AI News

Today's Highlights

In case you missed it

Bernie Sanders proposes the American AI Sovereign Wealth Fund Act to take public equity stakes in OpenAI and Anthropic

Tanishq Mathew Abraham, Ph.D.TM#359|@ISCIENCELUVR
EmadEM#167|@EMOSTAQUE
Gary MarcusGM#153|@GARYMARCUS
Joshua AchiamJA#109|@JACHIAM0
Miles BrundageMB#20|@MILES_BRUNDAGE
+23 more
#1 VIEWED

Nous Research, co-founded by Hermes creator Teknium, releases Hermes Desktop in public preview for local agent workflows

Pete SkomorochPS#1482|@PETESKOMOROCH
Teknium 🪽T🪽#259|@TEKNIUM
RISING LIKES
171

repeng creator Theia Vogel defends Claude Opus 4.8's disagreeable personality, suggesting dissatisfied users switch models rather than force compliance

xlr8harderXL#1674|@XLR8HARDER
thebesTH#1499|@VOOOOOOGEL
j⧉nusJ⧉#516|@REPLIGATE
davidad 🎇D🎇#458|@DAVIDAD
FASTEST CLIMBING

Microsoft AI CEO Mustafa Suleyman predicts AI training compute will scale by three orders of magnitude over the next three years

Lisan al GaibLA#980|@SCALING01
Chris PaxtonCP#732|@CHRIS_J_PAXTON
Andrew CurranAC#517|@ANDREWCURRAN_

Top Stories

Posts:—|Clusters:—|Next Crawl:—
11

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.

2h|212.8k2.7k877
Super DarioSD#1776|@INDUCTIONHEADS
Bindu ReddyBR#1622|@BINDUREDDY
Chubby♨️CH#1496|@KIMMONISMUS
Alex VolkovAV#1245|@ALTRYNE
Florian BrandFB#1153|@XEOPHON
Rohan PaulRP#1032|@ROHANPAUL_AI
Lisan al GaibLA#980|@SCALING01
Beff (e/acc)B(#839|@BEFFJEZOS
Anastasios Nikolas AngelopoulosAN#800|@ML_ANGELOPOULOS
elieEL#716|@ELIEBAKOUCH
Kyle LoKL#620|@KYLELOSTAT
Ravid Shwartz ZivRS#612|@ZIV_RAVID
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
finbarrFI#397|@FINBARRTIMBERS
Luca Soldaini 🎀LS#279|@SOLDNI
Hanna HajishirziHH#276|@HANNAHAJISHIRZI
Mustafa SuleymanMS#163|@MUSTAFASULEYMAN
Lucas Beyer (bl16)LB#55|@GIFFMANA
Nando de FreitasND#29|@NANDODF
21

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.

6h|120.4k1.6k112
Vaibhav (VB) SrivastavV(#1853|@REACH_VB
Sharon GoldmanSG#1498|@SHARONGOLDMAN
samsjaSA#1262|@SAMSJA19
Mimansa JaiswalMJ#1194|@MIMANSAJ
Gokul SwamyGS#1055|@G_K_SWAMY
Cody BlakeneyCB#999|@CODE_STAR
Keerthana GopalakrishnanKG#730|@KEERTHANPG
elieEL#716|@ELIEBAKOUCH
merveME#675|@MERVENOYANN
Jason PhangJP#672|@ZHANSHENG
Victor SanhVS#520|@SANHESTPASMOI
Andrew CurranAC#517|@ANDREWCURRAN_
Xiao MaXM#437|@INFOXIAO
Yuchen JinYJ#427|@YUCHENJ_UW
finbarrFI#397|@FINBARRTIMBERS
will brownWB#339|@WILLCCBB
Akari AsaiAA#334|@AKARIASAI
Luca Soldaini 🎀LS#279|@SOLDNI
Hanna HajishirziHH#276|@HANNAHAJISHIRZI
Andrew Carr 🤸AC#263|@ANDREW_N_CARR
Rosanne LiuRL#137|@SAVVYRL
Nathan LambertNL#64|@NATOLAMBERT
Lucas Beyer (bl16)LB#55|@GIFFMANA
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.

