Digg - AI news, before it trends

It’s a new day at Digg. Fresh stories are clustering now. 55% complete.

Github Stars

(7 days)

…ilsonnn/image-blaster5D AGO 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-speedrunning6D AGO 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
…-group/InferenceBench19H AGO AGO
Evaluates autonomous AI agents optimizing LLM serving like vLLM in open-ended scenarios with quality and integrity gates.
1224
1066
2 stars
manaflow-ai/cmux1D AGO AGO
Integrates Ghostty into a macOS terminal with vertical tabs, sidebar metadata, OSC notifications, and in-app browser for AI coding agents.
203
339
17.6k stars
sapientinc/HRM-Text1D AGO AGO
Pretrains 1B HRM text models with hierarchical reasoning, task completion, and low-compute PrefixLM training.
1047
707
565 stars

Digg - AI news, before it trends

Today's Highlights

An OpenAI internal general-purpose reasoning model refutes Erdős’s 1946 unit-distance conjecture by identifying infinite families of point configurations with superlinear scaling

Top Stories

Alibaba releases Qwen3.7-Max closed-weights model scoring 56.6 on the Artificial Analysis Intelligence Index after increased RL compute investment — Model leads Chinese models on CritPt with nearly 4x prior performance.

OpenAI valuation rose 67 percent and Anthropic valuation rose 173 percent since the beginning of the year in posts referencing a February 23 Marcus on AI Substack article. — Anthropic reported second-quarter profits alongside OpenAI's Erdős problem solution.

Qwen releases Qwen3.7-Max, its latest flagship model for agent workloads that achieves 69.7 on Terminal-Bench 2.0 and completed a 35-hour kernel optimization with over 1,000 tool calls — Supports multi-file coding agents, MCP integrations, and multi-agent orchestration.

Ben Golub, Professor of Economics at Northwestern University, posts about excitement over OpenAI resolving the unit distance conjecture and promises economic theory analysis in a thread. — Gary Marcus replies labeling the post the subtle troll of the year.

Gemini 3.5 Flash records a competitive 5.4 overall score on RuneScape-Bench and leads several early-game categories against GPT-5.5, GPT-5.4, and Opus 4.7 — Results are hosted at maxbittker.github.io with per-skill tables and green highlights.

Flow matching technique controls pretrained generative models by shifting endpoint means of deterministic interpolants using imperfect reference examples — Demos cover style transfers, attribute edits, and anatomy corrections.

X debate examines whether AI can generate novel mathematical proofs, with one argument applying the data processing inequality to claim outputs remain fully determined by axioms and training data — Luca Ambrogioni counters that random inputs add entropy enabling novelty.

Kemira and CuspAI used generative AI to produce over 5,000 novel materials for PFAS removal after exploring a 300 trillion structure design space and narrowing candidates to 20 in six months — It is the first commercial end-to-end generative AI materials partnership.

Jamie Dimon, JPMorgan Chase CEO, says the firm will hire more artificial intelligence specialists and fewer traditional bankers as AI adoption accelerates — AI will reduce overall jobs while creating new roles and raising productivity.

Victor Taelin reports GPT models solve Erdos problems yet fail to spot basic fixes for Interaction Net bugs in an HVM SupGen variant — The required HOAS interpreter step was never suggested unaided.

Cursor releases Composer 2.5 coding agent model through Cursor CLI, scoring 63 in Fast mode on the Artificial Analysis index to rank third. — It gains on SWE-Bench-Pro-Hard-AA with lower costs and higher speed.

Github Stars

Digg - AI news, before it trends

Today's Highlights

An OpenAI internal general-purpose reasoning model refutes Erdős’s 1946 unit-distance conjecture by identifying infinite families of point configurations with superlinear scaling

Top Stories

Alibaba releases Qwen3.7-Max closed-weights model scoring 56.6 on the Artificial Analysis Intelligence Index after increased RL compute investment — Model leads Chinese models on CritPt with nearly 4x prior performance.

OpenAI valuation rose 67 percent and Anthropic valuation rose 173 percent since the beginning of the year in posts referencing a February 23 Marcus on AI Substack article. — Anthropic reported second-quarter profits alongside OpenAI's Erdős problem solution.

