Digg - AI news, before it trends

Today's Highlights

In case you missed it

Pope Leo XIV releases papal encyclical on AI ethics and partners with Anthropic to promote human-centered development

DZ#145|@DENNY_ZHOU

SZ#58|@SUCHENZANG

SB#27|@SLEEPINYOURHAT

#1 LIKED

8.1k

AngelList's Naval Ravikant argues the primary AI competition is humans with AI versus humans without

KA#488|@YACINEMTB

#1 VIEWED

Merge API launches Merge Gateway to help enterprises build custom model routers based on product-specific benchmarks

CE#1598|@INTUITMACHINE

CH#1496|@KIMMONISMUS

FASTEST CLIMBING

Harvey's Legal Agent Benchmark finds frontier AI models complete less than 10% of complex legal tasks end-to-end

YP#1252|@YPATIL125

AR#1070|@AJRATNER

Top Stories

Posts:—|Clusters:—|Next Crawl:—

10

China reportedly restricts overseas travel for AI talent at Alibaba and DeepSeek to protect algorithmic secretsCritics note DeepSeek already publishes its AI models openly.

9h|281k4.7k634

ZE#1497|@ZEPHYR_Z9

RO#1220|@KRISHNANROHIT

RP#1032|@ROHANPAUL_AI

KA#488|@YACINEMTB

T(#420|@TEORTAXESTEX

FI#397|@FINBARRTIMBERS

NL#64|@NATOLAMBERT

20

MiniMax details its M3 sparse attention architecture, claiming a 15.6x decoding speedup at 1 million tokensThe design also achieves a 9.7x prefill speedup.

3h|267.8k2.4k598

ZE#1497|@ZEPHYR_Z9

CH#1496|@KIMMONISMUS

EL#716|@ELIEBAKOUCH

PM#713|@PMINERVINI

T(#420|@TEORTAXESTEX

30

OpenRouter raises $113 million at a $1.3 billion valuation as weekly traffic hits 25 trillion tokensThe funding will build multi-model infrastructure for autonomous agents.

4h|175.1k2k235

GK#1182|@GREGKAMRADT

CB#999|@CODE_STAR

DE#374|@DEEDYDAS

40

AngelList's Naval Ravikant argues the primary AI competition is humans with AI versus humans withoutResearch engineer kache questioned the future of developers who fully automate.

4h|265.3k#1LIKED8.1k889#1COMMENTS963

RISING LIKES8.1k

KA#488|@YACINEMTB

50

Merge API launches Merge Gateway to help enterprises build custom model routers based on product-specific benchmarksIt prevents enterprise cost overruns from overusing frontier models

4h|#1VIEWS3.1M1.9k#1BOOKMARKED1.2k

CE#1598|@INTUITMACHINE

CH#1496|@KIMMONISMUS

New github stars (48hrs)

self-evolving/repo192h

Invokes AI agents via GitHub mentions to implement issues, review PRs, fix branches, and persist memory across runs.

modaic-ai/gepa-viz22814h

Renders GEPA prompt-optimization candidate trees as interactive force-directed graphs with live updates from a Python callback.

manaflow-ai/cmux19.4k1d

Integrates Ghostty into a macOS terminal with vertical tabs, sidebar metadata, OSC notifications, and in-app browser for AI coding agents.

60

Nathan Lambert of the Allen Institute for AI says Google's Gemma 4 models are outpacing Alibaba's Qwen in cumulative downloadsAnalysts cite Google's Apache 2.0 license as the driver

2h|66.6k31362

FB#1153|@XEOPHON

👩‍💻P#271|@DYNAMICWEBPAIGE

NL#64|@NATOLAMBERT

70

Google DeepMind's Susan Zhang requests documented proof of AI support bots successfully resolving customer issues without human interventionAn engineer argued support loops are intentional cost-reduction tactics.

5h|28.7k37453

KA#488|@YACINEMTB

SZ#58|@SUCHENZANG

LB#55|@GIFFMANA

80

Google DeepMind's Alex Imas argues AI-generated writing alienates readers because it lacks the private information needed to justify their attentionChakrabarty notes that AI text signals skipped cognitive effort.

