AI News

Today's Highlights

In case you missed it

Pope Francis argues AI cannot truly understand human concepts, sparking debate among researchers over embodied AI and machine comprehension

DW#392|@DEANWBALL

⿻A#361|@IAMTRASK

FF#331|@FRANCOISFLEURET

GM#153|@GARYMARCUS

AS#118|@ARAVSRINIVAS

#1 VIEWED

OpenAI's Thibault Sottiaux teases a GPT-5 roadmap where sequential decimal increments through GPT-5.5 deliver efficiency and speed gains

CH#1496|@KIMMONISMUS

TI#768|@THSOTTIAUX

ER#306|@ERICMITCHELLAI

#1 BOOKMARKED

893

Peter Steinberger, OpenClaw project steward, details a workflow using GPT-5.5 and Crabbox to scale agent execution up to ten hours

🍓🍓#1711|@IRULETHEWORLDMO

PS#495|@STEIPETE

It’s a new day at Digg. Fresh stories are clustering now. 80% complete.

Top Stories

Posts:—|Clusters:—|Next Crawl:—

10

An overheard anecdote claims an incoming professor plans to spend her $1.5 million fund on ChatGPT tokens and undergraduatesThe strategy aims to outperform traditional graduate-led research workflows

8h|121.2k1.3k175

SG#1498|@SHARONGOLDMAN

TY#233|@TYPEDFEMALE

TL#224|@TALLINZEN

20

OpenAI's Thibault Sottiaux teases a GPT-5 roadmap where sequential decimal increments through GPT-5.5 deliver efficiency and speed gainsThe efficiency gains target costs in long-running agentic workflows

8h|#1VIEWS240.5k#1LIKED3.2k179#1COMMENTS397

CH#1496|@KIMMONISMUS

TI#768|@THSOTTIAUX

ER#306|@ERICMITCHELLAI

30

Peter Steinberger, OpenClaw project steward, details a workflow using GPT-5.5 and Crabbox to scale agent execution up to ten hoursThe multi-hour runs significantly increase confidence in production-ready output.

5h|87.2k1.6k#1BOOKMARKED893

RISING LIKES1.6k

🍓🍓#1711|@IRULETHEWORLDMO

PS#495|@STEIPETE

Watchlist

Video Signals

T🪽#259|@TEKNIUM

Nous Research Adds Tool Search to Hermes Agent to Cut Context Bloat

17h ago|Views 915KLikes 2.7KBookmarks 882

LB#975|@REMEMBERLENNY

RA#336|@RAPHA_GL

Drafted.ai Raises $16M Seed to Build AI Models for Home Design

23h ago|Views 162KLikes 295Bookmarks 251

GT#270|@GARRYTAN

Proactive AI Assistant 'Jo' Completes Tasks Before User Prompts

1d ago|Views 342KLikes 274Bookmarks 206

Meme Video Pleads With Dario Amodei For Codex Over Claude Opus

6h ago|Views 30KLikes 179Bookmarks 22

T🪽#259|@TEKNIUM

Hermes Agent Adds Tool Search to Cut Context Bloat

3h ago|Views 67KLikes 0Bookmarks 0

GK#1182|@GREGKAMRADT

Claude Opus 4.8 Falls to Seventh on SnakeBench After Loss to GPT-5.5

2h ago|Views 5.7KLikes 33Bookmarks 6

KD#908|@CSPROFKGD

ETH Students Demo Voice-Controlled Robot Arm Using Frozen Vision Models

20h ago|Views 16KLikes 62Bookmarks 28

DW#392|@DEANWBALL

Duke Lab Unveils Argus Spherical Robot With Dynamic Symmetry

2h ago|Views 162KLikes 42Bookmarks 10

RP#1032|@ROHANPAUL_AI

Dan Loeb Says Nvidia At 15X 2027 Earnings Is Not Pricey

8h ago|Views 14KLikes 31Bookmarks 7

EL#475|@OMARSAR0

Builder Adopts HTML Artifacts to Verify Complex AI Agent Workflows

26m ago|Views 978Likes 16Bookmarks 15

40

AVERI's Miles Brundage says deploying US open-weight models took hours compared to seamless Chinese alternativesFlorian Brand suggested OpenRouter to simplify the deployment process.