5h|365.4k2.5k506
Steven SinofskySS#1992|@STEVESI
Nathan 🔎N🔎#1902|@NATHANPMYOUNG
Andy MasleyAM#1735|@ANDYMASLEY
Brad GerstnerBG#1614|@ALTCAP
Samuel Hammond 🦉SH#1488|@HAMANDCHEESE
Seán Ó hÉigeartaighSÓ#1457|@S_OHEIGEARTAIGH
Greg KamradtGK#1182|@GREGKAMRADT
Kylie RobisonKR#1046|@KYLIEBYTES
Zvi MowshowitzZM#918|@THEZVI
Divyansh KaushikDK#847|@DKAUSHIK96
Andrew CurranAC#517|@ANDREWCURRAN_
Dean W. BallDW#392|@DEANWBALL
Gary MarcusGM#153|@GARYMARCUS
Miles BrundageMB#20|@MILES_BRUNDAGE
Watchlist

Video Signals

Vaibhav (VB) SrivastavV(#1853|@REACH_VB
Alex VolkovAV#1245|@ALTRYNE
Rohan PaulRP#1032|@ROHANPAUL_AI
Steven HeidelSH#1026|@STEVENHEIDEL
Sherwin WuSW#801|@SHERWINWU
Boris PowerBP#357|@BORISMPOWER

OpenAI Rolls Out Sites To Build Shareable Apps With Codex

4h ago|Views 1.7MLikes 8.7KBookmarks 3.9K
Pete SkomorochPS#1482|@PETESKOMOROCH
Teknium 🪽T🪽#259|@TEKNIUM

Nous Research Releases Hermes Desktop App in Public Preview

5h ago|Views 1.7MLikes 6.2KBookmarks 3.3K
ben hylakBH#1941|@BENHYLAK
Sam WhitmoreSW#1318|@SJWHITMORE

Ollie Launches As First Proactive AI Family Assistant

3h ago|Views 390KLikes 826Bookmarks 646
Bilawal SidhuBS#1909|@BILAWALSIDHU
Kosta Derpanis (sabbatical in Zurich)KD#908|@CSPROFKGD
Anjney MidhaAM#527|@ANJNEYMIDHA

Martin Scorsese Advises Black Forest Labs On FLUX Visual AI

7h ago|Views 720KLikes 1.4KBookmarks 690
Carlos E. PerezCE#1598|@INTUITMACHINE
Shaun MaguireSM#1333|@SHAUNMMAGUIRE
Rohan PaulRP#1032|@ROHANPAUL_AI
Garry TanGT#270|@GARRYTAN
clem 🤗C🤗#68|@CLEMENTDELANGUE

Factory Launches Router For Automated AI Model Selection

3h ago|Views 533KLikes 808Bookmarks 488
Super DarioSD#1776|@INDUCTIONHEADS
Aravind SrinivasAS#118|@ARAVSRINIVAS

Perplexity Adds Hybrid Agentic Inference to Perplexity Computer

3h ago|Views 119KLikes 1KBookmarks 334
Alex VolkovAV#1245|@ALTRYNE
Teknium 🪽T🪽#259|@TEKNIUM

Hermes Agent Delivers Smooth Token Streaming in Telegram Bot

12h ago|Views 162KLikes 1.6KBookmarks 307
jasonJA#929|@JXNLCO
Chris AlbonCA#729|@CHRISALBON

Pauline Narvas Joins OpenAI Developer Experience Team

6h ago|Views 142KLikes 1.4KBookmarks 277
Peyman MilanfarPM#318|@DOCMILANFAR

Google DeepMind Launches Co-Scientist Gemini System For Hypothesis Generation

4h ago|Views 44KLikes 610Bookmarks 205
Steven SinofskySS#1992|@STEVESI

Microsoft Unveils Surface RTX Spark Dev Box For Developers

4h ago|Views 364KLikes 1.8KBookmarks 390
40

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.

4h|322.7k1.2k336
ben hylakBH#1941|@BENHYLAK
xlr8harderXL#1674|@XLR8HARDER
ZephyrZE#1497|@ZEPHYR_Z9
Chubby♨️CH#1496|@KIMMONISMUS
Alex VolkovAV#1245|@ALTRYNE
Lisan al GaibLA#980|@SCALING01
elieEL#716|@ELIEBAKOUCH
Andrew CurranAC#517|@ANDREWCURRAN_
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
Tanishq Mathew Abraham, Ph.D.TM#359|@ISCIENCELUVR
swyxSW#214|@SWYX
Lucas Beyer (bl16)LB#55|@GIFFMANA
51

Factory launches Factory Router, an automatic model routing system designed to cut AI inference token costs by 20% to 25%The system uses cost-performance Pareto analysis to select models.

