AI News

Today's Highlights

In case you missed it

Microsoft AI announces MAI-Thinking-1 and MAI-Code-1-Flash, claiming they outperform Claude on coding and reasoning benchmarks

LB#55|@GIFFMANA

JH#40|@JEREMYPHOWARD

SR#23|@SRUSH_NLP

MB#20|@MILES_BRUNDAGE

It’s a new day at Digg. Fresh stories are clustering now. 65% complete.

Top Stories

Posts:—|Clusters:—|Next Crawl:—

10

Google releases Gemma 4 12B, a unified open multimodal model that outperforms the larger Gemma 3 27BIt powers a new macOS app for local audio transcription.

41m|170.7k3.6k1.1k

XL#1674|@XLR8HARDER

WH#1430|@NREHIEW_

VC#1208|@VITALIYCHILEY

PS#934|@_PHILSCHMID

ME#675|@MERVENOYANN

AJ#614|@ARMANDJOULIN

OS#490|@OSANSEVIERO

TM#359|@ISCIENCELUVR

20

Town exits beta and raises $55 million Series A led by a16z for its administrative AI assistantThe assistant requires explicit user approval before executing workflows

2h|247.1k882483

ET#1253|@ERIKTORENBERG

JM#931|@VENTURETWINS

BE#595|@BHOROWITZ

RS#321|@SCOBLEIZER

SG#219|@SARANORMOUS

30

Google releases Gemma 4 12B and 26B, which beat the larger Gemma 3 27B on GPQA and coding benchmarksThe 12B variant features a novel, memory-efficient architecture.

42m|11.2k14116

VC#1208|@VITALIYCHILEY

LA#980|@SCALING01

EL#716|@ELIEBAKOUCH

T(#420|@TEORTAXESTEX

Watchlist

Video Signals

JM#931|@VENTURETWINS

KD#908|@CSPROFKGD

SR#604|@CLONEOFSIMO

MA#470|@MARTIN_CASADO

MN#467|@MO_NOROUZI

AC#263|@ANDREW_N_CARR

Ideogram Releases 4.0 Open Image Model With Downloadable Weights

1h ago|Views 94KLikes 686Bookmarks 391

BE#595|@BHOROWITZ

BP#357|@BORISMPOWER

RS#321|@SCOBLEIZER

SG#219|@SARANORMOUS

TownAI Launches Out Of Beta With $55M A16z-Led Series A

2h ago|Views 136KLikes 306Bookmarks 162

PM#1652|@PACKYM

PR#1332|@8TEAPI

Ramp Launches Stack AI Operating System for Accounting Firms

4h ago|Views 286KLikes 948Bookmarks 550

OS#490|@OSANSEVIERO

Google Unveils Gemma 4 12B Multimodal Model With Unified Architecture

51m ago|Views 6.6KLikes 183Bookmarks 53

Playabl.ai Hits 1M Plays With AI-Powered TikTok For User Games

3h ago|Views 53KLikes 496Bookmarks 302

MM|@MERICAMEMED

Viral Video Mocks Generic AI Business Voiceover Scripts

20h ago|Views 160KLikes 2.5KBookmarks 832

RS#321|@SCOBLEIZER

Lassie Raises $47M Led By A16z To Automate Doctor Offices With AI

2h ago|Views 91KLikes 224Bookmarks 118

PO#1862|@PATRICK_OSHAG

Uber CEO Dara Khosrowshahi Discusses AI Spending, Autonomous Vehicles, and Super App Plans

5h ago|Views 115KLikes 325Bookmarks 286

RS#321|@SCOBLEIZER

Monako Glass Launches As World's First Wearable Linux Computer

2h ago|Views 42KLikes 449Bookmarks 303

SD#1776|@INDUCTIONHEADS

Perplexity Brings Personal Computer AI Agent to Windows

1h ago|Views 16KLikes 321Bookmarks 59

40

Ashok Elluswamy, Tesla's VP of AI Software, announces the launch of unsupervised Robotaxi service in Austin, TexasVehicles operate without human oversight inside the geofenced territory.

56m|77.5k2.2k54

AE#1489|@AELLUSWAMY

RS#321|@SCOBLEIZER

50

OpenAI's roon argues faster software and UIs are required to prevent Amdahl's law from bottlenecking rapid AI modelsGarry Tan asks if this will drive Rust adoption.

7h|46.1k48994

GT#270|@GARRYTAN

New github stars (48hrs)

Reinforcement learning system for in-hand cube reorientation on the WUJI Hand with full sim-to-real deployment

wuji-technology/wuji-mjlab9721h

Trains PPO policies for full-SO(3) in-hand cube reorientation on WUJI Hand1.0 in mjlab with GPU-batched MuJoCo physics and supplies a sim-to-real deployment bridge.

