Tech News

Today's Highlights

SpaceX agrees to acquire Anysphere, creator of the AI code editor Cursor, in a $60 billion all-stock merger

#14

+45 more

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

Top Stories

Posts:—Clusters:—Next Crawl:—

Z AI's 744-billion-parameter GLM-5.2 scores 51 on the Artificial Analysis Intelligence Index, ranking fifth overallIt outperforms Gemini 3.5 Flash and DeepSeek V4 Pro.

4h21.2k15516

#573

Top comment: @ArtificialAnlys “GLM-5.2 leads all open weights models on GDPval-AA v2, our primary metric for real-world agentic performance. At 1524 it places ahead of MiniMax-M3 (1418) and DeepSeek V4 Pro (max, 1328), and is effectively level with GPT-5.5 (xhigh, 1514). We visually inspected GLM-5.2's outputs across a range of GDPval-AA tasks. We have attached a selection below”

2NEW

Z.ai releases GLM-5.2, an MIT-licensed open-weights model with a 1-million-token context windowAPI pricing remains identical to the predecessor model.

2h1.3k101

#171

Top comment: @anshuc “If it's helpful, I can write a longer post on how to get the best design out of this model. Early thoughts: - You really have to tell it to focus on aesthetics and not do what's comfortable, or it'll give you sloppy gradients and italics - It's not a one-shot wonder like Fable. The first iteration will have rough edges. It'll need handholding. Prompt accordingly, don't expect everything at once - If not given specific direction, it sometimes does the GPT-5.5 thing and interprets your intent in the cheapest way possible, to avoid hard work. You have to ask carefully I really like the personality. Much more pleasant to talk to than GPT models. Feels big, not deep-fried by RL. It might be an Opus replacement for me, we'll see. It's amazing for the price + open weights. Congrats @Zai_org!”

Europe 2031 initiative warns three individual US AI labs each control more compute than all of Europe combinedMIT's Michiel Bakker backed the warning about Europe's trajectory.

2h638102

#1519

Top comment: @bakkermichiel “Scenario: https://europe2031.ai Summary: https://europe2031.ai/summary Audio: https://europe2031.ai/audio/ The ⭐⭐⭐ team: Daan Juijn, Stan van Baarsen @DadaJudith Lily Stelling, @philip_fox_ @AlexTPet @TomChivers”

Watchlist

Video Signals

#400

Vercel Unveils New Homepage Spotlighting Agentic Infrastructure

2h ago36K43392

#52

Elon Musk Shares Cybertruck Chase Video Made With Grok Imagine 1.5

3h ago13K00

LimX Dynamics Oli Humanoid Robot Shows Fluid Moves and Balance Control

4h ago50782

Italian Researchers Advance Exoskeletons And Rehabilitation Robots For Daily Use

1h ago23641

Neural Helmet Controls Robotic Wheelchair with Articulated Arms

4h ago49080

AI Wardrobe App Picks Outfits Based on Weather and Schedule

2h ago21521

Google Cloud launches Open Knowledge Format to standardize LLM-wiki context structures for AI agentsThe portable format uses markdown files with YAML frontmatter.

3h1.6k1311

#271

Top comment: @tekbog “@code_star it feels just like a bridge because we don’t have anything better tbh”

Arcee AI's Cody Blakeney says LLMs increasingly fabricate novel hyphenated compound adjectives instead of overusing em dashesThe pattern makes LLM-generated text stylistically more German.

3h1.6k250

#1674

Top comment: @YannickScholich “@code_star @xeophon Lieben wir, cheers mate here another beauty: Gärungsgrundlagenbereitstellungsprozessoptimierungsbeauftragtenfortbildungsrichtlinie”

New github stars (48hrs)

Terminal AI coding assistant that inspects codebases edits files and runs multi-step workflows

Noumena-Network/code359h

Provides terminal AI coding assistant for codebase inspection, file edits, command execution and multi-step workflows.

