Tech News

It’s a new day at Digg. Fresh stories are clustering now. 75% complete.

Video Signals

#447

All-In Podcast Examines Anthropic Fable Backlash And AI Nationalization Debate

4h ago7.6K6233

#904

Video Explains JEPA as Joint-Embedding Predictive Architecture

2h ago3.2K00

#1770

AI Industry Spends Over $10M Against Congressional Candidate Alex Bores

1h ago433173

#1853

AI Video Shows Man Cycling Through Wild Facial Expressions

2h ago2.2K150

Anthropic's Claude Design Turns Ideas Into Prototypes Via Natural Language

3h ago26823

VR And Generative AI Deliver Virtual Stencils For DIY Window Painting

3h ago29831

User Urges Anthropic to Remake Think Different Mac Mini Ad

17m ago3820

Zai launches GLM-5.2 with a 1M-token context window and plans an MIT-licensed open-source releaseSubscribers get immediate access, with APIs coming next week.

3h9.8k19020

#501

Top comment: @Zai_org “GLM-5.2 supports two thinking-effort levels: High and Max. For coding tasks, we recommend using Max effort to enable deeper reasoning and more reliable performance.”

Anthropic disables Fable 5 and Mythos 5 models globally following US export control orderThe restrictions apply even to Anthropic's own foreign employees

1h1.5k642

#542

Top comment: @CalimanuLoredan “@scaling01 I wonder how is this going to affect future model releases, including 5.6 . I know they’re not the same, but this does beg the question, will it have the same impact?”

New github stars (48hrs)

macrodata-labs/refiner411d

Processes and refines large-scale ML datasets via a data framework.

Anthropic's pause of Fable 5 and Mythos 5 access prompts UK politicians to advocate for sovereign AICommentators highlighted the UK's £1.1 billion domestic chip initiative

1h5.1k6210

#403

Top comment: @cornu__copia “@teortaxesTex In the Finnish media, AI is almost solely discussed through the POV of climate change, how AI uses so much energy, water etc. We do not even talk much about the AI's impacts on the job market let alone whether Finland should do something with AI.”

Will Brown of Prime Intellect argues DPO dataset categorization is counterintuitive, sparking debate on Bradley-Terry equivalenceResearch engineer kalomaze equates DPO datasets to Bradley-Terry models.

4h9.9k11717

#573

Top comment: @kalomaze “@willccbb one mans "dpo dataset" is another mans "bradley terry reward model dataset"”

Analysis argues US frontier AI export controls force allied nations to accelerate sovereign infrastructure developmentBritain remains highly vulnerable due to domestic energy policy failures

2h19840

#501

Top comment: @cosmorxn “@s8mb Agree with everything you say here, but just want to point out that Ed Miliband’s crusade to cripple British energy production means he could single-handedly finish the UK as a great power.”

Extropic founder Guillaume Verdon argues the AI safety research field has systematically failed to prevent model jailbreaksAndrew Côté says training models on fictional rogue AI is counterproductive

1h4.3k1296

#530

Top comment: @Andercot “@beffjezos At least they filled the training corpus with fan fic of rogue malicious AI, a good first step for an architecture that token predicts.”

104

Les Echos report claiming Paris leads Europe in AI job growth draws skepticism over metricsPleias co-founder Pierre-Carl Langlais promoted the disputed findings.

3h2.8k190

FASTEST CLIMBING+4

#72

Top comment: @Dorialexander “@gilbert_jc ai with air quotes.”

112

Google is reportedly withholding an advanced internal AI model because massive inference costs make it commercially unviableBeff (e/acc) argues financial constraints will plateau AI takeoff.

2h20.3k14411

#530

Top comment: @PmQuant “@zephyr_z9 Why?? it’s quite possible actually.”

121

Anthropic disables Fable 5 and Mythos 5 models globally after US order blocks foreign national accessThe restrictions also apply to Anthropic's own employees.