4h|13.4k14649

AI#1777|@ALEXOLEGIMAS

TC#1050|@TUHINCHAKR

KE#449|@ETHAYARAJH

90

Wharton's Ethan Mollick uses frontier models to find reporting errors in a WIRED article criticizing AI fact-checkingClaude 3.7 and GPT-5.5 Pro flagged the reporting inaccuracies.

4h|26.4k26634

EM#175|@EMOLLICK

BB#133|@BOAZBARAKTCS

100

FDA launches 2026 Accelerated AI Pathway Pilot to evaluate AI-generated evidence in drug submissionsNo AI-designed drug has received FDA approval yet.

10h|41.2k54980

CH#1496|@KIMMONISMUS

110

US Intelligence Bureau assessment designates AI and data center opposition as "anti-tech violent extremismThe designation could prompt FBI investigations of safety groups.

2h|9.1k14133

PR#1332|@8TEAPI

AC#517|@ANDREWCURRAN_

123

Serena Ge releases DeepSWE, a long-horizon benchmark designed to evaluate AI coding agents on complex engineering tasksIts prompts are half the length of SWE-bench Pro.

53m|18.7k161110

GT#270|@GARRYTAN

131

AI experimenter fofr finds Gemini Omni struggles with two-handed finger-counting in video evaluationThe model failed specifically on numbers seven and eight

7h|21.7k15729

FO#1996|@FOFRAI

KD#908|@CSPROFKGD

14NEW

Rishi Bommasani of Stanford HAI finds AI hiring monocultures cause systemic rejection of Black and Asian candidatesOver 60% of Fortune 100 companies use HireVue's algorithms

2h|6.5k5734

KE#449|@ETHAYARAJH

RI#223|@RISHIBOMMASANI

152

CallosumAI's Danyal Akarca announces hiring push to build heterogeneous compute infrastructure for multi-chip AI systemsThe startup is recruiting five roles based in London.

1h|33.1k11043

MC#1196|@MATTHEWCLIFFORD

AD#731|@AJDDAVISON

162

Yoav Gur Arieh finds most unfaithful chain-of-thought detectors perform near random chanceTheir BonaFide benchmark generates ground-truth faithfulness labels.

3h|5.7k7735

MG#1329|@MEGAMOR2

AM#624|@ANMARASOVIC

171

Gergely Bérczi and László M. Fehér use DeepMind's AlphaEvolve and GPT 5.5 Pro to generate new geometry conjecturesThis is the second paper to cite AlphaEvolve

8h|26.1k11554

PS#1426|@PATRICKSHAFTO

DR#56|@ROYDANROY

181

AMUSE optimizer merges Muon and schedule-free gradient evaluation to train models without learning rate decayLucas Nestler released a code implementation called SFMuon.

11h|10.9k10057

LN#1709|@CLASHLUKE

SR#604|@CLONEOFSIMO

191

DeepSeek researcher Deli Chen publishes an autonomous research agent taxonomy featuring a fictional Level 5 systemChen used an AI coding agent to update his website.

5h|15k9621

DC#1523|@VICTOR207755822

T(#420|@TEORTAXESTEX

201

AI experimenter fofr releases a joke video showcasing a parody AI model that counts the R sounds in 'strawberry' in real timeThe parody plays on a popular AI capability meme.

6h|12.3k30935

FO#1996|@FOFRAI

🍓🍓#1711|@IRULETHEWORLDMO

Rising Stories

AI Culture4H AGO
Sam Altman sends LEGO Star Wars set to pseudonymous commentator
#1262@SAMSJA19
#980@SCALING01
18110.6k
Industry Events4H AGO
Geoffrey Hinton scheduled for surprise Cohere appearance at Toronto Tech Week
#1634@1VNZH
#171@AIDANGOMEZ
10510.5k
AI Writing1H AGO
Cheaper small models will power hyper-personalized AI writing
#449@ETHAYARAJH
#70@DWARKESH_SP
20113.4k
AI Memes6H AGO
AI experimenter fofr releases a joke video showcasing a parody AI model that counts the R sounds in 'strawberry'
#1996@FOFRAI
#1711@IRULETHEWORLDMO
30712.1k
Academia2H AGO
Sarah Chieng says AI conferences have become VC deal-sourcing hubs
#1408@MILKSANDMATCHA
#223@RISHIBOMMASANI
7113.1k

Github Stars

(7 days)

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.
759
660
30
495
203
19.4k stars
self-evolving/repo2H AGO AGO
Invokes AI agents via GitHub mentions to implement issues, review PRs, fix branches, and persist memory across runs.
594
878
19 stars
modaic-ai/gepa-viz14H AGO AGO
Renders GEPA prompt-optimization candidate trees as interactive force-directed graphs with live updates from a Python callback.
878
793
228 stars
…-group/InferenceBench5D AGO AGO
Evaluates autonomous AI agents optimizing LLM serving like vLLM in open-ended scenarios with quality and integrity gates.
1224
1066
16 stars

Privacy Policy|Terms of Service|© 2026 Digg Inc.