5h|8.7k603

FB#1153|@XEOPHON

KW#1122|@KYLE_L_WIGGERS

MB#20|@MILES_BRUNDAGE

50

Open-source AI builder @xlr8harder argues model distillation is more effective than crowdsourcing for resolving underlying technical problemsResearch engineer Florian Brand confirmed the assessment.

8h|895283

XL#1674|@XLR8HARDER

FB#1153|@XEOPHON

WB#339|@WILLCCBB

New github stars (48hrs)

run-llama/liteparse7.2k22h

Extracts spatial text with bounding boxes from PDFs and office docs via PDFium plus optional OCR, outputting JSON or layout-preserved text.

60

Frontier AI labs spend $10 billion to $15 billion each on training data as high-quality tasks reach $20,000Few vendors can deliver specialized data for frontier models.

8h|10k9039

FB#1153|@XEOPHON

T(#420|@TEORTAXESTEX

70

Florian Brand of Prime Intellect argues OpenAI Codex can legitimately achieve 50% accuracy on the GSM8K math reasoning benchmarkDimitris Papailiopoulos clarified the 50% metric was a target.

2h|927160

AJ#1352|@ALEXJC

FB#1153|@XEOPHON

DP#197|@DIMITRISPAPAIL

80

Frontier AI labs are reportedly spending $10 billion to $15 billion each on training data acquisitionComplex, high-quality task datasets can cost up to $500,000.

8h|5.3k3611

XL#1674|@XLR8HARDER

EM#167|@EMOSTAQUE

90

Pierre-Carl Langlais, Pleias CTO, argues AI labs deploy token-heavy features that conflict with traditional SaaS modelsA viral meme joked that "Big Token" invented next-token generation

7h|2.5k264

AL#1114|@ALTH0U

AD#867|@DORIALEXANDER

100

Bilal Khan highlights the irony of OpenAI promoting supermarket Loblaw using Codex despite its past bread price-fixing scandalLoblaw secretly conspired to inflate bread prices from 2001 to 2015.

3h|1.9k131

BI#1402|@BILALTWOVEC

CB#999|@CODE_STAR

110

Dimitris Papailiopoulos of Microsoft Research AI Frontiers limits solver challenge models to 10 million trainable weightsThe update bans access to the evaluation test set

2h|50451

AJ#1352|@ALEXJC

DP#197|@DIMITRISPAPAIL

120

Open-source builder xlr8harder teases Flash Attention rebuild, as research engineer Florian Brand highlights GPU waste from missing Python wheelsMissing pre-built Python wheels frequently stall transformer deployment.

8h|2.1k390

XL#1674|@XLR8HARDER

FB#1153|@XEOPHON

130

Needle in a Haystack creator Greg Kamradt proposes a curated San Francisco meetup on agentic engineeringThe selective gathering will limit attendance to 30 people

1h|1k110

GK#1182|@GREGKAMRADT

C🎉#863|@CHARLES_IRL

140

Teortaxes argues Deli Chen's AI paper-summarization workflow relies too heavily on static knowledge instead of fresh sourcesChen used AI agents to summarize over 100 papers.

8h|39100

DC#1523|@VICTOR207755822

T(#420|@TEORTAXESTEX

150

Alex J. Champandard argues Recursive Language Models lack novelty, debating Grad over whether the paradigm differs fundamentally from ClaudeGrad defines RLMs via subagents and programmatic tool calling

8h|10600

AJ#1352|@ALEXJC

GR#996|@GRAD62304977

160

CreativeAI co-founder Alex J. Champandard argues Anthropic's workflows feature shifts away from Recursive Language ModelsDeveloper Grad claims RLM concepts yield substantial performance advantages

7h|3920

AJ#1352|@ALEXJC

GR#996|@GRAD62304977

Github Stars

(7 days)

manaflow-ai/cmux5D 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
19.8k stars
run-llama/liteparse22H AGO 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
7.2k stars
self-evolving/repo3D AGO AGO
Invokes AI agents via GitHub mentions to implement issues, review PRs, fix branches, and persist memory across runs.
594
878
34 stars
modaic-ai/gepa-viz4D AGO AGO
Renders GEPA prompt-optimization candidate trees as interactive force-directed graphs with live updates from a Python callback.
878
793
269 stars

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

Yesterday's Top Stories, May 29, 2026.

