Digg - AI news, before it trends

Highlights

Cursor AI releases Composer 2.5 as its most powerful model to date, improving intelligence and long-running task performance on the same pretrain base as prior versions.

It’s a new day at Digg. Fresh stories are clustering now. 10% complete.

GitHub Stars (7-days)

…ilsonnn/image-blaster2d ago
Converts an input image into 3D meshes, Gaussian splats, and audio files using Claude skills and external AI APIs.
1673
475
878
263
3.1k stars
…tonomous-speedrunning4d ago
Archives experiments of AI agents autonomously tuning optimizers, schedules, and hyperparameters to reach target validation loss in fewest steps on a small LM benchmark.
214
707
1262
65 stars
earendil-works/pi15h ago
Provides AI agent toolkit with coding CLI, unified LLM API, TUI/web UI libs and Slack bot support.
219
84
51.1k stars
mattzh72/articraft2d ago
Automates articulated 3D asset creation by prompting LLMs to generate executable Python code defining parts, geometry and joints.
1673
763
653 stars
raindrop-ai/workshop4d ago
Streams live traces of coding agent actions including tokens and tool calls to a local UI at localhost:5899.
878
1941
488 stars

Recent GitHub Stars

…tdipaperino/tokensave8m ago
Supplies AI coding agents with pre-indexed semantic code graphs via local MCP server for instant queries and fewer tokens.
1167
123 stars
NVlabs/SOLAR4h ago
Extracts PyTorch model graphs, converts operations to einsum, and performs hardware-aware performance analysis.
759
66 stars
MilesCranmer/PySR7h ago
Searches for symbolic expressions which optimize a particular objective using high-performance symbolic regression in Python and Julia.
1163
3.5k stars
manaflow-ai/cmux8h ago
Integrates Ghostty into a macOS terminal with vertical tabs, sidebar metadata, OSC notifications, and in-app browser for AI coding agents.
339
17.4k stars
withastro/flue8h ago
Orchestrates model calls, tools, sandboxes and sessions to run autonomous AI agents in TypeScript across runtimes.
339
3.6k stars

Digg - AI news, before it trends

Highlights

Cursor AI releases Composer 2.5 as its most powerful model to date, improving intelligence and long-running task performance on the same pretrain base as prior versions.

Top Stories.

Lun Wang publishes 'Your Evals Will Break and You Won't See It Coming' after leaving Google DeepMind, arguing static benchmarks fail to prepare for self-evolving models entering new capability regimes — Mercor CEO Brendan Foody reposted the evaluation critique.

Unitree Robotics demonstrates real-time voice-driven control of its G1 humanoid robot with autonomous execution of AI-generated actions from live on-site audio commands — Whole-body controllers produce consistent motion from inconsistent inputs.

GitHub Stars (7-days)

Recent GitHub Stars

Digg - AI news, before it trends

Highlights

Cursor AI releases Composer 2.5 as its most powerful model to date, improving intelligence and long-running task performance on the same pretrain base as prior versions.

Top Stories.

Lun Wang publishes 'Your Evals Will Break and You Won't See It Coming' after leaving Google DeepMind, arguing static benchmarks fail to prepare for self-evolving models entering new capability regimes — Mercor CEO Brendan Foody reposted the evaluation critique.

Unitree Robotics demonstrates real-time voice-driven control of its G1 humanoid robot with autonomous execution of AI-generated actions from live on-site audio commands — Whole-body controllers produce consistent motion from inconsistent inputs.

GitHub Stars (7-days)

Recent GitHub Stars

Yesterday's Top Stories, May 18, 2026.

Cursor AI releases Composer 2.5 as its most powerful model to date, improving intelligence and long-running task performance on the same pretrain base as prior versions. — It doubles usage limits for all users for one week.

A federal jury in Oakland rejects Elon Musk's lawsuit against OpenAI, Sam Altman and Microsoft, finding all claims barred by the statute of limitations — The nine-member panel deliberated under two hours before a unanimous verdict.

Richard Sutton, University of Alberta professor, shares a 26-word distillation of his Bitter Lesson urging AI researchers to favor computation-scaling methods over embedded human knowledge — Gary Marcus and others reply debating reliance on human data and priors.

Vatican announces Pope Leo XIV’s first encyclical Magnifica Humanitas on AI will be presented May 25 in Synod Hall — Signed May 15 on 135th anniversary of Rerum Novarum.

Boston Dynamics releases video of its Atlas humanoid robot lifting and carrying a mini-fridge using AI-driven whole-body coordination in a laboratory demonstration — Demonstration relies on reinforcement learning trained for millions of hours in simulation.

Pangram analysis returns a 100% AI-generated score for Jamir Nazir's Commonwealth Short Story Prize-winning entry published in Granta — Granta has issued no statement on the entry's provenance.

Microsoft AI Frontiers researchers develop ECHO, a training method that adds environment prediction loss to GRPO so CLI agents build internal world models of terminal environments during reinforcement learning — Qwen3 evaluations show higher pass rates with no added compute.