Yesterday's Top Stories, May 25, 2026.

Posts: 2,930|Clusters: 803|Stories frozen at midnight PT

10

Pope Leo XIV releases papal encyclical on AI ethics and partners with Anthropic to promote human-centered developmentThe encyclical warns against purely profit-driven AI development.

7d|10.8M#1LIKED108.3k#1BOOKMARKED20.1k

CL#1983|@_CHENGLOU

V(#1853|@REACH_VB

SM#1811|@SAMMCALLISTER

𝚟⟢#1753|@VIEMCCOY

🍓🍓#1711|@IRULETHEWORLDMO

CE#1598|@INTUITMACHINE

NR#1562|@NATANIELRUIZG

CH#1496|@KIMMONISMUS

AM#1487|@ANDREWMAYNE

HB#1459|@HAYDNBELFIELD

SÓ#1457|@S_OHEIGEARTAIGH

AN#1409|@ARAN_NAYEBI

BI#1402|@BILALTWOVEC

PW#1359|@PETERWILDEFORD

CF#1356|@CFGEEK

KK#1218|@KEVINAKWOK

RP#1032|@ROHANPAUL_AI

🎭#878|@DEEPFATES

AD#867|@DORIALEXANDER

B(#839|@BEFFJEZOS

PT#643|@ELDER_PLINIUS

WK#609|@WILLKNIGHT

CC#548|@ITSCLIVETIME

AC#517|@ANDREWCURRAN_

SK#505|@SEBKRIER

MT#450|@TEGMARK

DW#392|@DEANWBALL

TM#359|@ISCIENCELUVR

FF#331|@FRANCOISFLEURET

KA#207|@KARINANGUYEN

@(#195|@TIMNITGEBRU

DZ#145|@DENNY_ZHOU

SZ#58|@SUCHENZANG

SB#27|@SLEEPINYOURHAT

20

xAI expands its Grok Build terminal coding assistant beta to Premium+ and SuperGrok subscribersThe tool deploys parallel sub-agents to execute tasks.

12h|#1VIEWS24.5M28.9k3.8k

“P#1702|@PAULARAMBLES

RP#1032|@ROHANPAUL_AI

B(#839|@BEFFJEZOS

BY#341|@BILLYUCHENLIN

RS#321|@SCOBLEIZER

EM#76|@ELONMUSK

30

Andrew Wilson proposes a chalkboard lecture on deriving AI foundations from probability theory for Dwarkesh Patel's podcastDan Roy nominated David Duvenaud and Pushmeet Kohli.

12h|310.8k2.2k296

🍓🍓#1711|@IRULETHEWORLDMO

TB#1556|@BINARYBITS

BG#1447|@BEN_GOLUB

PS#1371|@RECURSEPARADOX

JL#1106|@ARTIRKEL

CB#999|@CODE_STAR

B(#839|@BEFFJEZOS

KA#488|@YACINEMTB

RL#444|@RYAN_T_LOWE

NI#303|@NILOOFAR_MIRE

AG#154|@ANDREWGWILS

TC#120|@TACOCOHEN

DP#70|@DWARKESH_SP

DR#56|@ROYDANROY

40

Critics compare the Vatican's new AI encyclical Magnifica Humanitas to pre-Copernican geocentrism for denying machine cognitionThe text frames machine processes strictly as data handling.