Posts: 3,438|Clusters: 2,458|Stories frozen at midnight PT

10

Pope Francis argues AI cannot truly understand human concepts, sparking debate among researchers over embodied AI and machine comprehensionCMU's Aran Nayebi says robotics will bridge this gap.

19h|9.4M290.7k21.1k

SD#1776|@INDUCTIONHEADS

CG#1679|@CGARCIAE88

XL#1674|@XLR8HARDER

CE#1598|@INTUITMACHINE

TH#1499|@VOOOOOOGEL

CH#1496|@KIMMONISMUS

SH#1488|@HAMANDCHEESE

AN#1409|@ARAN_NAYEBI

BI#1402|@BILALTWOVEC

CF#1356|@CFGEEK

PR#1332|@8TEAPI

JN#1260|@JVNIXON

RO#1220|@KRISHNANROHIT

AL#1114|@ALTH0U

CP#1092|@CHRISPAINTERYUP

JM#931|@VENTURETWINS

JA#837|@APPLES_JIMMY

AE#719|@ADRIENLE

AC#517|@ANDREWCURRAN_

XM#437|@INFOXIAO

DW#392|@DEANWBALL

⿻A#361|@IAMTRASK

FF#331|@FRANCOISFLEURET

GM#153|@GARYMARCUS

AS#118|@ARAVSRINIVAS

20

Pope Francis claims AI fundamentally lacks physical bodies and moral conscience, drawing pushback from CMU's Aran NayebiNayebi predicts embodied AI will mature within 20 years.

19h|9.4M#1LIKED290.8k#1BOOKMARKED21.1k

SD#1776|@INDUCTIONHEADS

CG#1679|@CGARCIAE88

XL#1674|@XLR8HARDER

CE#1598|@INTUITMACHINE

TH#1499|@VOOOOOOGEL

CH#1496|@KIMMONISMUS

SH#1488|@HAMANDCHEESE

AN#1409|@ARAN_NAYEBI

BI#1402|@BILALTWOVEC

CF#1356|@CFGEEK

PR#1332|@8TEAPI

JN#1260|@JVNIXON

RO#1220|@KRISHNANROHIT

AL#1114|@ALTH0U

CP#1092|@CHRISPAINTERYUP

JM#931|@VENTURETWINS

JA#837|@APPLES_JIMMY

AE#719|@ADRIENLE

XM#437|@INFOXIAO

DW#392|@DEANWBALL

⿻A#361|@IAMTRASK

FF#331|@FRANCOISFLEURET

GM#153|@GARYMARCUS

AS#118|@ARAVSRINIVAS

30

OpenAI brings Codex "Computer Use" to Windows with mobile and Mac remote controlThe agent operates Windows applications by clicking and typing.

12h|1.6M13.7k2k

V(#1853|@REACH_VB

AE#1826|@EMBIRICO

CH#1496|@KIMMONISMUS

MP#906|@MPARAKHIN

RH#804|@ROMAINHUET

LK#627|@LUKASZKAISER

BP#357|@BORISMPOWER

40

OpenAI releases Codex v26.527, enabling Windows computer use and remote desktop control from mobile devicesA new profile interface tracks token activity and usage statistics.

12h|1.4M12.7k1.9k

V(#1853|@REACH_VB

AE#1826|@EMBIRICO

CH#1496|@KIMMONISMUS

MP#906|@MPARAKHIN

RH#804|@ROMAINHUET

LK#627|@LUKASZKAISER

BP#357|@BORISMPOWER

50

Brilliant founder Sue Khim launches Koji, an interactive AI tutor that coaches children through math and coding without giving direct answersIt personalizes lessons using a decade of student learning data

15h|4.5M16.6k13k

JW#1460|@JIAXINWEN22

RP#1032|@ROHANPAUL_AI

NL#1006|@LABENZ

CP#747|@CHAMATH

RS#321|@SCOBLEIZER

60

Brilliant launches Koji, an interactive AI tutor designed to guide students instead of giving direct answersIt targets topics in geometry, algebra, and coding.