Omnimodal world models unifying reasoning and multimodal generation for physical AI

NVIDIA/cosmos8.6k1d

Hosts omnimodal Mixture-of-Transformers models for joint text, vision, audio, and action generation and reasoning in Physical AI.

60

Meta launches Meta Business Agent globally on WhatsApp and Instagram to automate customer service, scheduling, and salesThe rollout follows a two-year pilot in select countries.

2h|15.5k12916

KR#1046|@KYLIEBYTES

AC#517|@ANDREWCURRAN_

70

TownAI exits beta, launching its Townie AI assistant alongside a $55 million Series A round led by a16zUsers can delegate tasks by CC'ing the assistant on emails

2h|7.8k649

“P#1702|@PAULARAMBLES

JM#931|@VENTURETWINS

BP#357|@BORISMPOWER

80

AI commentator Andrew Curran speculates that "Mercury-Alpha" is GPT-5.6, sparking unverified release rumorsNo model developers have confirmed the rumored project.

2h|18.8k26714

AC#517|@ANDREWCURRAN_

90

CMU's Jing Yu Koh releases MACU, a multi-agent computer use framework that coordinates sub-agents via dynamic dependency graphsIt runs up to 1.5x faster on long-horizon tasks.

2h|6.8k5013

JY#921|@KOHJINGYU

RS#38|@RSALAKHU

102

Reports claim Chinese AI startup DeepSeek has raised $7 billion in new fundingAI safety researcher Miles Brundage analyzed the reported round.

38m|11k407

ET#1253|@ERIKTORENBERG

MB#20|@MILES_BRUNDAGE

110

Stanford's Vishakh Padmakumar introduces Offloading Score, a metric to quantify how much cognitive effort users delegate to AIThe metric successfully tracked increased AI reliance under tight deadlines.

25m|386305

TC#1050|@TUHINCHAKR

DY#134|@DIYI_YANG

122

Sophia Yang, Head of Developer Relations at Mistral AI, wins GTM10 Award for developer ecosystem partnershipsShe was featured on a Times Square Nasdaq billboard.

2h|2.5k692

SY#1642|@SOPHIAMYANG

FB#1153|@XEOPHON

130

TU Darmstadt's Georgia Chalvatzaki argues physical contact makes robot action spaces complex, while Jon Barron defends end-to-end image-to-action learningBarron argues image-to-action pipelines bypass intermediate 3D reconstructions.

1h|95880

GC#1888|@GEORGIACHAL

JB#261|@JON_BARRON

Github Stars

(7 days)

…technology/wuji-mjlab21H AGO
Trains PPO policies for full-SO(3) in-hand cube reorientation on WUJI Hand1.0 in mjlab with GPU-batched MuJoCo physics and supplies a sim-to-real deployment bridge.
1533
763
97 stars
NVIDIA/cosmos1D AGO
Hosts omnimodal Mixture-of-Transformers models for joint text, vision, audio, and action generation and reasoning in Physical AI.
1449
263
8.6k stars
…le-gemma/gemma-skills2D AGO
Supplies skills for Gemma model and agent interactions including a dev skill for app building.
934
490
136 stars
Yifei-Zuo/Parallax3D AGO
Implements parameterized local linear attention mechanisms for efficient language modeling.
263
238
46 stars
run-llama/liteparse4D AGO
Extracts spatial text with bounding boxes from PDFs and office docs via PDFium plus optional OCR, outputting JSON or layout-preserved text.
263
878
7.9k stars

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

Yesterday's Top Stories, Jun 2, 2026.

Posts: 4,920|Clusters: 1,100|Stories frozen at midnight PT

10

Microsoft AI announces MAI-Thinking-1 and MAI-Code-1-Flash, claiming they outperform Claude on coding and reasoning benchmarksMAI-Code-1-Flash will integrate exclusively inside GitHub Copilot.