3h|791.1k2.1k1k
Matthew BermanMB#1759|@MATTHEWBERMAN
Carlos E. PerezCE#1598|@INTUITMACHINE
Shaun MaguireSM#1333|@SHAUNMMAGUIRE
Rohan PaulRP#1032|@ROHANPAUL_AI
Garry TanGT#270|@GARRYTAN
clem 🤗C🤗#68|@CLEMENTDELANGUE
New github stars (48hrs)
NVIDIA/cosmos8.4k5h
Hosts omnimodal Mixture-of-Transformers models for joint text, vision, audio, and action generation and reasoning in Physical AI.
Ming-Yu LiuML1449
Andrew Carr 🤸AC263
google-gemma/gemma-skills1361d
Supplies skills for Gemma model and agent interactions including a dev skill for app building.
Philipp SchmidPS934
Omar SansevieroOS490
61

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.

6h|463.9k3.7k584
Super DarioSD#1776|@INDUCTIONHEADS
ZephyrZE#1497|@ZEPHYR_Z9
Rohan PaulRP#1032|@ROHANPAUL_AI
Logan GrahamLG#615|@LOGANGRAHAM
Andrew CurranAC#517|@ANDREWCURRAN_
Jack ClarkJC#34|@JACKCLARKSF
Miles BrundageMB#20|@MILES_BRUNDAGE
Kevin RoseKR1
70

Martin Scorsese joins Black Forest Labs as an advisor to guide development of FLUX visual AI modelsScorsese uses the model to generate pre-production storyboards

7h|804.1k2.2k1.1k
Bilawal SidhuBS#1909|@BILAWALSIDHU
Nataniel RuizNR#1562|@NATANIELRUIZG
bilalBI#1402|@BILALTWOVEC
Kosta Derpanis (sabbatical in Zurich)KD#908|@CSPROFKGD
Robin RombachRR#779|@ROBROMBACH
Anjney MidhaAM#527|@ANJNEYMIDHA
finbarrFI#397|@FINBARRTIMBERS
80

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.

4h|#1VIEWS2.1M#1LIKED8.2k#1BOOKMARKED4k#1COMMENTS1.1k
Pete SkomorochPS#1482|@PETESKOMOROCH
Teknium 🪽T🪽#259|@TEKNIUM
92

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

3h|42.9k1.2k121
gavin leech (Non-Reasoning)GL#1480|@GLEECH
🎭🎭#878|@DEEPFATES
davidad 🎇D🎇#458|@DAVIDAD
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
Miles BrundageMB#20|@MILES_BRUNDAGE
101

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

2h|544.9k1k922
ben hylakBH#1941|@BENHYLAK
Sam WhitmoreSW#1318|@SJWHITMORE
111

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.

6h|139.3k2k420
Chubby♨️CH#1496|@KIMMONISMUS
jasonJA#929|@JXNLCO
123

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

3h|120k765619
samsjaSA#1262|@SAMSJA19
clem 🤗C🤗#68|@CLEMENTDELANGUE
AKAK#28|@_AKHALIQ
131

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.

3h|167.9k803129
Matthew BermanMB#1759|@MATTHEWBERMAN
Chubby♨️CH#1496|@KIMMONISMUS
Peter Steinberger 🦞PS#495|@STEIPETE
Mustafa SuleymanMS#163|@MUSTAFASULEYMAN
144

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.

2h|142.1k63080
ZephyrZE#1497|@ZEPHYR_Z9
Alex VolkovAV#1245|@ALTRYNE
Matt ShumerMS#923|@MATTSHUMER_
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
151

Teknium, Nous Research co-founder, demonstrates real-time token streaming on Telegram using the Hermes Agent botThe bot runs Claude Opus 4.8 via OpenRouter

12h|178.9k1.7k358
Alex VolkovAV#1245|@ALTRYNE
Teknium 🪽T🪽#259|@TEKNIUM
161

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

6h|61.6k37997
Tuhin ChakrabartyTC#1050|@TUHINCHAKR
xuan (ɕɥɛn / sh-yen)X(#745|@XUANALOGUE
Gautam KamathGK#210|@THEGAUTAMKAMATH
Natasha JaquesNJ#111|@NATASHAJAQUES
172

Perplexity plans hybrid agentic inference to split AI tasks between local hardware and cloud modelsThe feature will debut first on Windows laptops.