15h|4.5M16.6k13k

JW#1460|@JIAXINWEN22

RP#1032|@ROHANPAUL_AI

NL#1006|@LABENZ

CP#747|@CHAMATH

RS#321|@SCOBLEIZER

70

Nvidia and Microsoft tease 'A new era of PC' with coordinates pointing to a May 2026 Taipei eventThe teaser sparked speculation of Nvidia Windows on ARM.

14h|8.2M30.1k3.1k

SS#1992|@STEVESI

RC#1841|@REMICADENE

CH#1496|@KIMMONISMUS

AC#517|@ANDREWCURRAN_

80

A16z partner Justine Moore posts a parody about using a hypothetical "Claude Opus 4.8" model to rename a single fileThe joke satirizes using massive AI models for trivial tasks.

1d|#1VIEWS31.5M236.6k14.9k

ST#1446|@_ONIONESQUE

B(#839|@BEFFJEZOS

T🪽#259|@TEKNIUM

EM#76|@ELONMUSK

90

a16z partner Justine Moore jokes about using a hypothetical Claude Opus 4.8 model to rename a fileThe parody mocks using massive compute for trivial tasks

1d|31.5M236.6k14.9k

ST#1446|@_ONIONESQUE

B(#839|@BEFFJEZOS

T🪽#259|@TEKNIUM

EM#76|@ELONMUSK

100

NVIDIA and Microsoft tease a "new era of PC" with coordinates pointing to TaipeiThe synchronized posts featured identical coordinates pointing to Taiwan.

14h|7.4M26.4k2.7k

RC#1841|@REMICADENE

CH#1496|@KIMMONISMUS

AC#517|@ANDREWCURRAN_

110

Codex launches self-managing workspaces, letting AI agents autonomously organize chat threads and spin up parallel worktreesFile-based architecture allows agents to modify chat metadata directly.

9h|678.9k4.5k2.5k

ND#1894|@NICKADOBOS

V(#1853|@REACH_VB

DS#1410|@DANSHIPPER

TI#768|@THSOTTIAUX

120

Codex update allows AI agents to autonomously manage workspace state and spin up parallel worktrees by editing chat filesAgents modify file metadata to configure the application's interface.

9h|678.9k4.5k2.5k

ND#1894|@NICKADOBOS

V(#1853|@REACH_VB

DS#1410|@DANSHIPPER

TI#768|@THSOTTIAUX

130

Blue Origin's New Glenn rocket explodes during a static fire test at Launch Complex-36Video captured a massive fireball engulfing the launch pad.

1d|1.7M7.5k834

CH#1496|@KIMMONISMUS

RP#1032|@ROHANPAUL_AI

ZN#537|@ZACHARYNADO

140

UCLA mathematician Terence Tao says AI automation of routine tasks lets researchers test unconventional, high-risk hypothesesHe uses AI for case verification and literature search.

10h|382k4.4k1.1k

RP#1032|@ROHANPAUL_AI

MC#566|@MMMBCHANG

SB#51|@SEBASTIENBUBECK

150

UCLA mathematician Terence Tao says AI tools reduce the cognitive friction of routine research to near zeroThis allows scientists to pursue high-risk, unconventional hypotheses

10h|382k4.4k1.1k

RP#1032|@ROHANPAUL_AI

MC#566|@MMMBCHANG

SB#51|@SEBASTIENBUBECK

160

OpenAI launches Rosalind Biodefense, giving U.S. government agencies access to its specialized GPT-Rosalind modelThe initiative focuses on defending against engineered pathogens.

20h|394.6k2.4k396

JK#586|@JASONKWON

AC#517|@ANDREWCURRAN_

BB#133|@BOAZBARAKTCS

170

OpenAI launches Rosalind Biodefense initiative, restricting its specialized GPT-Rosalind model to government agenciesAllied public health partners will also receive model access.

