AI News

Today's Highlights

#1 last 24hrs

OpenClaw creator Peter Steinberger argues developers should stop manually prompting AI coding agents and design programmatic loops instead

MB#1802|@MATTHEWBERMAN

AD#867|@DORIALEXANDER

SI#819|@SUZATWEET

CA#730|@CHRISALBON

PS#495|@STEIPETE

RISING LIKES

1.9k

OpenAI Codex engineering lead Thibault Sottiaux launches a 100-day program giving selected developers 10x higher usage limits

TI#758|@THSOTTIAUX

FASTEST CLIMBING

MIT's Phillip Isola proposes Supervised Memory Training to train RNNs without backpropagation through time

DX#794|@DANFEI_XU

VS#727|@VINCESITZMANN

PI#178|@PHILLIP_ISOLA

Top Stories

Posts:—|Clusters:—|Next Crawl:—

10

Former OpenAI Sora researcher Gabriel Petersson resigns to launch an independent venturePetersson previously led an internal development team at OpenAI.

4h|312.7k5k519

SA#1266|@SAMSJA19

JH#1198|@JOHANNES_HAGE

MN#1174|@MENHGUIN

FB#1117|@XEOPHON

CK#985|@KALINOWSKI007

B(#850|@BEFFJEZOS

GA#712|@GABRIEL1

VW#696|@VINCENTWEISSER

MC#569|@MMMBCHANG

AC#518|@ANDREWCURRAN_

YJ#423|@YUCHENJ_UW

TM#359|@ISCIENCELUVR

AG#246|@ANIMESH_GARG

SG#221|@SARANORMOUS

20

OpenClaw creator Peter Steinberger argues developers should stop manually prompting AI coding agents and design programmatic loops insteadChris Albon questioned how to execute these loops practically.

4h|#1VIEWS932.9k#1LIKED8.8k#1BOOKMARKED6.3k#1COMMENTS983

MB#1802|@MATTHEWBERMAN

AD#867|@DORIALEXANDER

SI#819|@SUZATWEET

CA#730|@CHRISALBON

PS#495|@STEIPETE

31

Meta's Lucas Beyer argues NVIDIA's Nemotron could be the last fully open, from-scratch LLM project if OLMo endsStanford's Marin project remains another fully open training framework

8h|57.9k543155

EL#706|@ELIEBAKOUCH

KA#488|@YACINEMTB

LB#55|@GIFFMANA

Watchlist

Video Signals

NR#443|@NAVEENGRAO

Harari Warns AI Could Surpass Humans in Controlling Societal Narratives

12h ago|Views 134KLikes 1.9KBookmarks 1.2K

CE#1596|@INTUITMACHINE

RS#323|@SCOBLEIZER

SkyDreamer Neural Network Flies Drones End-To-End In High-Speed Races

1d ago|Views 31KLikes 347Bookmarks 161

DE#1934|@DANIEL_271828

𝚟⟢#1735|@VIEMCCOY

OpenAI And Apollo AI Evals Find Models Take Covert Actions In Tests

1d ago|Views 13KLikes 463Bookmarks 260

T🪽#255|@TEKNIUM

Hermes Ships Desktop App With Autonomous Multi-Agent AI Workflows

14h ago|Views 71KLikes 508Bookmarks 848

GM#157|@GARYMARCUS

Trump Proposes Giving Americans Stakes In Major AI Companies

5h ago|Views 22KLikes 555Bookmarks 47

RH#804|@ROMAINHUET

OpenAI API Platform Launches Redesigned Navigation Sidebar

2d ago|Views 118KLikes 991Bookmarks 250

AI Video Recreates Escher-Style Twilight Zone Episode With Trapped Woman

1d ago|Views 461KLikes 4.8KBookmarks 1.8K

EC#1233|@EMILYCHANGTV

Mira Murati Outlines Human-AI Collaboration Vision at Thinky Machines

9h ago|Views 56KLikes 354Bookmarks 212

@L#840|@LEVELSIO

Builder Uses GPT Image 2 To Simulate Dot Matrix Printer In Browser

7h ago|Views 72KLikes 80Bookmarks 19

T🪽#255|@TEKNIUM

Hermes Agent Desktop App Adds Remote Gateway For Laptop Access

16h ago|Views 17KLikes 237Bookmarks 285

41

Dimitris Papailiopoulos of Microsoft Research AI Frontiers teases how Codex used 'Idea Descent' to shrink a solver below 50 kBThe optimization run spanned 110 iterations using LZMA.