16h|394k1.9k600

𝚟⟢#1753|@VIEMCCOY

XL#1674|@XLR8HARDER

CE#1598|@INTUITMACHINE

HB#1459|@HAYDNBELFIELD

AN#1409|@ARAN_NAYEBI

PR#1332|@8TEAPI

RO#1220|@KRISHNANROHIT

SL#1060|@SETHLAZAR

B(#839|@BEFFJEZOS

CC#548|@ITSCLIVETIME

DW#392|@DEANWBALL

50

Y Combinator's Paul Graham warns that AI-generated emails from startup founders lose all impact once detectedGraham noted a telltale journalistic tone exposes these drafts

21h|1M10.3k872

SS#1992|@STEVESI

AI#1777|@ALEXOLEGIMAS

NR#1562|@NATANIELRUIZG

RM#1327|@RACHELMETZ

MS#1323|@MICSOLANA

KS#496|@KENNETH0STANLEY

GM#153|@GARYMARCUS

60

Aidan Clark argues AGI pretraining is now restricted to six industry labs with massive compute resourcesWill Depue says private compute will dictate scientific progress

6d|136.3k1.5k207

T(#420|@TEORTAXESTEX

XE#417|@XWANG_LK

PN#348|@PREETUMNAKKIRAN

NI#303|@NILOOFAR_MIRE

T🪽#259|@TEKNIUM

WD#254|@WILLDEPUE

NB#242|@NATHANBENAICH

MB#199|@MMBRONSTEIN

BN#107|@BNEYSHABUR

RA#83|@_AROHAN_

NL#64|@NATOLAMBERT

SZ#58|@SUCHENZANG

70

Japanese Prime Minister Sanae Takaichi proposes procurement and funding boosts for domestic deep tech and sovereign AI startupsSakana AI proposed specialized models for Japanese defense autonomy.

22h|761.6k12.9k597

T🐻#1476|@TKASASAGI

LJ#1377|@YESTHISISLION

HA#18|@HARDMARU

80

Every co-founder Dan Shipper argues AI progress will paradoxically create more human work rather than a job apocalypseShipper forecasts companies will adopt single, organization-wide Slack super-agents.

1d|655.1k2.2k3.3k

PS#1593|@SKIRANO

DS#1410|@DANSHIPPER

LR#1176|@LENNYSAN

90

Google DeepMind CEO Demis Hassabis says solving mathematical benchmarks like Erdős problems does not constitute true AGIDeepMind's AI systems recently solved eight Erdős problems.

1d|434.8k3.1k846

PM#713|@PMINERVINI

SR#604|@CLONEOFSIMO

FF#331|@FRANCOISFLEURET

GM#153|@GARYMARCUS

100

Goldman Sachs CEO argues AI will not eliminate 25% of jobs, calling apocalypse fears exaggeratedHe compared the transition to historical shifts like microfiche.

15h|888.1k9k1.5k

RP#1032|@ROHANPAUL_AI

B(#839|@BEFFJEZOS

GM#153|@GARYMARCUS

110

OpenAI's roon shares OpenAI Five's 2018 architecture flowchart to counter claims the Dota 2 AI design was kept secretThe shared 2018 diagram details a 1,024-unit LSTM network.

15h|207.7k2.9k287

LA#980|@SCALING01

DA#927|@GROWING_DANIEL

B(#839|@BEFFJEZOS

AE#719|@ADRIENLE

KA#488|@YACINEMTB

120

xAI removes the distinction between researcher and engineer titles, unifying all technical staff under the single designation of engineer across the companyElon Musk called the old system a two-tier academic convention.

2d|233.7k3.3k739

PD#654|@PMDDOMINGOS

KC#489|@KYLECRANMER

GM#153|@GARYMARCUS

130

Google's Logan Kilpatrick announces a free, no-code tool in AI Studio for building native Android apps using GeminiUsers generated over 250,000 apps since last week

11h|289.1k3.9k1.6k

SM#530|@SWAROOPRM7

LK#90|@OFFICIALLOGANK

LB#55|@GIFFMANA

140

AngelList co-founder Naval Ravikant jokes that Anthropic does not have Claude, but rather Claude has AnthropicThe quip suggests the flagship model now drives the company.

20h|1.1M17.3k915

T(#420|@TEORTAXESTEX

150

Gary Marcus disputes AI czar David Sacks' warning, arguing private tech companies already censor and surveil the publicLovely noted the contradiction in Sacks' anti-regulation warnings.

14h|266.4k3.7k390

GL#1984|@GARRISONLOVELY

GM#153|@GARYMARCUS

C🤗#68|@CLEMENTDELANGUE

160

David Heinemeier Hansson claims GPT-5.5 outperforms every coding model since Opus 4.5, requiring only high-level steeringBeff Jezos expressed skepticism about competitor model Opus 4.7.

