code - first seen 2026-06-11
- openai/codex β Lightweight coding agent that runs in your terminal - first seen 2026-06-24
- LongE2V: Long-Horizon Event-based Video Reconstruction, Prediction, and Frame Interpolation with Video Diffusion Models - first seen 2026-07-10
- Why doesn't the ML research community limit the number of submissions per author? [D] - first seen 2026-07-10
- anthropics/claude-code β Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands. - first seen 2026-07-02
- ... plus 221 more repeated items in processed data