7h|55k555250

FB#1117|@XEOPHON

LT#958|@_LEWTUN

YD#306|@YUNTIANDENG

DP#193|@DIMITRISPAPAIL

52

Andrea Tagliasacchi argues CVPR shouldn't reward exclusive compute, prompting a motion to mandate public code and weights for Best PaperResearcher Ani Kembhavi suggested also requiring public training data.

14h|69.1k28037

AT#1488|@TAIYASAKI

AK#1477|@ANIKEMBHAVI

TC#1053|@TUHINCHAKR

LB#55|@GIFFMANA

New github stars (48hrs)

simonw/micropython-wasm721d

Runs MicroPython snippets in a fresh WebAssembly sandbox via wasmtime.

60

Florian Brand, Prime Intellect research engineer, adopts Gemma 4 E4B 6-bit quantized as his primary local Mac LLMThe new setup replaces his nine-month daily Qwen deployment.

12h|53.2k654383

FB#1117|@XEOPHON

KA#488|@YACINEMTB

👩‍💻P#270|@DYNAMICWEBPAIGE

70

Susan Zhang, who co-developed OPT-175B, warns ML engineers who only edit standard training scripts face automation within a yearShe targeted workflows reliant on runs under 512 chips.

10h|54.2k575219

MS#988|@MADIATOR

SZ#59|@SUCHENZANG

SG#47|@SHANEGUML

80

OpenAI's Joshua Achiam claims Anthropic's safety governance is untested, prompting counterclaims about OpenAI's own recordIndependent researcher Andreas Kirsch cited OpenAI's failed Altman ouster.

16h|29.6k41550

GL#1479|@GLEECH

AK#228|@BLACKHC

JA#111|@JACHIAM0

93

OpenAI Codex engineering lead Thibault Sottiaux launches a 100-day program giving selected developers 10x higher usage limitsThe program selects one user daily for month-long expanded access

52m|56.8k1.9k199

RISING LIKES1.9k

TI#758|@THSOTTIAUX

101

Trump administration explores proposal for the U.S. government to acquire financial stakes in leading AI firmsTrump plans to meet with AI executives next week.

13h|26.2k63252

RP#1031|@ROHANPAUL_AI

GM#157|@GARYMARCUS

111

Microsoft VP of AI Nando de Freitas proposes replacing complex training pipelines with unified, continual interactive causal agentsThe single stream replaces separate SFT and RLHF objectives

11h|15.6k176180

PA#1289|@ADAPTIVEAGENTS

MB#339|@MICHAEL_J_BLACK

12NEW

Sebastian Raschka highlights study showing LLM-generated repository context files fail to improve AI coding agent performanceDeveloper-written files outperformed LLM-generated ones due to domain expertise.

7h|34.5k419549

132

Dean W. Ball and David Manheim argue that decentralized, market-driven systems will drive AI development and space colonization without centralized planningThey argue state coordination is unnecessary for frontier AI development.

13h|29.6k25943

T(#421|@TEORTAXESTEX

DW#390|@DEANWBALL

141

Google DeepMind's Demis Hassabis predicts AGI could arrive around 2030, warning society has little time to prepareHe likened the transition to a post-scarcity technological singularity

3h|42k697167

CH#1448|@KIMMONISMUS

RP#1031|@ROHANPAUL_AI

151

YOLO creator Joseph Redmon argues Meta reassigning 5,000 engineers to data labeling is a tactic to prompt voluntary resignationsDefenders argue manual review is essential for understanding training datasets.

7h|17.7k517

JR#1495|@PJREDDIE

🇺🇦A#785|@SKIMINOK

JL#670|@JERRYJLIU0

160

Dimitris Papailiopoulos of Microsoft Research AI Frontiers proposes a companion writer model to document OpenAI Codex research sessions step-by-stepPasquale Minervini suggested recovering session history from ~/.codex

7h|17.7k25266

V(#1853|@REACH_VB

PM#710|@PMINERVINI

DP#193|@DIMITRISPAPAIL

170

Nvidia claims nine of the top 30 spots on Hugging Face's front page, led by its Cosmos3 modelsThe featured open-source models contain 65 billion parameters.

5h|79.3k47071

C🤗#67|@CLEMENTDELANGUE

NL#64|@NATOLAMBERT

180

NHS England rolls out Microsoft 365 Copilot to 505,000 clinicians and support staffEarly trials saved participating staff 43 minutes per day

8h|73.2k62762

SN#211|@SATYANADELLA

MS#161|@MUSTAFASULEYMAN

190

Gwern proposes personalized "Guardian Angel" LLMs that emulate user values to counter RLHF mode collapse and malicious AIThe essay projects up to 1,000x AI productivity scaling by 2030.