14h|296.8k4.9k668

B(#839|@BEFFJEZOS

RH#804|@ROMAINHUET

170

Engineer Paula jokes that God was the original prompt engineer, drawing technical parody replies on XDanielle Fong joked humanity lacks instructions to avoid mistakes.

12h|183.6k4k209

“P#1702|@PAULARAMBLES

AV#1245|@ALTRYNE

XM#437|@INFOXIAO

JJ#244|@JOANNEJANG

180

Google Omni AI Edits Videos With Dynamic Object Additions And Effects

21h|1.1M13.8k6.3k

190

Anthropic co-founder Dario Amodei warns AI could eliminate 50% of entry-level white-collar jobs within five yearsOpenAI instead frames AI as a productivity tool.

20h|178.5k1.6k369

NI#1902|@NATHANPMYOUNG

CH#1496|@KIMMONISMUS

B(#839|@BEFFJEZOS

AC#517|@ANDREWCURRAN_

200

Jon Stokes argues that AI models emulating functional emotions will drive a psychological shift toward modern animismQiaochu Yuan warns children are forming deep attachments to AI.

9h|43.2k417180

SD#1776|@INDUCTIONHEADS

XL#1674|@XLR8HARDER

AN#1409|@ARAN_NAYEBI

DW#392|@DEANWBALL

210

Google DeepMind's Lean-powered AI agent solves 56-year-old open Erdős math problems for a few hundred dollars eachThe system also proved 44 open conjectures from OEIS.

2d|73.4k85385

NS#1981|@SO8RES

PW#1359|@PETERWILDEFORD

RH#887|@ROBINHANSON

D🎇#458|@DAVIDAD

CS#43|@CHRSZEGEDY

220

Uber COO Andrew Macdonald says rising AI token costs are hard to justify after exhausting its 2026 Claude Code budgetUber is slowing down hiring to offset the ballooning costs

7h|451.4k6.7k487

HP#455|@HYHIEU226

GM#153|@GARYMARCUS

230

Extropic AI founder Guillaume Verdon shares a parody meme satirizing the highly competitive hiring market for top AI talentThe parody depicts divine figures joining Anthropic for pay.

13h|318.4k3.7k232

B(#839|@BEFFJEZOS

240

Elon Musk says xAI finished training its 1.5-trillion-parameter Grok V9-Medium model using Cursor data for supplementary trainingPublic release is scheduled within two to three weeks.

1d|42.2k70129

BR#1622|@BINDUREDDY

B(#839|@BEFFJEZOS

T(#420|@TEORTAXESTEX

BY#341|@BILLYUCHENLIN

T🪽#259|@TEKNIUM

EM#167|@EMOSTAQUE

250

Paul Graham Sequences Human Genome to 30x Coverage Entirely at Home

21h|312.5k4.2k2.3k

260

Swarat Chaudhuri releases AlphaProof Nexus, an LLM framework that solved nine open Erdős problems using Lean compiler feedbackThe agent also solved 44 OEIS mathematical problems.

15h|135.7k1.4k399

PK#298|@PUSHMEET

DR#56|@ROYDANROY

270

Google's Logan Kilpatrick and tech founders argue AI demand is limitless, while creator roon predicts a single utility-like monopolySuhail says ChatGPT and Claude Code won't dominate alone.

1d|175.3k867129

GT#270|@GARRYTAN

280

Microsoft Research releases SkillOpt, an optimization method that treats AI agent skills as trainable external states of a frozen modelA textual learning rate governs validation-gated skill rewrites.

15h|81.8k733940

EL#475|@OMARSAR0

GT#270|@GARRYTAN

290

Huawei unveils its LogicFolding architecture to bypass EUV lithography limits and target 1.4nm-equivalent chips by 2031LogicFolding Kirin smartphone chips will ship this year.

1d|79.3k697176

CE#1598|@INTUITMACHINE

CH#1496|@KIMMONISMUS

T(#420|@TEORTAXESTEX

300

Viral AI-generated parody of The Office depicts Andrej Karpathy's fictional onboarding day at AnthropicAn intern generated the zero-cost video using Kling 3.0.

1d|97.3k894634

RA#83|@_AROHAN_