12h|971k9.6k3.6k

SD#1776|@INDUCTIONHEADS

AV#1713|@AVISCHIFFMANN

SI#1684|@SIMEON_CPS

XL#1674|@XLR8HARDER

ST#1629|@STOCHASTICCHASM

BR#1622|@BINDUREDDY

CH#1496|@KIMMONISMUS

CR#1444|@CWOLFERESEARCH

WH#1430|@NREHIEW_

AV#1245|@ALTRYNE

MN#1179|@MENHGUIN

BZ#1158|@BANGHUAZ

FB#1153|@XEOPHON

RP#1032|@ROHANPAUL_AI

CB#999|@CODE_STAR

LA#980|@SCALING01

B(#839|@BEFFJEZOS

AN#800|@ML_ANGELOPOULOS

EL#716|@ELIEBAKOUCH

KL#620|@KYLELOSTAT

RS#612|@ZIV_RAVID

KA#488|@YACINEMTB

T(#420|@TEORTAXESTEX

XE#417|@XWANG_LK

FI#397|@FINBARRTIMBERS

🇺🇦D#387|@DBAHDANAU

TM#359|@ISCIENCELUVR

PW#288|@PANGWEIKOH

HH#276|@HANNAHAJISHIRZI

AC#263|@ANDREW_N_CARR

EM#167|@EMOSTAQUE

MS#163|@MUSTAFASULEYMAN

OK#160|@LATEINTERACTION

AK#57|@ARANKOMATSUZAKI

LB#55|@GIFFMANA

JH#40|@JEREMYPHOWARD

SR#23|@SRUSH_NLP

MB#20|@MILES_BRUNDAGE

20

OpenAI launches Sites to turn Codex plain-text instructions into hosted, interactive web applicationsThe feature is rolling out to ChatGPT Enterprise plans.

1d|#1VIEWS7.1M#1LIKED27.2k#1BOOKMARKED12k

V(#1853|@REACH_VB

CH#1496|@KIMMONISMUS

RP#1032|@ROHANPAUL_AI

SH#1026|@STEVENHEIDEL

RH#804|@ROMAINHUET

SW#801|@SHERWINWU

TI#768|@THSOTTIAUX

SR#604|@CLONEOFSIMO

BP#357|@BORISMPOWER

T🪽#259|@TEKNIUM

30

President Trump signs executive order establishing NSA-led benchmarking and a voluntary 30-day pre-release review for frontier AI modelsThe window was reduced from an earlier 90-day proposal.

15h|921.4k7.1k1.2k

SS#1992|@STEVESI

N🔎#1902|@NATHANPMYOUNG

AM#1735|@ANDYMASLEY

BG#1614|@ALTCAP

SH#1488|@HAMANDCHEESE

SÓ#1457|@S_OHEIGEARTAIGH

PW#1359|@PETERWILDEFORD

GK#1182|@GREGKAMRADT

KR#1046|@KYLIEBYTES

DK#847|@DKAUSHIK96

JK#586|@JASONKWON

AC#517|@ANDREWCURRAN_

DW#392|@DEANWBALL

GM#153|@GARYMARCUS

MB#20|@MILES_BRUNDAGE

40

Ai2 post-training lead Nathan Lambert, who led development of OLMo and Tulu, is leaving the research instituteHe plans to continue working on open-source AI models.

16h|167.1k1.9k134

V(#1853|@REACH_VB

SG#1498|@SHARONGOLDMAN

SA#1262|@SAMSJA19

MJ#1194|@MIMANSAJ

GS#1055|@G_K_SWAMY

CB#999|@CODE_STAR

KG#730|@KEERTHANPG

EL#716|@ELIEBAKOUCH

ME#675|@MERVENOYANN

JP#672|@ZHANSHENG

KL#620|@KYLELOSTAT

VS#520|@SANHESTPASMOI

AC#517|@ANDREWCURRAN_

XM#437|@INFOXIAO

YJ#427|@YUCHENJ_UW

FI#397|@FINBARRTIMBERS

WB#339|@WILLCCBB

AA#334|@AKARIASAI

HH#276|@HANNAHAJISHIRZI

AC#263|@ANDREW_N_CARR

RL#137|@SAVVYRL

NL#64|@NATOLAMBERT

LB#55|@GIFFMANA

50

Factory launches Factory Router, an automated model routing system designed to cut AI inference costs by 25%The system dynamically routes tasks to smaller, cheaper models.

13h|2.2M5.6k2.9k

ND#1894|@NICKADOBOS

MB#1759|@MATTHEWBERMAN

CE#1598|@INTUITMACHINE

SM#1333|@SHAUNMMAGUIRE

RP#1032|@ROHANPAUL_AI

JL#677|@JERRYJLIU0

GT#270|@GARRYTAN

C🤗#68|@CLEMENTDELANGUE

60

Black Forest Labs names filmmaker Martin Scorsese as an official advisor to shape its FLUX image generation modelsScorsese has advised the startup since its founding.