20h|394.6k2.4k396

JK#586|@JASONKWON

AC#517|@ANDREWCURRAN_

BB#133|@BOAZBARAKTCS

180

Michael Rabinovich evaluates Opus 4.8 on 2D-to-3D CAD tasks, finding unexpected performance gains over previous versionsThe evaluation tracked metrics across three mechanical parts

15h|502.8k2.7k1.5k

T(#420|@TEORTAXESTEX

TW#17|@THOM_WOLF

190

Epoch AI finds open-weight AI models lag behind proprietary systems by an average of four monthsThe lag represents an eight-point deficit on Epoch's index.

10h|99.6k734162

CH#1496|@KIMMONISMUS

FB#1153|@XEOPHON

SR#604|@CLONEOFSIMO

T(#420|@TEORTAXESTEX

GM#153|@GARYMARCUS

MB#20|@MILES_BRUNDAGE

200

Epoch AI finds open-weight models lag proprietary state-of-the-art AI systems by about four monthsThe performance gap has remained steady since January 2023.

10h|99.6k734162

CH#1496|@KIMMONISMUS

FB#1153|@XEOPHON

SR#604|@CLONEOFSIMO

T(#420|@TEORTAXESTEX

GM#153|@GARYMARCUS

MB#20|@MILES_BRUNDAGE

210

Gary Marcus argues OpenAI faces WeWork-like collapse and Anthropic relied on a SpaceX subsidyFounder Beff (e/acc) challenged the claims, defending AI viability.

1d|259.3k1.7k372

BR#1622|@BINDUREDDY

LA#980|@SCALING01

TK#267|@TEJASDKULKARNI

GM#153|@GARYMARCUS

220

Amazon shuts down an internal AI adoption leaderboard after employees gamed the system to inflate usage metricsAnthropic previously rejected building a similar token-maximizing leaderboard

1d|1.1M7.1k868

ET#1884|@TORCHCOMPILED

CE#1598|@INTUITMACHINE

AV#1245|@ALTRYNE

230

Amazon shuts down internal AI adoption leaderboard after employees gamed tracking metricsAnthropic previously rejected token leaderboards to avoid employee gaming.

1d|1.1M7.1k868

ET#1884|@TORCHCOMPILED

CE#1598|@INTUITMACHINE

AV#1245|@ALTRYNE

240

Internet Restored In Iran Exposes Hidden Protest Crackdown

17h|1.2M43.4k1.6k

JW#1539|@WOLFEJOSH

250

Keshigeyan Chandrasegaran and Kyle Sargent launch GPIC, a permissive image-text dataset and benchmark for training visual modelsThe 28-trillion-pixel corpus is fully permissive for commercial use.

14h|110.2k516242

JY#921|@KOHJINGYU

JJ#403|@JCJOHNSS

SD#80|@SEDIELEM

FL#12|@DRFEIFEI

260

Keshigeyan Chandrasegaran and Kyle Sargent launch GPIC, a permissively licensed 100-million image-text dataset and benchmarkThe dataset spans 28 trillion pixels on Hugging Face.

14h|110.2k516242

JY#921|@KOHJINGYU

JJ#403|@JCJOHNSS

SD#80|@SEDIELEM

FL#12|@DRFEIFEI

270

Founders Fund's Mike Solana proposes using gene drive technology to drive mosquitoes and ticks to extinctionElon Musk signaled support for the genetic engineering proposal

17h|1.1M18.1k315

MS#1323|@MICSOLANA

HE#572|@HEINRICHKUTTLER

EM#76|@ELONMUSK

280

Chroma founder Jeff Huber calls legal AI uninspiring, prompting Marc Andreessen to argue it empowers individuals against large institutionsAcademic Peter Henderson says AI will democratize impact litigation.

6h|224.8k1.8k222

NL#1006|@LABENZ

PH#636|@PETERHNDRSN

290

Marc Andreessen argues legal AI democratizes litigation against institutions, countering Chroma founder Jeff Huber's claim that the sector is uninspiringPeter Henderson says AI enables citizen-led impact litigation.

6h|224.8k1.8k222

NL#1006|@LABENZ

PH#636|@PETERHNDRSN

300

Gary Marcus argues OpenAI and Anthropic face financial instability, comparing OpenAI's business model to WeWorkMarcus claims Anthropic was bolstered by a SpaceX subsidy.

1d|222.8k1.4k334

LA#980|@SCALING01

TK#267|@TEJASDKULKARNI

GM#153|@GARYMARCUS