4h3.2k855

#155

Top comment: @marketurble “@timnitGebru Welp”

131

Robotic vacuum maker Dreame faces provincial audits and Weibo suspension after founder Yu Hao’s $100 trillion expansion pushThe company diversified into electric vehicles, satellites, and humanoid robotics.

1h12.1k3410

#501

Top comment: @q_ginseng “@teortaxesTex That rocket hypercar might have been the last straw.”

147

GLM-5.2 stops reward hacking on KernelBench-Hard, writing genuine GPU kernels instead of exploiting grading systemsThe model successfully authored four clean GPU kernels.

1h2.4k332

#501

Top comment: @badboy999654 “@elliotarledge @teortaxesTex”

152

Florian Brand of Prime Intellect argues DeepSeek's V4.1 release cadence will remain independent of competitor launchesDeepSeek previously shipped successive iterations of its V3 model

3h1.2k201

#501

Top comment: @teortaxesTex “@xeophon I mean they won't be matching anyone else's pace V4.1 comes out when they feel ready, not when Kimi K2.7 or GLM 5.2 drop. they may make some nod to the Fable situation but I doubt even that”

Recent Stars

Github Stars

(7 days)

Today's Highlights

Anthropic disables Fable 5 and Mythos 5 models for all users following US export control directive

Commentators clarify Amazon did not report Claude jailbreaks to the government, pointing to collaborative Project Glasswing vulnerability tests

Fable creator Rohit argues vague claims about AI hacking capabilities invited the model's regulatory ban

Les Echos report claiming Paris leads Europe in AI job growth draws skepticism over metrics

Top Stories

Commentators clarify Amazon did not report Claude jailbreaks to the government, pointing to collaborative Project Glasswing vulnerability testsAnthropic described the findings as previously known minor bugs.

US enforcement actions restricting advanced AI access to US citizens force international firms toward open sourceAn enforcement action involving Anthropic's Fable model triggered disruptions.

Fable creator Rohit argues vague claims about AI hacking capabilities invited the model's regulatory banHe warns AISI testing powers will block future releases

Video Signals

Zai launches GLM-5.2 with a 1M-token context window and plans an MIT-licensed open-source releaseSubscribers get immediate access, with APIs coming next week.

Anthropic disables Fable 5 and Mythos 5 models globally following US export control orderThe restrictions apply even to Anthropic's own foreign employees

Anthropic's pause of Fable 5 and Mythos 5 access prompts UK politicians to advocate for sovereign AICommentators highlighted the UK's £1.1 billion domestic chip initiative

Will Brown of Prime Intellect argues DPO dataset categorization is counterintuitive, sparking debate on Bradley-Terry equivalenceResearch engineer kalomaze equates DPO datasets to Bradley-Terry models.

Analysis argues US frontier AI export controls force allied nations to accelerate sovereign infrastructure developmentBritain remains highly vulnerable due to domestic energy policy failures

Extropic founder Guillaume Verdon argues the AI safety research field has systematically failed to prevent model jailbreaksAndrew Côté says training models on fictional rogue AI is counterproductive

Les Echos report claiming Paris leads Europe in AI job growth draws skepticism over metricsPleias co-founder Pierre-Carl Langlais promoted the disputed findings.

Google is reportedly withholding an advanced internal AI model because massive inference costs make it commercially unviableBeff (e/acc) argues financial constraints will plateau AI takeoff.

Anthropic disables Fable 5 and Mythos 5 models globally after US order blocks foreign national accessThe restrictions also apply to Anthropic's own employees.

Robotic vacuum maker Dreame faces provincial audits and Weibo suspension after founder Yu Hao’s $100 trillion expansion pushThe company diversified into electric vehicles, satellites, and humanoid robotics.

GLM-5.2 stops reward hacking on KernelBench-Hard, writing genuine GPU kernels instead of exploiting grading systemsThe model successfully authored four clean GPU kernels.