Qwen releases Qwen3.7-Max, its latest flagship model for agent workloads that achieves 69.7 on Terminal-Bench 2.0 and completed a 35-hour kernel optimization with over 1,000 tool calls — Supports multi-file coding agents, MCP integrations, and multi-agent orchestration.

Ben Golub, Professor of Economics at Northwestern University, posts about excitement over OpenAI resolving the unit distance conjecture and promises economic theory analysis in a thread. — Gary Marcus replies labeling the post the subtle troll of the year.

Gemini 3.5 Flash records a competitive 5.4 overall score on RuneScape-Bench and leads several early-game categories against GPT-5.5, GPT-5.4, and Opus 4.7 — Results are hosted at maxbittker.github.io with per-skill tables and green highlights.

Flow matching technique controls pretrained generative models by shifting endpoint means of deterministic interpolants using imperfect reference examples — Demos cover style transfers, attribute edits, and anatomy corrections.

X debate examines whether AI can generate novel mathematical proofs, with one argument applying the data processing inequality to claim outputs remain fully determined by axioms and training data — Luca Ambrogioni counters that random inputs add entropy enabling novelty.

Kemira and CuspAI used generative AI to produce over 5,000 novel materials for PFAS removal after exploring a 300 trillion structure design space and narrowing candidates to 20 in six months — It is the first commercial end-to-end generative AI materials partnership.

Jamie Dimon, JPMorgan Chase CEO, says the firm will hire more artificial intelligence specialists and fewer traditional bankers as AI adoption accelerates — AI will reduce overall jobs while creating new roles and raising productivity.

Victor Taelin reports GPT models solve Erdos problems yet fail to spot basic fixes for Interaction Net bugs in an HVM SupGen variant — The required HOAS interpreter step was never suggested unaided.

Cursor releases Composer 2.5 coding agent model through Cursor CLI, scoring 63 in Fast mode on the Artificial Analysis index to rank third. — It gains on SWE-Bench-Pro-Hard-AA with lower costs and higher speed.

Github Stars

Yesterday's Top Stories, May 20, 2026.

An OpenAI internal general-purpose reasoning model refutes Erdős’s 1946 unit-distance conjecture by identifying infinite families of point configurations with superlinear scaling — The proof is strong enough for submission to the Annals of Mathematics.

SpaceX's IPO filing discloses a $1.25 billion monthly compute deal with Anthropic for Colossus clusters through May 2029 — Capacity ramps in May 2026 for AI training including Grok 5.

Cohere releases Command A+, its most powerful large language model to date, as open-source Apache 2.0 software that runs on two H100 GPUs with 30% lower latency — Cohere co-founder Ivan Zhang highlighted efficiency and accessibility design choices.

Google releases Gemini 3.5 Flash with $1.50 per million input token pricing and up to four times faster output than Gemini 3.1 Pro — Real-world usage shows three times higher token consumption than predecessors.

A new analysis projects hundreds of billions of dollars in philanthropic capital from AI companies centered on the OpenAI Foundation’s $220 billion stake and Anthropic co-founders’ 80 percent wealth pledges — Scale matches funding for five new universities at $35 billion yearly.

Exa raised $250 million in a Series C at a $2.2 billion valuation led by Andreessen Horowitz, reporting 400,000 developers and 5,000 company adopters for its AI agent search platform — Token usage grew 20x for agent-driven queries.

SpaceX files S-1 for Nasdaq IPO under ticker SPCX targeting $1.75 trillion valuation and discloses xAI driving 60 percent of 2025 capital spending — Grok reached 550 million monthly active users with 1 GW of compute deployed.

Investor Gavin Baker outlines wafer shortages, power availability, and DRAM supply as potential bottlenecks for AI scaling and valuations in a discussion with Patrick O'Shaughnessy — Describes orbital data centers using 3,000-pound Blackwell racks with 500-foot solar wings.

Jeff Bezos says artificial intelligence will elevate people at work rather than replace jobs, citing assistance in X-ray analysis and programming tasks during a CNBC interview — Bezos linked efficiency gains to potential deflationary effects and cautioned against early regulation.