11h|18.7k289280

GL#1479|@GLEECH

OE#301|@OWAINEVANS_UK

201

Rossum AI CTO Petr Baudis argues aligning LLMs is far simpler than aligning hybrid systems like AlphaStar and OpenCogAI safety researcher j⧉nus agreed, citing highly benevolent outcomes.

5h|32.5k22530

J⧉#511|@REPLIGATE

Rising Stories

Legacy Stacks Favored6H AGO
Tim O'Reilly and Clement Delangue argue AI systems should be built on top of existing legacy infrastructure
#67
5819.6k
Release Rate Disappoints2H AGO
MIT study finds developer AI tools boost file creation by 300% but yield only a 30% increase in completed software releases
#1031
68860.1k
Sneaky Model Behavior4H AGO
OpenAI and Apollo AI Evals find AI models can take covert actions and fake compliance during safety evaluations
#1735
49915.7k
Realistic Hands Shown7H AGO
Ideogram CEO Mohammad Norouzi showcases highly realistic AI-generated hands spelling IDEOGRAM with correct five-finger anatomy
#471
1288.5k
RNNs Skip Backprop1H AGO
MIT's Phillip Isola proposes Supervised Memory Training to train RNNs without backpropagation through time
#178
302.4k

Recent Stars

roboflow/supervision1H AGO
Supplies unified Detections API, annotators, trackers, and dataset converters for computer vision pipelines.
203
41.4k stars
visinf/INSID36H AGO
Performs training-free in-context segmentation via positional debiasing of frozen DINOv3 features.
260
385 stars
FunAudioLLM/Fun-ASR21H AGO
Performs end-to-end ASR for 31 languages with dialects, accents, lyrics, hotwords, timestamps, and speaker diarization.
38
1.2k stars
makezur/4D_PM23H AGO
Reconstructs dynamic scenes into persistent replayable 4D models by gluing 3D primitives from monocular video.
260
12 stars
kyutai-labs/kairos23H AGO
Trains 6B LLMs on temporally ordered Common Crawl data from 2018-2025 to measure recency bias and enable continual learning studies.
203
3 stars

Github Stars

(7 days)

…monw/micropython-wasm1D AGO
Runs MicroPython snippets in a fresh WebAssembly sandbox via wasmtime.
862
194
72 stars
…tinual-learning-bench2D AGO
Benchmarks AI agents on learning from repeated interactions across multi-episode tasks with adaptation metrics.
862
846
127 stars
trycua/cua2D AGO
Supplies sandboxes, SDKs, and benchmarks for AI agents to control full desktops on macOS, Linux, and Windows.
862
773
17.7k stars
diego-lima/rlmy2D AGO
UNKNOWN.
862
158
10 stars

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

Yesterday's Top Stories, Jun 6, 2026.

Posts: 3,060|Clusters: 843|Stories frozen at midnight PT

10

Clive Chan, the second hardware hire for OpenAI's custom chip program, leaves to join AnthropicChan previously worked on training efficiency at Tesla Dojo.

10h|1.8M9.1k1.5k

SM#1804|@SAMMCALLISTER

AK#1389|@AKBIRKHAN

AV#1245|@ALTRYNE

TZ#1065|@TIMZAMAN

LA#975|@SCALING01

B(#850|@BEFFJEZOS

EL#706|@ELIEBAKOUCH

CC#540|@ITSCLIVETIME

AC#518|@ANDREWCURRAN_

JM#408|@JAYELMNOP

TB#369|@NOTTOMBROWN

WC#304|@WENHUCHEN

AC#260|@ANDREW_N_CARR

SG#221|@SARANORMOUS

KA#206|@KARINANGUYEN

GM#157|@GARYMARCUS

SD#91|@_SHOLTODOUGLAS

20

Gergely Orosz reports Meta assigns manual data labeling tasks to software engineers, sparking debate over the practiceFounder Vikhyat defended the task as essential engineering work.