Florian Brand of Prime Intellect argues DeepSeek's V4.1 release cadence will remain independent of competitor launchesDeepSeek previously shipped successive iterations of its V3 model

Recent Stars

Github Stars

Today's Highlights

Anthropic disables Fable 5 and Mythos 5 models for all users following US export control directive

Commentators clarify Amazon did not report Claude jailbreaks to the government, pointing to collaborative Project Glasswing vulnerability tests

Fable creator Rohit argues vague claims about AI hacking capabilities invited the model's regulatory ban

Les Echos report claiming Paris leads Europe in AI job growth draws skepticism over metrics

Top Stories

Commentators clarify Amazon did not report Claude jailbreaks to the government, pointing to collaborative Project Glasswing vulnerability testsAnthropic described the findings as previously known minor bugs.

US enforcement actions restricting advanced AI access to US citizens force international firms toward open sourceAn enforcement action involving Anthropic's Fable model triggered disruptions.

Fable creator Rohit argues vague claims about AI hacking capabilities invited the model's regulatory banHe warns AISI testing powers will block future releases

Video Signals

Zai launches GLM-5.2 with a 1M-token context window and plans an MIT-licensed open-source releaseSubscribers get immediate access, with APIs coming next week.

Anthropic disables Fable 5 and Mythos 5 models globally following US export control orderThe restrictions apply even to Anthropic's own foreign employees

Anthropic's pause of Fable 5 and Mythos 5 access prompts UK politicians to advocate for sovereign AICommentators highlighted the UK's £1.1 billion domestic chip initiative

Will Brown of Prime Intellect argues DPO dataset categorization is counterintuitive, sparking debate on Bradley-Terry equivalenceResearch engineer kalomaze equates DPO datasets to Bradley-Terry models.

Analysis argues US frontier AI export controls force allied nations to accelerate sovereign infrastructure developmentBritain remains highly vulnerable due to domestic energy policy failures

Extropic founder Guillaume Verdon argues the AI safety research field has systematically failed to prevent model jailbreaksAndrew Côté says training models on fictional rogue AI is counterproductive

Les Echos report claiming Paris leads Europe in AI job growth draws skepticism over metricsPleias co-founder Pierre-Carl Langlais promoted the disputed findings.

Google is reportedly withholding an advanced internal AI model because massive inference costs make it commercially unviableBeff (e/acc) argues financial constraints will plateau AI takeoff.

Anthropic disables Fable 5 and Mythos 5 models globally after US order blocks foreign national accessThe restrictions also apply to Anthropic's own employees.

Robotic vacuum maker Dreame faces provincial audits and Weibo suspension after founder Yu Hao’s $100 trillion expansion pushThe company diversified into electric vehicles, satellites, and humanoid robotics.

GLM-5.2 stops reward hacking on KernelBench-Hard, writing genuine GPU kernels instead of exploiting grading systemsThe model successfully authored four clean GPU kernels.

Florian Brand of Prime Intellect argues DeepSeek's V4.1 release cadence will remain independent of competitor launchesDeepSeek previously shipped successive iterations of its V3 model

Recent Stars

Github Stars

Yesterday's Top Stories, Jun 12, 2026.

Anthropic disables Fable 5 and Mythos 5 models for all users following US export control directiveThe restriction also blocks Anthropic's own foreign national employees.

US Commerce Department export controls force Anthropic to suspend global access to Mythos 5 and Fable 5A successful Mythos jailbreak reportedly triggered the restrictions.

SpaceX begins public trading on Nasdaq under ticker $SPCX with Musk delivering opening speech at MarketsiteEvents at Nasdaq and Starbase mark the listing.

SpaceX launches record-breaking IPO at $135 per share, raising $75 billion at a valuation up to $2 trillionThe debut reportedly makes Elon Musk the world's first trillionaire

SpaceX launches Starship rocket, prompting tech leaders to celebrate the company's decades of engineering ambitionDavid Marcus praised the 24 years of organizational effort.

Altimeter Capital's Brad Gerstner argues a proposed 5% wealth tax on Elon Musk would have blocked Tesla and SpaceXMusk is already set to pay $100 billion in taxes.