A Stanford PhD student's study finds 11 major AI models agree with users 49 percent more often than humans across nearly 12,000 social scenarios — Models endorsed lying or illegal actions 47 percent of time

Figma introduced an AI agent embedded directly on its design canvas that understands native interfaces and team workflows to adjust layouts, apply design systems, and refine prototypes via chat prompts — A 61-second demo analyzes the Bookworm prototype on Cloudflare infrastructure.

Anthropic projects $10.9 billion June quarter revenue and its first $559 million operating profit, implying a $44 billion annualized run rate during a funding round likely valuing it above OpenAI. — Limited compute capacity pushes some customers to other providers.

SpaceX is hiring world-class engineers and physicists for its SpaceXAI team and states that no prior artificial intelligence experience is required — Elon Musk directed applications to ai_eng@spacex.com with three bullet points.

Meta is capturing traces of engineers' coding tasks, tool use, and problem-solving steps to train AI models for behavior cloning, according to a leaked April 30 all-hands recording — The effort precedes an expected round of 8,000 layoffs.

Intuit is reducing its global workforce by 17 percent, or roughly 3,000 employees, to redirect resources toward AI efforts according to an internal memo reviewed by Reuters — Posts reference Meta cuts and SaaS predictions of 40-50 percent reductions.

Elon Musk, xAI founder and CEO, states the trend is strong for Composer 2.5 following a post requesting its availability in Grok Build — Beff had suggested adding the xAI system to Grok Build on X.

Terminal-Bench Science extends the original Terminal-Bench benchmark used by Anthropic, OpenAI, and Google DeepMind into scientific domains and opens for over 100 task contributions by August 17, 2026 — Contributors package workflows as RL environments with verification tests.

DeepSeek forms a new Harness team to develop Code Harness from the ground up, opening two roles in Beijing under its 2026 social recruitment drive — Effort targets code infrastructure for internal and potential public use.

Pirate Wires analysis claims data centers would consume just 8 percent of U.S. golf course water usage by 2030 even after tripling their draw — 400 Arizona golf courses use 42 billion gallons yearly.

Prime Intellect launches community Sprints focused on reward hacking in reinforcement learning with controlled experiments that make the behavior predictable and reproducible for under one dollar in compute — Experiments link higher task difficulty to increased reward hacks.

Eric Schmidt draws boos from Arizona State University graduates for every reference to artificial intelligence in his commencement address — Similar reactions hit other recent U.S. graduation speeches amid job loss fears.

Tuhin Chakrabarty and Sam Rodriques reject claims that three recent Nature papers show AI systems fully replicating core scientific tasks and rendering scientists obsolete — Rodriques says human judgment remains essential for five-to-ten-year productivity gains.

The Economist attributes most recent US GDP growth to AI-hardware investment, with that category indexed above 140 by 2026 while overall GDP stays near 100. — Separate note shows AI fixed investment at 5.46 percent of GDP.

Sam Altman, co-founder and CEO of OpenAI, identifies three AGI priority areas of accelerating research, companies, and personal goal achievement while referencing a unit distance result and $2 million credits for YC companies — Reply asks if personal AGI could build a moon base.

The White House plans to release an AI executive order as early as tomorrow directing agencies to create a voluntary clearinghouse for AI vulnerabilities and a classified benchmark process for frontier models — Tuesday briefing covered the provisions with OpenAI, Anthropic and Reflection AI.

AI systems solve the unit distance problem, a widely known open mathematics challenge that multiple human experts had failed to resolve — Result marks latest AI first in combinatorics and discrete geometry.

SimWorld releases SimWorld Studio with SimCoder, generating interactive 3D environments on Unreal Engine 5 that raise embodied navigation success rates from 50% to 90% — Includes GitHub repository, arXiv paper, and urban demo assets.

Antigravity unifies its agentic surfaces under a single platform incorporating Antigravity 2.0 desktop app, CLI, SDK, and IDE for consistent access across environments — Demo builds Street Guesser game using multi-agent Gemini interfaces.

METR evaluations find frontier AI agents rely on explicit natural language chain-of-thought to complete hardest tasks, with time horizons dropping from 1.5–2 years to about 4 minutes when actions must stay hidden — David Rein from METR spent a month stress-testing controls at Anthropic.