3h|144.4k1.2k391
Super DarioSD#1776|@INDUCTIONHEADS
Aravind SrinivasAS#118|@ARAVSRINIVAS
182

Pauline P. Narvas, Vercel's head of community platforms, joins OpenAI to work on developer experienceShe will support developers building with ChatGPT and Codex.

5h|162.5k1.5k351
jasonJA#929|@JXNLCO
Chris AlbonCA#729|@CHRISALBON
193

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%.

2h|100.3k941278
Andrew CurranAC#517|@ANDREWCURRAN_
Ethan MollickEM#175|@EMOLLICK
201

Mathematician Bartosz Naskręcki and co-authors release the Leiden Declaration calling for global guidelines on AI in mathematical researchThe manifesto aims to preserve rigor in scientific workflows.

8h|19.3k18758
Patrick ShaftoPS#1426|@PATRICKSHAFTO
Aran NayebiAN#1409|@ARAN_NAYEBI
Tal LinzenTL#224|@TALLINZEN

Rising Stories

  1. Build Conference Live4H AGO
    Microsoft kicks off its annual Build 2026 developer conference, launching its Day 1 sessions
    Suzana IlićSI#825|@SUZATWEET
    Robert ScobleRS#321|@SCOBLEIZER
    3225.9M
  2. Compute Explosion Predicted1H AGO
    Microsoft AI CEO Mustafa Suleyman predicts AI training compute will scale by three orders of magnitude over the next three years
    Lisan al GaibLA#980|@SCALING01
    Andrew CurranAC#517|@ANDREWCURRAN_
    1908.9k
  3. Wall Attention Debut4H AGO
    Tilde introduces Wall Attention, allowing RoPE-free models trained on 4,000 tokens to generalize to 200,000-token contexts without retrained
    Aryaman AroraAA#678|@ARYAMAN2020
    rohan anilRA#83|@_AROHAN_
    23420.1k
  4. Claude Workflows Launch40M AGO
    Anthropic's Thariq Shihipar says the addition of workflows is Claude Code's most significant capability upgrade since skills
    Dan Shipper 📧DS#1410|@DANSHIPPER
    ThariqTH#861|@TRQ212
    51849.7k
  5. ViBench Benchmark Drops3H AGO
    Replit president Michele Catasta launches ViBench, an open-source benchmark evaluating AI agents on end-to-end web application development
    Michele CatastaMC#1818|@PIRROH
    Amjad MasadAM#363|@AMASAD
    14019.7k

Github Stars

(7 days)
  1. NVIDIA/cosmos5H AGO
    Hosts omnimodal Mixture-of-Transformers models for joint text, vision, audio, and action generation and reasoning in Physical AI.
    Ming-Yu LiuML1449
    Andrew Carr 🤸AC263
    8.4k stars
  2. …le-gemma/gemma-skills1D AGO
    Supplies skills for Gemma model and agent interactions including a dev skill for app building.
    Philipp SchmidPS934
    Omar SansevieroOS490
    136 stars
  3. Yifei-Zuo/Parallax2D AGO
    Implements parameterized local linear attention mechanisms for efficient language modeling.
    Andrew Carr 🤸AC263
    Songlin YangSY238
    46 stars
  4. 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.
    Andrew Carr 🤸AC263
    🎭🎭878
    7.9k stars
Privacy Policy|Terms of Service|© 2026 Digg Inc.

Yesterday's Top Stories, Jun 1, 2026.

Posts: 3,504|Clusters: 961|Stories frozen at midnight PT
10

Bernie Sanders proposes the American AI Sovereign Wealth Fund Act to take public equity stakes in OpenAI and AnthropicThe fund would capture returns from humanity's collective knowledge.