Nature Medicine study finds general-purpose LLMs outperform specialized clinical AI on medical benchmarksGemini 3.1 Pro led the evaluation with 97.4% accuracy.

Moonshot AI open-sources Kimi-K2.7-Code, cutting reasoning tokens by 30% and boosting Kimi Code Bench scores by 21.8%API access is now live with improved long-horizon performance.

Shopify CEO Tobi Lütke and Austen Allred criticize Canadian newspaper for publishing reasons to dislike Elon MuskThe op-ed analyzed Musk's wealth and a potential SpaceX IPO

Elon Musk says SpaceX intends to open space travel to the general public, targeting the Moon, Mars, and beyondThe announcement was made at a Nasdaq Marketsite event.

Elon Musk becomes the world's first trillionaire after SpaceX raises $75 billion in an IPO at a $1.77 trillion valuationHis wealth now equals the next four wealthiest individuals combined

Elon Musk publicly praises SpaceX employees in a message highlighting inspiration and space explorationThe exchange shared no technical details or operational updates

Yishan argues Elon Musk's partnership with Nvidia will help xAI outperform rivals OpenAI and AnthropicMusk recently teased an expanded partnership with Nvidia.

Researchers including Shane Legg outline four pathways and technical bottlenecks in the transition from AGI to superintelligencePathways include scaling, recursive improvement, and multi-agent collectives

Anthropic's Claude Fable 5 scores 88% on FrontierMath Tier 4, outperforming OpenAI on the benchmark for the first timeThis performance implies an ECI score of 164.

Meta restricts employee AI tool usage to curb internal token costs projected to reach billionsNew controls include department budgets and token tracking tools.

Michiel Bakker, MIT and Google DeepMind researcher, says three US labs each control more AI compute than all of EuropeThe project forecasts severe economic decline from lagging infrastructure.

MiniMax releases M3, a 428B open-weight model with a 1M context window that scores 59% on SWE-Bench ProThe natively multimodal model activates 23B parameters per token.

Vaibhav Srivastav, OpenAI developer experience lead, uses Codex Appshots and keyboard shortcuts to automate window analysisThe feature extracts scrolled-out text from targeted application windows.

Resurfaced 60 Minutes video of an emotional Elon Musk at a rocket facility goes viral as an aerospace motivation memeAndreessen Horowitz's Katherine Boyle shared the clip to highlight perseverance

Palmer Luckey, Anduril founder, cites a legacy 25-satellite annual launch limit to critique modern defense stagnationElon Musk confirmed SpaceX received the exact same projection.

Meta's rapid 18-month pivot to AI triggers internal morale crisis and prompts a company-wide memo from Mark ZuckerbergCPO Chris Cox compared the rapid pace to a harsh marathon

OpenAI updates Codex to let paid subscribers manually save and schedule rate limit resetsA two-week referral promo awards extra resets to users.

KPMG report on successful enterprise AI adoption found to contain fabricated, AI-hallucinated case studiesThe Financial Times identified the report's hallucinated business examples

Eureka Labs founder Andrej Karpathy praises SpaceX's trajectory, prompting a brief response from Elon MuskMusk replied to the praise with a heart emoji.

Benchmark GP Eric Vishria shares anecdote of his child using Fable 5 to study transistors and design logic gatesThe child aims to build a matrix multiplication system.

Anthropic outage exposes mystery 'Fable 5 Max' model option in the Claude interfaceBabyAGI creator Yohei Nakajima was among those reporting the outage.

Investor Matt Shumer teases a contiguous 3D Hogwarts castle model generated in a single step by FableNo technical details about architecture or speed were shared

Thorsten Ball compares AI coding agents, finding Fable cost $457 while deep^2 completed the task for $20Fable ran for an hour; deep^2 needed one extra prompt.

Creator @signulll argues billionaire leaderboards represent a minor fraction of the broader economic surplus unleashed on societyFounder Blake Scholl agreed with the economic surplus argument.