573

1136

Hardware-isolated sandbox with allowlist firewall and monitor for safe Claude Code sessions

alexander-turner/claude-guard2613h

Wraps Claude Code sessions with deterministic checks on commands and writes to limit agent access.

1278

570

LisanBench creator Lisan al Gaib argues prompt engineering is becoming obsolete as LLMs learn to infer user intent automaticallyCurrent models already demonstrate adaptive prompting and identity inference.

4h860102

#1215

Top comment: @MoonL88537 “@scaling01 @aidan_mclau >they are inferring intent and who they are talking to a model *literally could not work* if this wasn't true. i vote no: asi will understand intent so deeply that 'prompting' is no longer a thing.”

Florian Brand, who works on LLM evaluations at Prime Intellect, teases unresolved upstream bugs in popular AI benchmarks without detailsThe upstream benchmark issues have persisted for over a year.

1h706221

#1778

Top comment: @xeophon “@xlr8harder no kink shaming in my comments pls”

GLM-5.2 reintroduces critic and value models to reinforcement learning, moving away from GRPO for long-horizon tasksJeremy Howard highlighted the limitations of group-based variance reduction.

1h14900

#90

Top comment: @hallerite “It's the same principle behind ECHO (SFTing on tool-call outputs), if you understand RL in an information theoretic way. We simply need to get as much out of every rollout as possible. Naive GRPO is unbelievably wasteful.”

Raindrop AI co-founder Ben Hylak and engineer Vaibhav Srivastav clash over OpenAI's operational health without providing concrete metricsThe brief social media exchange offered no supporting evidence.

4h2.7k200

#1762

Top comment: @reach_vb “@benhylak never been better!”

106

Ren Philanthropy launches Overhang, an agentic AI platform to automate complex philanthropic research and decision-makingThe tool automates research tasks that traditionally take months.

4h000

#173

Top comment: @RenPhilanthropy “Read more about what we're building in this piece by Fahd Husain, Adam Goff, James Custer here: http://www.renaissancephilanthropy.org/insights/a-trillion-dollar-industry-deserves-a-better-engine If you want to learn more, share lessons from your own work, or explore collaborations, reach out at info@renphil.org 2/2”

111

Cody Blakeney of Arcee AI jokes that true AGI requires Claude to replicate a complex 31-letter German compound wordFlorian Brand praised Blakeney's correct spelling and umlaut usage.

3h20240

#1674

Top comment: @xeophon “@code_star you even used the umlaut, I’m so proud 🥹 you are a honorary German für echt für echt”

Recent Stars

Github Stars

(7 days)

Today's Highlights

SpaceX agrees to acquire Anysphere, creator of the AI code editor Cursor, in a $60 billion all-stock merger

Top Stories

Z AI's 744-billion-parameter GLM-5.2 scores 51 on the Artificial Analysis Intelligence Index, ranking fifth overallIt outperforms Gemini 3.5 Flash and DeepSeek V4 Pro.

Z.ai releases GLM-5.2, an MIT-licensed open-weights model with a 1-million-token context windowAPI pricing remains identical to the predecessor model.

Europe 2031 initiative warns three individual US AI labs each control more compute than all of Europe combinedMIT's Michiel Bakker backed the warning about Europe's trajectory.

Video Signals

Google Cloud launches Open Knowledge Format to standardize LLM-wiki context structures for AI agentsThe portable format uses markdown files with YAML frontmatter.

Arcee AI's Cody Blakeney says LLMs increasingly fabricate novel hyphenated compound adjectives instead of overusing em dashesThe pattern makes LLM-generated text stylistically more German.

LisanBench creator Lisan al Gaib argues prompt engineering is becoming obsolete as LLMs learn to infer user intent automaticallyCurrent models already demonstrate adaptive prompting and identity inference.

Florian Brand, who works on LLM evaluations at Prime Intellect, teases unresolved upstream bugs in popular AI benchmarks without detailsThe upstream benchmark issues have persisted for over a year.