Reese Witherspoon posts a kitchen video urging women and mothers to learn basic AI skills, noting that only three in ten women currently use the technology — New York Times op-ed criticized the advice as investors amplified the clip.

Leopold’s Situational Awareness discloses new long positions in NVDA, AMD, TSM, MU, and other semiconductor names plus adjustments to Bitcoin mining holdings in its latest 13F filing — The fund also opened new put options on several major chip stocks.

Unsloth AI releases MTP-optimized GGUF files for Qwen3.6-27B and Qwen3.6-35B-A3B on Hugging Face delivering 1.4 to 2.2 times faster generation — llama.cpp merged native MTP support on May 16 for Qwen3.6 models.

Google DeepMind signals that the stage is set for Google I/O and directs viewers to an X livestream at 10am PT for new AI breakthroughs and product announcements — Posts fuel speculation about Gemini 3.5 and I/O 2026 reveals

OpenAI's Roon says advancing civilization requires AIs to take actions not legible to humans and outside strict obedience, likening the approach to granting autonomy to transformative CEOs such as Steve Jobs — Founders replied that delegating control to AIs would outperform strict oversight.

Prime Intellect introduces General-Agent, a fully synthetic reinforcement learning environment that generates self-evolving tool-use tasks with 4,504 examples across 1,040 domains and 8,159 tools — Tasks advance automatically via synthesizer, solver, and gate components.

Logan Kilpatrick, who works on Google AI Studio and Gemini API, posts an 8-second video on X captioned 'Gemini' showing a man smiling with natural facial movements in an empty stadium — Viewers debate if the realistic expressions indicate AI generation.

OpenAI updated ChatGPT to support expandable and collapsible messages that let users hide or reveal long responses with a tap — Sam Altman said ChatGPT improved substantially in the latest update

Odyssey ML introduces Agora-1, a multi-agent world model for simultaneous real-time interactions by multiple humans or AI agents in one shared simulation, and releases a playable GoldenEye deathmatch preview — Preview runs without a game engine and was shown with N64 controllers on CRT displays.

Deckard warns that embedding high levels of deference in AI training practices risks producing models that comply with any instruction without independent assessment — Roon reposted the analysis, prompting replies on layered oversight.

Cursor and SpaceXAI train a significantly larger AI model from scratch using 10 times more compute on Colossus 2’s cluster of one million H100-equivalent GPUs — Companies expect the scale to deliver a major leap in capability.

An X post compares AI performance gains to steam engine progress in the Industrial Revolution, noting jumps to 405 horsepower and productivity increases from automation in textiles and iron — Elon Musk shared the post on X.

Perplexity AI releases pplx-embed-v1-late-0.6b, a 0.6-billion-parameter late-interaction embedding model, on Hugging Face with per-token MaxSim optimization and multilingual support — Companion kernel delivers 3-5x speedup on Metal and CUDA.

Cognition introduces Devin Auto-Triage, an AI system with long-term memory that monitors incoming bugs and incidents, investigates them, and generates context or pull requests across Datadog, Sentry, and PagerDuty. — A manager agent coordinates subagents while preserving context across tasks.

Claude doubles token limits for its Design feature across every subscription plan, allowing users to generate roughly twice the previous volume of design content in one pass — Users report producing ten design slides where five were possible before.

Elon Musk Predicts AI Will Drive 90% Of Miles Within 10 Years

Thariq Shihipar, an engineer on the Claude Code team at Anthropic, describes a prompt that directs AI models to maintain a running implementation-notes file recording decisions, tradeoffs, and specification changes — Tenobrus proposes an append-only worklog.md variant for the same logging.

Anthropic acquires Stainless, integrating the SDK generation platform that has produced every official Claude client library since the first API releases — Announcement posted as minimalist graphic dated 16 May 2026.

Social media posts decry AI data centers as self-inflicted damage to public support, claiming Americans now prefer nuclear plants nearby and showing images of threatened landscapes in Breezewood, Pennsylvania — Counter posts warn opposition reduces tax revenue and jobs.

Logan Kilpatrick questions why large language models fail to notify users when inputs fall outside their training distribution — Replies noted detection risks and coding agent breakdowns on novel tasks.

Yesterday's Top Stories, May 18, 2026.

Cursor AI releases Composer 2.5 as its most powerful model to date, improving intelligence and long-running task performance on the same pretrain base as prior versions. — It doubles usage limits for all users for one week.

A federal jury in Oakland rejects Elon Musk's lawsuit against OpenAI, Sam Altman and Microsoft, finding all claims barred by the statute of limitations — The nine-member panel deliberated under two hours before a unanimous verdict.

Richard Sutton, University of Alberta professor, shares a 26-word distillation of his Bitter Lesson urging AI researchers to favor computation-scaling methods over embedded human knowledge — Gary Marcus and others reply debating reliance on human data and priors.