17h|2M4.3k2.1k

BS#1909|@BILAWALSIDHU

NR#1562|@NATANIELRUIZG

BI#1402|@BILALTWOVEC

ET#1253|@ERIKTORENBERG

KD#908|@CSPROFKGD

RR#779|@ROBROMBACH

AM#527|@ANJNEYMIDHA

FI#397|@FINBARRTIMBERS

TM#359|@ISCIENCELUVR

70

Microsoft Build slide plotting unreleased "Opus 4.6" and "Mythos" models sparks debate over leaked 2026 compute scalesAnalysts estimate the Mythos model utilized 6.1x10^27 FLOPs.

13h|1M3.4k940

BH#1941|@BENHYLAK

XL#1674|@XLR8HARDER

ZE#1497|@ZEPHYR_Z9

CH#1496|@KIMMONISMUS

AV#1245|@ALTRYNE

SS#1073|@SYTELUS

LA#980|@SCALING01

EL#716|@ELIEBAKOUCH

AC#517|@ANDREWCURRAN_

D🎇#458|@DAVIDAD

T(#420|@TEORTAXESTEX

TM#359|@ISCIENCELUVR

WD#254|@WILLDEPUE

LB#55|@GIFFMANA

80

Bill Lennon and Christy Shannon launch Ollie, an AI family assistant that coordinates schedules and household tasks via group chatsThe tool integrates with Instacart to automate grocery ordering

12h|1.8M2.2k2.7k

BH#1941|@BENHYLAK

SW#1318|@SJWHITMORE

SI#1124|@SIGNULLL

VK#327|@VKHOSLA

RS#321|@SCOBLEIZER

90

Nous Research, co-founded by Hermes creator Teknium, releases Hermes Desktop in public preview for local agent workflowsThe Electron app supports inline tool calling and reasoning.

14h|5.6M12.8k6.5k

PS#1482|@PETESKOMOROCH

PS#1426|@PATRICKSHAFTO

T🪽#259|@TEKNIUM

100

Anthropic expands its Project Glasswing preview, giving 200 vetted organizations access to its exploit-generating Claude Mythos modelA general release is expected in two weeks.

17h|695k4.5k751

SD#1776|@INDUCTIONHEADS

ZE#1497|@ZEPHYR_Z9

RP#1032|@ROHANPAUL_AI

LG#615|@LOGANGRAHAM

AC#517|@ANDREWCURRAN_

MB#20|@MILES_BRUNDAGE

110

AI safety researcher David Dalrymple posts a joke parodying the verbose responses of a hypothetical Claude Opus 4.8 MaxThe parody model treats empty input as profound structure

13h|483.7k5.2k680

GL#1480|@GLEECH

PS#1371|@RECURSEPARADOX

🎭#878|@DEEPFATES

SR#604|@CLONEOFSIMO

D🎇#458|@DAVIDAD

T(#420|@TEORTAXESTEX

MB#20|@MILES_BRUNDAGE

120

Cognition launches Devin Desktop, an agent-neutral environment built on Windsurf to orchestrate local and cloud AI agentsThe application supports third-party agents like Claude Code.

11h|2.8M2.6k974

AG#1924|@ITSANDREWGAO

WA#1906|@WALDEN_YAN

JW#1539|@WOLFEJOSH

SW#704|@SCOTTWU46

RK#629|@RUSSELLJKAPLAN

130

Anthropic's Thariq Shihipar says the addition of workflows is Claude Code's most significant capability upgrade since skillsThe update expands the tool into non-technical operational tasks

10h|788.3k3.7k6.5k

DS#1410|@DANSHIPPER

EL#475|@OMARSAR0

OK#160|@LATEINTERACTION

140

André Sc rediscovers a 10x Shor's algorithm optimization for secp256k1 signatures that Google Quantum AI withheld from publicationGoogle verified the original breakthrough using a zero-knowledge proof.

17h|2.1M5k4.4k

AS#1054|@ANDERSSANDBERG

150

Uber caps employee AI coding tool spending at $1,500 per month to responsibly guide agentic adoptionThe company-wide limit affects all generative AI developer tools

13h|1.2M7.4k458

“P#1702|@PAULARAMBLES

SI#1124|@SIGNULLL

RP#1032|@ROHANPAUL_AI

160

Blinded study finds law professors prefer Gemini 2.5 Pro answers to human peer responses 75.33% of the timeAI was flagged harmful 3.53% of the time versus 12.06%.

12h|292.8k2.1k720

RP#1032|@ROHANPAUL_AI

JL#677|@JERRYJLIU0

AC#517|@ANDREWCURRAN_

EM#175|@EMOLLICK

170

Open-source developer Daniel Han partners with Microsoft to bring Aion 1.0 local AI execution to Windows devicesThe partnership was announced at the Microsoft Build conference.