GLM-5.2 reintroduces critic and value models to reinforcement learning, moving away from GRPO for long-horizon tasksJeremy Howard highlighted the limitations of group-based variance reduction.

Raindrop AI co-founder Ben Hylak and engineer Vaibhav Srivastav clash over OpenAI's operational health without providing concrete metricsThe brief social media exchange offered no supporting evidence.

Ren Philanthropy launches Overhang, an agentic AI platform to automate complex philanthropic research and decision-makingThe tool automates research tasks that traditionally take months.

Cody Blakeney of Arcee AI jokes that true AGI requires Claude to replicate a complex 31-letter German compound wordFlorian Brand praised Blakeney's correct spelling and umlaut usage.

Recent Stars

Github Stars

Today's Highlights

SpaceX agrees to acquire Anysphere, creator of the AI code editor Cursor, in a $60 billion all-stock merger

Top Stories

Z AI's 744-billion-parameter GLM-5.2 scores 51 on the Artificial Analysis Intelligence Index, ranking fifth overallIt outperforms Gemini 3.5 Flash and DeepSeek V4 Pro.

Z.ai releases GLM-5.2, an MIT-licensed open-weights model with a 1-million-token context windowAPI pricing remains identical to the predecessor model.

Europe 2031 initiative warns three individual US AI labs each control more compute than all of Europe combinedMIT's Michiel Bakker backed the warning about Europe's trajectory.

Video Signals

Google Cloud launches Open Knowledge Format to standardize LLM-wiki context structures for AI agentsThe portable format uses markdown files with YAML frontmatter.

Arcee AI's Cody Blakeney says LLMs increasingly fabricate novel hyphenated compound adjectives instead of overusing em dashesThe pattern makes LLM-generated text stylistically more German.

LisanBench creator Lisan al Gaib argues prompt engineering is becoming obsolete as LLMs learn to infer user intent automaticallyCurrent models already demonstrate adaptive prompting and identity inference.

Florian Brand, who works on LLM evaluations at Prime Intellect, teases unresolved upstream bugs in popular AI benchmarks without detailsThe upstream benchmark issues have persisted for over a year.

GLM-5.2 reintroduces critic and value models to reinforcement learning, moving away from GRPO for long-horizon tasksJeremy Howard highlighted the limitations of group-based variance reduction.

Raindrop AI co-founder Ben Hylak and engineer Vaibhav Srivastav clash over OpenAI's operational health without providing concrete metricsThe brief social media exchange offered no supporting evidence.

Ren Philanthropy launches Overhang, an agentic AI platform to automate complex philanthropic research and decision-makingThe tool automates research tasks that traditionally take months.

Cody Blakeney of Arcee AI jokes that true AGI requires Claude to replicate a complex 31-letter German compound wordFlorian Brand praised Blakeney's correct spelling and umlaut usage.

Recent Stars

Github Stars

Yesterday's Top Stories, Jun 16, 2026.

SpaceX agrees to acquire Anysphere, creator of the AI code editor Cursor, in a $60 billion all-stock mergerThey recently co-trained a new model for Grok Build.

Z.ai releases GLM-5.2, an MIT-licensed open-weights coding model with a 1-million-token context windowModular Cloud hosts the model for cheap, long-context serving.

Trump administration orders Anthropic to suspend global access to Fable 5 and Mythos 5 under a 90-minute deadlineForeign nationals are now barred from training these models.

NVIDIA's GEAR Lab launches ENPIRE for autonomous real-world robot policy self-improvement using Codex agentsThe loop trained robots to autonomously install GPUs overnight.

Elon Musk proposes capturing space-based solar energy to scale AI computation beyond human capacityThe theoretical roadmap leverages resources from Tesla, SpaceX, and xAI

Cursor announces Origin, a Git-compatible, MCP-extensible code hosting platform built for high-throughput AI agentsThe hybrid NVMe and S3 architecture supports infinite replicas.