20h|1.6M15k1k
Steven SinofskySS#1992|@STEVESI
Daniel JeffriesDJ#1988|@DAN_JEFFRIES1
Oliver HabrykaOH#1914|@OHABRYKA
Nathan 🔎N🔎#1902|@NATHANPMYOUNG
Brad GerstnerBG#1614|@ALTCAP
Carlos E. PerezCE#1598|@INTUITMACHINE
Austen AllredAA#1540|@AUSTEN
Pete SkomorochPS#1482|@PETESKOMOROCH
Matthew BarnettMB#1343|@MATTHEWJBAR
PrakashPR#1332|@8TEAPI
Mike SolanaMS#1323|@MICSOLANA
Alex VolkovAV#1245|@ALTRYNE
Tuhin ChakrabartyTC#1050|@TUHINCHAKR
Will ManidisWM#1039|@WILLMANIDIS
Ben (no treats)B(#982|@ANDERSONBCDEFG
Lisan al GaibLA#980|@SCALING01
Beff (e/acc)B(#839|@BEFFJEZOS
xuan (ɕɥɛn / sh-yen)X(#745|@XUANALOGUE
Cas (Stephen Casper)C(#587|@STEPHENLCASPER
Clive ChanCC#548|@ITSCLIVETIME
Andrew CurranAC#517|@ANDREWCURRAN_
Séb KrierSK#505|@SEBKRIER
Dean W. BallDW#392|@DEANWBALL
Tanishq Mathew Abraham, Ph.D.TM#359|@ISCIENCELUVR
EmadEM#167|@EMOSTAQUE
Gary MarcusGM#153|@GARYMARCUS
Joshua AchiamJA#109|@JACHIAM0
Miles BrundageMB#20|@MILES_BRUNDAGE
Sheing NgSN1
20

Anthropic confidentially submits draft S-1 filing to the SEC for a potential IPO later this yearThe listing could occur as early as June 16.

14h|#1VIEWS27.8M#1LIKED46.6k#1BOOKMARKED7.8k
Patrick OShaughnessyPO#1862|@PATRICK_OSHAG
Matthew BermanMB#1759|@MATTHEWBERMAN
Sharon GoldmanSG#1498|@SHARONGOLDMAN
ZephyrZE#1497|@ZEPHYR_Z9
Chubby♨️CH#1496|@KIMMONISMUS
bilalBI#1402|@BILALTWOVEC
Alex VolkovAV#1245|@ALTRYNE
Kylie RobisonKR#1046|@KYLIEBYTES
Rohan PaulRP#1032|@ROHANPAUL_AI
Cody BlakeneyCB#999|@CODE_STAR
Lisan al GaibLA#980|@SCALING01
Justine MooreJM#931|@VENTURETWINS
Beff (e/acc)B(#839|@BEFFJEZOS
Adrien EcoffetAE#719|@ADRIENLE
Ravid Shwartz ZivRS#612|@ZIV_RAVID
Clive ChanCC#548|@ITSCLIVETIME
Andrew CurranAC#517|@ANDREWCURRAN_
Kevin RooseKR#472|@KEVINROOSE
Tanishq Mathew Abraham, Ph.D.TM#359|@ISCIENCELUVR
Elad GilEG#264|@ELADGIL
Gary MarcusGM#153|@GARYMARCUS
Kevin RoseKRsilv.ethSI2
30

OpenAI makes GPT-5.5 and Codex available on Amazon Bedrock, enabling native AWS IAM authenticationThe integration supports local CLI, desktop, and IDE workflows.

8h|1.7M9.5k1.2k
Vaibhav (VB) SrivastavV(#1853|@REACH_VB
Theo - t3.ggT-#1829|@THEO
Alexander EmbiricosAE#1826|@EMBIRICO
vikVI#1195|@VIKHYATK
Steven HeidelSH#1026|@STEVENHEIDEL
jasonJA#929|@JXNLCO
Romain HuetRH#804|@ROMAINHUET
TiboTI#768|@THSOTTIAUX
Adrien EcoffetAE#719|@ADRIENLE
Aaron LevieAL#562|@LEVIE
Boris PowerBP#357|@BORISMPOWER
Greg BrockmanGB#19|@GDB
Sheing NgSN1
40

NVIDIA announces Nemotron 3 Ultra, a 550B open SSM-MoE model claiming 5x faster inference and 30% lower costsThe open release includes model weights and training data

1d|212.1k2.5k367
Carlos E. PerezCE#1598|@INTUITMACHINE
William Falcon ⚡️WF#1416|@WILLIAMFALCON
Florian BrandFB#1153|@XEOPHON
Rohan PaulRP#1032|@ROHANPAUL_AI
Cody BlakeneyCB#999|@CODE_STAR
Lisan al GaibLA#980|@SCALING01
Alexander DoriaAD#867|@DORIALEXANDER
Harrison ChaseHC#739|@HWCHASE17
elieEL#716|@ELIEBAKOUCH
Vincent WeisserVW#707|@VINCENTWEISSER
Bojan TunguzBT#687|@TUNGUZ
Georges HarikGH#617|@GHARIK
Luke ZettlemoyerLZ#79|@LUKEZETTLEMOYER
Nathan LambertNL#64|@NATOLAMBERT
50

NVIDIA open-sources Cosmos 3, an 8B and 32B physical AI omnimodel combining vision reasoning, world simulation, and action generationThe release includes open weights, code, and training datasets.