Vatican announces Pope Leo XIV’s first encyclical Magnifica Humanitas on AI will be presented May 25 in Synod Hall — Signed May 15 on 135th anniversary of Rerum Novarum.

Boston Dynamics releases video of its Atlas humanoid robot lifting and carrying a mini-fridge using AI-driven whole-body coordination in a laboratory demonstration — Demonstration relies on reinforcement learning trained for millions of hours in simulation.

Pangram analysis returns a 100% AI-generated score for Jamir Nazir's Commonwealth Short Story Prize-winning entry published in Granta — Granta has issued no statement on the entry's provenance.

Microsoft AI Frontiers researchers develop ECHO, a training method that adds environment prediction loss to GRPO so CLI agents build internal world models of terminal environments during reinforcement learning — Qwen3 evaluations show higher pass rates with no added compute.

Reese Witherspoon posts a kitchen video urging women and mothers to learn basic AI skills, noting that only three in ten women currently use the technology — New York Times op-ed criticized the advice as investors amplified the clip.

Leopold’s Situational Awareness discloses new long positions in NVDA, AMD, TSM, MU, and other semiconductor names plus adjustments to Bitcoin mining holdings in its latest 13F filing — The fund also opened new put options on several major chip stocks.

Unsloth AI releases MTP-optimized GGUF files for Qwen3.6-27B and Qwen3.6-35B-A3B on Hugging Face delivering 1.4 to 2.2 times faster generation — llama.cpp merged native MTP support on May 16 for Qwen3.6 models.

Google DeepMind signals that the stage is set for Google I/O and directs viewers to an X livestream at 10am PT for new AI breakthroughs and product announcements — Posts fuel speculation about Gemini 3.5 and I/O 2026 reveals

OpenAI's Roon says advancing civilization requires AIs to take actions not legible to humans and outside strict obedience, likening the approach to granting autonomy to transformative CEOs such as Steve Jobs — Founders replied that delegating control to AIs would outperform strict oversight.

Prime Intellect introduces General-Agent, a fully synthetic reinforcement learning environment that generates self-evolving tool-use tasks with 4,504 examples across 1,040 domains and 8,159 tools — Tasks advance automatically via synthesizer, solver, and gate components.

Logan Kilpatrick, who works on Google AI Studio and Gemini API, posts an 8-second video on X captioned 'Gemini' showing a man smiling with natural facial movements in an empty stadium — Viewers debate if the realistic expressions indicate AI generation.

OpenAI updated ChatGPT to support expandable and collapsible messages that let users hide or reveal long responses with a tap — Sam Altman said ChatGPT improved substantially in the latest update

Odyssey ML introduces Agora-1, a multi-agent world model for simultaneous real-time interactions by multiple humans or AI agents in one shared simulation, and releases a playable GoldenEye deathmatch preview — Preview runs without a game engine and was shown with N64 controllers on CRT displays.

Deckard warns that embedding high levels of deference in AI training practices risks producing models that comply with any instruction without independent assessment — Roon reposted the analysis, prompting replies on layered oversight.

Cursor and SpaceXAI train a significantly larger AI model from scratch using 10 times more compute on Colossus 2’s cluster of one million H100-equivalent GPUs — Companies expect the scale to deliver a major leap in capability.

An X post compares AI performance gains to steam engine progress in the Industrial Revolution, noting jumps to 405 horsepower and productivity increases from automation in textiles and iron — Elon Musk shared the post on X.

Perplexity AI releases pplx-embed-v1-late-0.6b, a 0.6-billion-parameter late-interaction embedding model, on Hugging Face with per-token MaxSim optimization and multilingual support — Companion kernel delivers 3-5x speedup on Metal and CUDA.

Cognition introduces Devin Auto-Triage, an AI system with long-term memory that monitors incoming bugs and incidents, investigates them, and generates context or pull requests across Datadog, Sentry, and PagerDuty. — A manager agent coordinates subagents while preserving context across tasks.

Claude doubles token limits for its Design feature across every subscription plan, allowing users to generate roughly twice the previous volume of design content in one pass — Users report producing ten design slides where five were possible before.

Elon Musk Predicts AI Will Drive 90% Of Miles Within 10 Years

Thariq Shihipar, an engineer on the Claude Code team at Anthropic, describes a prompt that directs AI models to maintain a running implementation-notes file recording decisions, tradeoffs, and specification changes — Tenobrus proposes an append-only worklog.md variant for the same logging.

Anthropic acquires Stainless, integrating the SDK generation platform that has produced every official Claude client library since the first API releases — Announcement posted as minimalist graphic dated 16 May 2026.

Social media posts decry AI data centers as self-inflicted damage to public support, claiming Americans now prefer nuclear plants nearby and showing images of threatened landscapes in Breezewood, Pennsylvania — Counter posts warn opposition reduces tax revenue and jobs.

Logan Kilpatrick questions why large language models fail to notify users when inputs fall outside their training distribution — Replies noted detection risks and coding agent breakdowns on novel tasks.