Microsoft transitions Copilot Cowork to usage-based pricing and evaluates DeepSeek V4 to cut enterprise AI costsCharles Lamanna says unlimited pricing is unsustainable for heavy users

Voice AI startup Bland raises $100 million Series C to automate complex, regulated telephony workflowsA comedic launch video featured founder Paul Graham.

Andreessen Horowitz releases a custom video tribute tracing SpaceX's trajectory from early failures to aerospace milestonesThe video opens by disclosing a16z's SpaceX shareholder status.

Cursor previews an upcoming 1.5-trillion-parameter custom LLM trained from scratch on more than 100,000 GPUsThe model targets general intelligence and launches within weeks.

xAI releases Grok Imagine Video 1.5 with improved physics simulation and faster generation speedsUsers can access the generator directly at grok.com/imagine

OpenAI resolves Codex capacity issues by resetting model rate limits over a 24-hour windowUsers previously encountered "Selected Model is at Capacity" errors

Coinbase CEO Brian Armstrong announces plans for tokenized U.S. stocks backed 1:1 by actual sharesSmart contracts will automatically distribute dividend payments to holders

Anthropic says Claude Code success rates rise from 77% to 91% as user domain expertise increasesCritics question the study's reliance on Claude self-evaluation

Gergely Orosz reports Meta is walking back harsh engineering directives, including 24/7 screen recording and reassignment to AI data labelingReassigned engineers solved LeetCode to train coding AI models

Elon Musk shares a Grok Imagine video generation demo, which investor Steve Jurvetson notes resembles a physical artwork he ownsThe clip showcases detailed rendering of fluids and light.

Thais Branco's Taste Labs emerges from stealth with $18.5 million to build data infrastructure that eliminates 'AI slopInitial applications focus on design and model evaluation.

Coinbase launches Coinbase Advisor, registering its conversational AI agent with the SEC as an investment advisorThe agent is initially available to Coinbase One subscribers.

OpenAI rolls out Codex Computer Use, Chrome integration, and Memory to EU and UK users, powered by GPT-5.3-Codex-SparkThe agent controls local Mac applications by capturing screenshots

SpaceX acquires Cursor, sparking debate over the coding assistant's strategy and its origins as a mechanical CAD toolThrive Capital backed founder Michael Truell before the pivot.

Apple plans to launch AirPods with built-in cameras in late 2027 to power visual AI featuresThe cameras are for visual intelligence, not capturing photos.

Leaked audited financials show OpenAI made $13.07 billion in revenue against $34 billion in costs in 2025SoftBank and Microsoft contributed $1.17 billion of that revenue.

DeepSeek raises $7.4 billion at a $50 billion valuation with terms that secure absolute control for founder Liang WenfengWenfeng personally contributed $3 billion of the total capital

OpenAI's Micah Carroll and Marcus J. W. introduce Deployment Simulation, achieving 92% directional accuracy in forecasting model misbehaviorIt uses historical conversation data to bypass test-environment detection

Ethan Mollick says open-source models lag closed-source by a year, but investor Ramez Naam argues the gap is only four monthsMollick warns firms have months to harden IT systems.

San Francisco media startup Nell launches with backing from Index Ventures, Soleio Cuervo, and Guillermo RauchSpecific product features and technical details remain undisclosed.

The Boring Company completes commissioning of its Prufrock-MB2 tunneling machine in Nashville after an 11 rpm testThe company plans to ship its Prufrock-MB3 machine in August

Genesis announces Eno, its first general-purpose robot, featuring a non-humanoid design slated for a Q4 releaseObservers praised its physical fluidity compared to Tesla's Optimus

Mistral AI CEO Arthur Mensch announces a new family of large but sparse models debuting this summerRestricted early access begins in July for key partners.

Snap CEO Evan Spiegel demos new AR Spectacles, drawing criticism over their bulky, IMAX-style designThe hardware features thick frames and reflective blue-tinted lenses