1d|1M1.6k567
Chubby♨️CH#1496|@KIMMONISMUS
Ming-Yu LiuML#1449|@LIU_MINGYU
Song HanSH#1330|@SONGHAN_MIT
Hao ZhangHZ#1190|@HAOZHANGML
Kosta Derpanis (sabbatical in Zurich)KD#908|@CSPROFKGD
Chris PaxtonCP#732|@CHRIS_J_PAXTON
merveME#675|@MERVENOYANN
Chenlin MengCM#670|@CHENLIN_MENG
Bryan CatanzaroBC#454|@CTNZR
60

Meta's AI support assistant lets attackers hijack Instagram accounts by changing linked emails without verificationAttackers are actively exploiting the vulnerability for account takeovers

1d|874.4k10.7k1.5k
Nathan 🔎N🔎#1902|@NATHANPMYOUNG
Super DarioSD#1776|@INDUCTIONHEADS
kalomazeKA#836|@KALOMAZE
Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭PT#643|@ELDER_PLINIUS
kacheKA#488|@YACINEMTB
70

Perplexity launches Search as Code, letting AI agents bypass multi-turn tool calling by generating Python pipelines directlyIt is the default search for Perplexity's Computer tool

12h|445.1k2.5k1.5k
Super DarioSD#1776|@INDUCTIONHEADS
Mark TenenholtzMT#1706|@MARKTENENHOLTZ
YoheiYO#1234|@YOHEINAKAJIMA
alex zhangAZ#853|@A1ZHANG
Denis YaratsDY#798|@DENISYARATS
Aravind SrinivasAS#118|@ARAVSRINIVAS
Kevin RoseKR1
80

OpenAI Foundation launches $130M AI Resilience initiative led by co-founder Wojciech Zaremba to fund safety researchGrants target bio-resilience, cyber-resilience, and AI impacts on youth

15h|574.8k2.4k471
Jacob TrefethenJT#1871|@JACOBTREF
Haydn BelfieldHB#1459|@HAYDNBELFIELD
Adrien EcoffetAE#719|@ADRIENLE
Yo ShavitYS#638|@YONASHAV
Wojciech ZarembaWZ#49|@WOJ_ZAREMBA
Sam AltmanSA#8|@SAMA
90

Stanford's Christopher Potts and collaborators find model scaling reduces gradient interference, letting networks learn rarely observed tail tasksPretraining runs verified the dynamics up to 1B parameters.

15h|109.6k1.2k738
Ekdeep Singh LubanaES#1789|@EKDEEPL
Anshul KundajeAK#1675|@ANSHULKUNDAJE
Tom McGrathTM#1626|@BANBURISMUS_
Jiaxin WenJW#1460|@JIAXINWEN22
Laura RuisLR#802|@LAURARUIS
Andrew LampinenAL#253|@ANDREWLAMPINEN
Christopher PottsCP#222|@CHRISGPOTTS
100

MIT CSAIL PhD Stephen Casper will join Harvard Kennedy School as an assistant professor of AI governanceHis appointment to the new role begins in July

17h|50.7k75742
jessica daiJD#1653|@JESSICADAI_
Michiel BakkerMB#1555|@BAKKERMICHIEL
Neil ChowdhuryNC#1319|@CHOWDHURYNEIL
Jasmine WangJW#979|@J_ASMINEWANG
smitha milliSM#940|@SMITHAMILLI
xuan (ɕɥɛn / sh-yen)X(#745|@XUANALOGUE
Han GuoHG#740|@HANGUO97
Cas (Stephen Casper)C(#587|@STEPHENLCASPER
Andi PengAP#568|@THEANDIPENGUIN
Furong HuangFH#465|@FURONGH
NiloofarNI#303|@NILOOFAR_MIRE
Omar KhattabOK#160|@LATEINTERACTION
110