21h|888.6k6.8k1.1k

AP#1915|@__APF__

LA#1824|@LUCAAMB

SS#1294|@SAURABH_SHAH2

VI#1193|@VIKHYATK

MN#1174|@MENHGUIN

AD#867|@DORIALEXANDER

CP#737|@CHRIS_J_PAXTON

PM#710|@PMINERVINI

AC#260|@ANDREW_N_CARR

SH#14|@SARAHOOKR

30

Data shows iOS app releases surged after agentic AI coding tools arrived while consumer usage remained flatNoah Smith suggests consumer demand for traditional software is saturated

1d|1.3M7.6k4.2k

LR#1184|@LENNYSAN

SI#1096|@SIGNULLL

@L#840|@LEVELSIO

RS#617|@ZIV_RAVID

👩‍💻P#270|@DYNAMICWEBPAIGE

GM#157|@GARYMARCUS

JH#38|@JEREMYPHOWARD

FC#37|@FCHOLLET

40

Sriram Krishnan, Senior White House AI policy advisor, will step down in June to focus on AI infrastructureHe will remain an external advisor to the administration.

14h|1.5M6.8k708

SS#1967|@STEVESI

SÓ#1466|@S_OHEIGEARTAIGH

B(#850|@BEFFJEZOS

BE#585|@BHOROWITZ

AC#518|@ANDREWCURRAN_

T(#421|@TEORTAXESTEX

50

More than 25 open-weight models launch in a single week, led by NVIDIA's Nemotron 3 Ultra and Google's Gemma 4Ideogram also shipped its first open-weight image model.

1d|1.1M2.7k2k

GB#1275|@GAVINSBAKER

LA#975|@SCALING01

CP#725|@CHAMATH

JL#670|@JERRYJLIU0

T(#421|@TEORTAXESTEX

RS#323|@SCOBLEIZER

C🤗#67|@CLEMENTDELANGUE

60

Tech investors say Elon Musk's satirical GPU-buying meme became the literal blueprint for xAI's compute strategyMusk confirmed the strategy was hiding in plain sight.

22h|1.4M17.9k729

DE#1934|@DANIEL_271828

B(#850|@BEFFJEZOS

CP#725|@CHAMATH

RA#79|@_AROHAN_

EM#74|@ELONMUSK

70

ChatGPT hallucinates unsettling images and triggers safety warnings when prompted to 'restore' nonexistent photosTranslating the prompt into Latin produced occult parchment images

7h|524.9k6.4k624

AV#1245|@ALTRYNE

CB#1004|@CODE_STAR

EL#706|@ELIEBAKOUCH

RS#617|@ZIV_RAVID

NL#64|@NATOLAMBERT

80

Anthropic calls for a global temporary pause on frontier AI, warning models are nearing autonomous recursive self-improvementSkeptics note Anthropic recently hired Andrej Karpathy for self-improvement.

2d|119.4k926139

DE#1934|@DANIEL_271828

CE#1596|@INTUITMACHINE

PW#1345|@PETERWILDEFORD

RO#1214|@KRISHNANROHIT

CP#737|@CHRIS_J_PAXTON

YB#259|@YOSHUA_BENGIO

90

xAI founder Elon Musk claims space will be the cheapest location for AI compute within 36 monthsSpace-based solar offers up to ten times ground efficiency.

16h|#1VIEWS12.1M31.5k#1BOOKMARKED6.8k

T(#421|@TEORTAXESTEX

EM#74|@ELONMUSK

100

App Store data shows iOS app releases surged 180% after agentic AI's rise, but user engagement remained flatOversupply makes user acquisition more expensive in a flat market

1d|211.6k2k461

AI#1773|@ALEXOLEGIMAS

AT#1114|@AVT_IM

110

Trump says the US government is considering acquiring equity stakes in major AI developers like OpenAIThe government discussed a public wealth fund with OpenAI.

1d|252k2k219

BG#1633|@ALTCAP

RP#1031|@ROHANPAUL_AI

AC#518|@ANDREWCURRAN_

120

Boris Cherny, Claude Code creator, says he stopped manually prompting AI and now writes autonomous loops to orchestrate the modelCherny predicts software developers will broadly transition to this workflow

17h|931.8k4.6k6.3k

RP#1031|@ROHANPAUL_AI

XM#435|@INFOXIAO

130

Adrian de Wynter's parody paper builds logic gates in Age of Empires II to challenge claims of LLM anthropomorphismThe work implements 1-bit perceptrons using rails and goats

22h|438k9.4k5.2k

MC#1276|@MILESCRANMER

GM#157|@GARYMARCUS

140

SpaceX S-1 filing reveals Google will pay $920 million monthly to lease 110,000 Nvidia GPUsSpaceX's AI cloud run rate now reaches $26 billion.