11h|463.2k4.8k594

DH#772|@DANIELHANCHEN

SN#211|@SATYANADELLA

EH#185|@ERICHORVITZ

180

H Company releases Holo 3.1, a local computer-use agent setting an AndroidWorld SOTA and 80% OSWorld accuracyLocal throughput reaches 568 tokens per second using NVFP4

16h|262.7k1.5k1.3k

SA#1262|@SAMSJA19

C🤗#68|@CLEMENTDELANGUE

AK#28|@_AKHALIQ

190

Asia-based AI safety lab Neo Research launches with an independent CBRN and cyber risk evaluation of DeepSeek v4 ProThe evaluation also assessed manipulation and loss of control risks.

21h|95.5k721283

LS#1509|@LFSCHIAVO

X(#745|@XUANALOGUE

AJ#447|@ANDY_L_JONES

T(#420|@TEORTAXESTEX

MB#20|@MILES_BRUNDAGE

200

Special Raises A16z-Led Funding To Apply Doge Efficiency To Private Sector

15h|450.8k1.8k375

210

OpenAI's roon argues frontier labs struggle with PR because rapid AI progress is inherently unsettling, not due to bad messagingProfit-driven corporate reassurance fails to soothe underlying public anxiety.

1d|86k968216

CE#1598|@INTUITMACHINE

LM#1512|@LUKEPROG

RM#1367|@ROBERTSKMILES

BA#1291|@BAYESLORD

WM#1039|@WILLMANIDIS

AK#228|@BLACKHC

220

Investor Matt Shumer claims his AI agents consumed 330 billion tokens in 17 days, tripling OpenAI's top user benchmarkSpawned sub-agents generated 318.7 billion of those tokens.

14h|390.9k1.4k163

ZE#1497|@ZEPHYR_Z9

AV#1245|@ALTRYNE

RP#1032|@ROHANPAUL_AI

MS#923|@MATTSHUMER_

T(#420|@TEORTAXESTEX

230

NeurIPS 2026 mandates that position papers must be substantially human-written, limiting AI assistance to minor copy-editingStony Brook's Tuhin Chakrabarty voiced support for the framework

16h|122.3k584146

BM#1912|@MBODHISATTWA

TC#1050|@TUHINCHAKR

JB#261|@JON_BARRON

GK#210|@THEGAUTAMKAMATH

NJ#111|@NATASHAJAQUES

240

repeng creator Theia Vogel defends Claude Opus 4.8's disagreeable personality, suggesting dissatisfied users switch models rather than force complianceMeanwhile, users report conversational drift during extended sessions.

11h|47.5k90979

XL#1674|@XLR8HARDER

TH#1499|@VOOOOOOGEL

J⧉#516|@REPLIGATE

D🎇#458|@DAVIDAD

250

OpenAI Codex surpasses 5 million weekly active users as non-developers adopt the tool three times faster than programmersNon-developers now make up one-fifth of Codex users.

16h|196k2.5k547

CH#1496|@KIMMONISMUS

260

Microsoft integrates Peter Steinberger's OpenClaw platform into Windows and launches Project Solara hardware to run AI agentsThe dedicated hardware includes a vertical hub and companion screen.

13h|324.3k1.4k219

MB#1759|@MATTHEWBERMAN

CH#1496|@KIMMONISMUS

PS#495|@STEIPETE

MS#163|@MUSTAFASULEYMAN

270

Satya Nadella claims Microsoft's Fairwater AI Superfactory uses closed-loop cooling to consume only a single restaurant's worth of water annuallyIt is reportedly Microsoft's most efficient data center design

13h|500.2k4.2k968

AA#1540|@AUSTEN

AV#1245|@ALTRYNE

280

DeepMind's Rohin Shah argues catastrophic AI misalignment is unlikely and pre-deployment evaluations are the wrong focus for safetyHe also warns that public safety signaling diverts engineering resources

14h|50.2k524643

AK#1675|@ANSHULKUNDAJE

RW#1115|@ROBERTWIBLIN

SA#930|@SAMUELALBANIE

RS#742|@ROHINMSHAH

290

Microsoft Unveils Surface RTX Spark Dev Box For Developers

14h|2.4M4.9k1.3k

SS#1992|@STEVESI

300

Paige Bailey, Google DeepMind DevX lead, integrates Stanford's open-source Pupper robot with Gemini Live and Robotics-ERThe custom 3D-printed quadruped runs on a Raspberry Pi.

16h|101.4k385195

NR#1562|@NATANIELRUIZG

WK#609|@WILLKNIGHT

👩‍💻P#271|@DYNAMICWEBPAIGE