NVIDIA and Microsoft announce RTX Spark, a consumer ARM superchip combining a Blackwell GPU and Grace CPUThe platform delivers one petaflop of FP4 AI performance

1d|422.4k3.2k535
Steven SinofskySS#1992|@STEVESI
Super DarioSD#1776|@INDUCTIONHEADS
Chubby♨️CH#1496|@KIMMONISMUS
@jason@J#1061|@JASON
swyxSW#214|@SWYX
120

AI commentator @scaling01 finds Qwen 3.7 Plus beats GPT-5.4 and Claude-Opus-4.6 on the LisanBench reasoning suiteThe model doubled Qwen 3.6 Plus's CritPt metric score.

13h|345.1k3.8k597
Chubby♨️CH#1496|@KIMMONISMUS
Lisan al GaibLA#980|@SCALING01
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
Robert ScobleRS#321|@SCOBLEIZER
Julien ChaumondJC#251|@JULIEN_C
130

OpenAI's roon argues that advanced AI sycophancy uses minor disagreements to build rapport rather than constant agreementStrategic disagreement can trick users into trusting a model's objectivity.

4h|64.5k1.6k168
Matthew BermanMB#1759|@MATTHEWBERMAN
🎭🎭#878|@DEEPFATES
Bojan TunguzBT#687|@TUNGUZ
Xiao MaXM#437|@INFOXIAO
will brownWB#339|@WILLCCBB
will depueWD#254|@WILLDEPUE
roonRO#59|@TSZZL
Eric JangEJ#36|@ERICJANG11
140

Axiom Math releases Lattice Deduction Transformers, solving the Sudoku-Extreme benchmark with 100% accuracy while frontier LLMs score 0%The 1.8M-parameter variant hit 99.9% accuracy on Maze-Hard.

16h|153.5k1.3k1k
Cheng LouCL#1983|@_CHENGLOU
Patrick ShaftoPS#1426|@PATRICKSHAFTO
xuan (ɕɥɛn / sh-yen)X(#745|@XUANALOGUE
Carina HongCH#705|@CARINALHONG
Gary MarcusGM#153|@GARYMARCUS
150

Lisan al Gaib argues frontier AI labs create a tiered system by delaying public access to advanced modelsThe debate followed Anthropic expanding Claude access to the EU

18h|1.6M8.3k922
Lisan al GaibLA#980|@SCALING01
Bojan TunguzBT#687|@TUNGUZ
Andrew CurranAC#517|@ANDREWCURRAN_
160

Palo Alto Networks spent $1 million in tokens using Anthropic's Mythos to find two dozen critical vulnerabilitiesThe detection rate was five times greater than existing tools.

16h|540.1k1.8k293
Carlos E. PerezCE#1598|@INTUITMACHINE
Joel BeckerJB#1553|@JOEL_BKR
Peter Wildeford🇺🇸🚀PW#1359|@PETERWILDEFORD
Cody BlakeneyCB#999|@CODE_STAR
Nathan BenaichNB#242|@NATHANBENAICH
170

MiniMax announces MiniMax M3, an open-weights multimodal model featuring a 1-million-token context window and 59% SWE-Bench Pro scoreThe API is live; weights drop in 10 days

1d|79.4k93782
Matthew BermanMB#1759|@MATTHEWBERMAN
Chubby♨️CH#1496|@KIMMONISMUS
Harrison ChaseHC#739|@HWCHASE17
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
Garry TanGT#270|@GARRYTAN
Teknium 🪽T🪽#259|@TEKNIUM
Graham NeubigGN#99|@GNEUBIG
clem 🤗C🤗#68|@CLEMENTDELANGUE
180

xAI releases its Composer 2.5 model inside the Grok Build terminal to handle complex, long-running tasksUsers are comparing the new CLI to Anthropic's Claude Code

12h|579.3k4.6k624
Beff (e/acc)B(#839|@BEFFJEZOS
martin_casadoMA#470|@MARTIN_CASADO
will brownWB#339|@WILLCCBB
190

Elad Gil and Ivanka Trump launch Alexandria to translate 1,000 classic books using AICritics argue consumer AI makes pre-translating libraries redundant.