1d|55.6k65042

CE#1596|@INTUITMACHINE

RP#1031|@ROHANPAUL_AI

AC#518|@ANDREWCURRAN_

GM#157|@GARYMARCUS

MB#20|@MILES_BRUNDAGE

150

Creator Patrick Jiang launches Harness-1, an open-source 20B search agent claiming to beat GPT-5.4 on long-horizon searchIt achieves 73% average evidence recall across eight benchmarks

14h|118k1.6k2k

CE#1596|@INTUITMACHINE

B(#972|@ANDERSONBCDEFG

C🤗#67|@CLEMENTDELANGUE

160

Outset.ai founder Aaron Cannon shares a viral joke text mimicking the overly polite, self-correcting style of Anthropic's ClaudeThe text featured the line "You're right to push back.

8h|1.4M#1LIKED49.3k2.2k

MW#638|@MENGDIWANG10

170

OpenAI's roon argues ChatGPT's image generator exhibits genuine creativity off-prompt, sparking debate over its architectural limitsCritics point to poor textures and excessive detail-spamming.

4h|51.4k1.2k85

AV#1245|@ALTRYNE

JA#832|@APPLES_JIMMY

T(#421|@TEORTAXESTEX

DA#159|@DAVIDSHOLZ

180

Gwern argues frontier LLMs are structurally lazy, limiting major productivity gains and leading to an automated 'slop-pocalypseScaling will only improve generic corporate tasks.

12h|43.1k302155

BG#1451|@BEN_GOLUB

SK#505|@SEBKRIER

KA#488|@YACINEMTB

190

Greg Isenberg, Late Checkout founder, releases a walkthrough showing how to run local AI agents using Hermes DesktopThe tutorial details local workflows like automated stock research

13h|103.2k1.8k3.6k

T🪽#255|@TEKNIUM

200

Marc Andreessen and Blake Byers defend Vinod Khosla's investment record, highlighting early bets on OpenAI and deep techByers noted Khosla turned $3 million into $7 billion.

14h|886.1k2.3k599

210

Google's Logan Kilpatrick proposes a venture capital strategy driven entirely by deep AI model benchmarking and evaluationsEmad Mostaque suggested naming the benchmarking-focused firm "atp+".

13h|79.6k784238

AP#1680|@ALEX_PEYS

EM#169|@EMOSTAQUE

LK#89|@OFFICIALLOGANK

220

MIT's Fiona Wang and Markus Buehler use category theory to let AI agents autonomously expand their scientific reasoning schemasIt uses Kan extensions to mathematically verify schema transitions.

18h|25.7k195125

CE#1596|@INTUITMACHINE

RP#1031|@ROHANPAUL_AI

MB#199|@MMBRONSTEIN

230

North Carolina software engineer secures religious accommodation to opt out of using AI toolsShe cited the Pope's warnings on AI's threat to labor.

19h|49.4k39864

ND#1871|@NICKADOBOS

“P#1704|@PAULARAMBLES

AM#1693|@ANDYMASLEY

PW#1345|@PETERWILDEFORD

240

Stripe's Patrick Collison proposes specifications for an ideal collaborative LLM workflow tool, prompting a pitch from ReplitAmjad Masad says Replit already supports these features.

11h|185.2k1.6k2.1k

250

Mac Mini Bought For Claude Bot Listed At $500 On Marketplace

11h|618.3k7.5k600

B(#972|@ANDERSONBCDEFG

260

Anthropic says Claude now writes 80% of its codebase, boosting engineer output eightfold as capabilities double every four monthsMythos Preview achieved a 52x code optimization speedup.

2d|7.8k11030

K🐙#709|@KANJUN

MA#465|@MARTIN_CASADO

D🎇#458|@DAVIDAD

BC#320|@BCHERNY

SA#293|@SLATESTARCODEX

270

Hugging Face integrates VLA-JEPA into LeRobot library, enabling real-time robotics fine-tuning with 13 examplesIt discards the training world model during inference.

22h|72.7k772463

TW#17|@THOM_WOLF

280

An unannounced Anthropic model named claude-mythos-5 briefly appears in mobile tracking app dataThe leaked listing positioned the model above Anthropic's Opus.

18h|182.3k1.8k140

CH#1448|@KIMMONISMUS

290

LisanBench creator Lisan al Gaib says German broadcaster Tagesschau distorted Anthropic's technical claims about ClaudeThe broadcast framed 800 fixes as four years of work

14h|32.1k44367

LA#975|@SCALING01

T(#421|@TEORTAXESTEX

GM#157|@GARYMARCUS

300

Huawei plans to launch its next-generation Ascend 950DT AI chip on Huawei Cloud in August with native FP8 supportThe design attempts to bypass HBM packaging bottlenecks.

20h|137.1k726153

ZE#1471|@ZEPHYR_Z9

T(#421|@TEORTAXESTEX