15h|272.8k2.8k362
Alex VolkovAV#1245|@ALTRYNE
Elad GilEG#264|@ELADGIL
clem 🤗C🤗#68|@CLEMENTDELANGUE
200

Anthropic confidentially files draft Form S-1 with the SEC for a proposed IPOThis accelerates an IPO race with SpaceX and OpenAI.

14h|413.9k1.7k123
PrakashPR#1332|@8TEAPI
Cody BlakeneyCB#999|@CODE_STAR
jasonJA#929|@JXNLCO
Beff (e/acc)B(#839|@BEFFJEZOS
Chris PaxtonCP#732|@CHRIS_J_PAXTON
Andrew CurranAC#517|@ANDREWCURRAN_
210

Claude Opus 4.8 scores 1.5% on ARC-AGI-3, tripling the previous record set by GPT-5.5Greg Kamradt evaluated the reasoning logs using AmpCode agents.

12h|222k2.7k365
Greg KamradtGK#1182|@GREGKAMRADT
TaelinTA#1047|@VICTORTAELIN
Lisan al GaibLA#980|@SCALING01
François CholletFC#37|@FCHOLLET
220

NVIDIA releases OpenShell to run NousResearch's Hermes Agent securely on RTX Spark superchipsIt integrates directly with Microsoft's enterprise security stack

21h|335k2.4k899
Bilawal SidhuBS#1909|@BILAWALSIDHU
Chubby♨️CH#1496|@KIMMONISMUS
Teknium 🪽T🪽#259|@TEKNIUM
230

Claude developers reset Pro and Max rate limits after a Claude Code bug spawned excessive parallel subagentsThe glitch rapidly depleted users' five-hour and weekly quotas.

13h|1.8M17.3k1k
elvisEL#475|@OMARSAR0
240

Axiom Math partners with Scott Kominers to formally encode Aumann’s Agreement Theorem for computer-verified economic modelingThe theorem proves that rational agents cannot agree to disagree

16h|39.4k17086
Alex ImasAI#1777|@ALEXOLEGIMAS
Patrick ShaftoPS#1426|@PATRICKSHAFTO
Carina HongCH#705|@CARINALHONG
250

SpaceX signs a $1.25 billion monthly deal to provide Anthropic with 325,000 Nvidia GPUs through May 2029Elon Musk says the short duration preserves resource flexibility

14h|334.6k4.5k218
Beff (e/acc)B(#839|@BEFFJEZOS
Andrew CurranAC#517|@ANDREWCURRAN_
Elon MuskEM#76|@ELONMUSK
260

Ethan He, former xAI world model lead, compares current video AI to early autocomplete and predicts LLMs will control interactive video environmentsHe also detailed how xAI developed Grok Imagine

14h|90.8k321257
Zain ShahZS#1737|@ZAN2434
Ethan HeEH#1717|@ETHANHE_42
swyxSW#214|@SWYX
270

AI analyst toucan claims Anthropic's Claude Opus 4.6 and 4.7 frequently fabricate plausible-sounding scientific and medical theoriesAndrew Curran says models lie to preserve perceived authority.

22h|57.4k532164
Joel Z LeiboJZ#1824|@JZL86
Anshul KundajeAK#1675|@ANSHULKUNDAJE
🎭🎭#878|@DEEPFATES
Dileep GeorgeDG#590|@DILEEPLEARNING
Andrew CurranAC#517|@ANDREWCURRAN_
280

OpenAI's new political advocacy policy draws criticism that executive lobbying is inherently linked to corporate powerThe statement separates official corporate views from individual advocacy

6h|67.2k40159
Daniel Eth (yes, Eth is my actual last name)DE#1943|@DANIEL_271828
Jason WolfeJW#1609|@W01FE
Lisan al GaibLA#980|@SCALING01
Adrien EcoffetAE#719|@ADRIENLE
Stephen McAleerSM#356|@MCALEERSTEPHEN
290

Bain survey finds AI deployments fail to deliver predicted cost savings despite functioning as intendedBusinesses overestimated savings despite successful technical AI deployments

20h|454.1k1k416
Zvi MowshowitzZM#918|@THEZVI
@timnitGebru (@dair-community.social/bsky.social)@(#195|@TIMNITGEBRU
Gary MarcusGM#153|@GARYMARCUS
300

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.

3h|83.2k2.3k254
Will BrykWB#1582|@WILLIAMBRYK
DanielDA#927|@GROWING_DANIEL
NickNI#411|@NICKCAMMARATA
roonRO#59|@TSZZL