AI News

Today's Highlights

In case you missed it

Cognition, creator of Devin, raises $1 billion at a $26 billion valuation with $492 million run-rate

SW#704|@SCOTTWU46

RK#629|@RUSSELLJKAPLAN

KA#207|@KARINANGUYEN

#1 VIEWED

Anthropic stages unreleased Claude Opus 4.8 model in its desktop application's model selector

CH#1496|@KIMMONISMUS

LA#980|@SCALING01

AC#517|@ANDREWCURRAN_

#1 BOOKMARKED

109

Elon Musk claims SpaceX's custom C-based AI training framework is 10x faster than JAX on 220,000 GB300 GPUs

MP#906|@MPARAKHIN

EL#716|@ELIEBAKOUCH

KA#488|@YACINEMTB

LB#55|@GIFFMANA

RISING LIKES

348

Mathematicians disprove the sum-product conjecture for real numbers using insights from OpenAI's unit distance conjecture counterexample

PS#1426|@PATRICKSHAFTO

BB#133|@BOAZBARAKTCS

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

Top Stories

Posts:—|Clusters:—|Next Crawl:—

10

Elon Musk claims SpaceX's custom C-based AI training framework is 10x faster than JAX on 220,000 GB300 GPUsCritics questioned the JAX benchmark and choice of C.

1h|88.4k813#1BOOKMARKED99

BI#1402|@BILALTWOVEC

JS#1251|@JSUAREZ

DA#927|@GROWING_DANIEL

MP#906|@MPARAKHIN

EL#716|@ELIEBAKOUCH

KA#488|@YACINEMTB

LB#55|@GIFFMANA

20

Robin Hanson argues current AI models show little indication of suffering despite debates over machine consciousnessMillions of Claude instances are generated and terminated daily.

1h|7.7k7515

SH#1488|@HAMANDCHEESE

GL#1480|@GLEECH

AN#1409|@ARAN_NAYEBI

PR#1332|@8TEAPI

RH#887|@ROBINHANSON

BB#133|@BOAZBARAKTCS

30

Anthropic stages unreleased Claude Opus 4.8 model in its desktop application's model selectorThe staging suggests an imminent official product launch

1h|#1VIEWS95.2k#1LIKED1.3k70#1COMMENTS90

CH#1496|@KIMMONISMUS

LA#980|@SCALING01

AC#517|@ANDREWCURRAN_

41

Analysis argues AI advancements accelerate human problem-solving, citing post-AlphaGo Go performance and combinatorics breakthroughsHumans used AI-derived techniques to solve unit distance problems.

1h|13.8k15332

DJ#1988|@DAN_JEFFRIES1

LA#980|@SCALING01

EL#475|@OMARSAR0

BP#357|@BORISMPOWER

51

Mathematicians disprove the sum-product conjecture for real numbers using insights from OpenAI's unit distance conjecture counterexampleThe discovery validates Timothy Gowers' prediction of human-AI math collaboration

1h|21.3k33976

RISING LIKES348

PS#1426|@PATRICKSHAFTO

BB#133|@BOAZBARAKTCS

New github stars (48hrs)

self-evolving/repo341d

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

60

Systems engineer Yacine argues that reading LLM outputs that cannot be immediately verified is highly detrimentalHe exempts structured formats like code and Lean proofs.

53m|32.2k50164

CB#999|@CODE_STAR

KA#488|@YACINEMTB

71

SpAItial Releases API For Programmatic 3D World Generation From Text And Images

1h|7k5125

BS#1909|@BILAWALSIDHU

MN#1235|@MATTNIESSNER

RS#321|@SCOBLEIZER

81

Elon Musk says SpaceX's Colossus deal with Anthropic is a 180-day lease, refuting rumors of a multi-year contractThe lease includes a 90-day mutual cancellation path

53m|83.4k37841

RP#1032|@ROHANPAUL_AI

KA#488|@YACINEMTB

94

Reactor Raises $59M To Launch Infrastructure For Real-Time World Models

50m|10.1k6317

FASTEST CLIMBING+4

RP#1032|@ROHANPAUL_AI

RS#321|@SCOBLEIZER

101

Lindsay M. Smith and Matt Wiemann launch DiscoverPhysics to evaluate LLM agents on scientific experimentation and physical law discoveryThe pipeline tests if LLMs can design scientific experiments.

1h|1.1k218

PI#376|@PAVEL_IZMAILOV

AG#154|@ANDREWGWILS

112

Microsoft plans to unveil homegrown AI models at Build, including a cheaper coding model for GitHub CopilotThe launch follows usage limit issues with Anthropic's Claude.

53m|6k7511

CH#1496|@KIMMONISMUS

AC#517|@ANDREWCURRAN_

122

Will Depue proposes a distance-based training penalty for weight interpretability, which Ashwinee Panda critiques as heavy-handedStandard training typically scrambles weight matrices learning identity functions.

53m|2.9k3120

AP#1365|@PANDAASHWINEE

WD#254|@WILLDEPUE

131

Hugging Face's Lewis Tunstall open-sources Repo2RLEnv to automatically convert GitHub issues and pull requests into RL environmentsIt generates simulated software engineering environments from any GitHub repository

53m|2.1k2113

LT#950|@_LEWTUN

EL#716|@ELIEBAKOUCH

140

John Schulman joins academic AI assistant Refine.ink as an advisor to guide human-AI scientific collaborationThe platform helps academic authors and reviewers analyze papers.

1h|2.5k272

BG#1447|@BEN_GOLUB

JS#11|@JOHNSCHULMAN2

150

Jack Parker-Holder argues that the rapid pace of AI development compresses reputation-building and destruction timelines tenfoldThe commentary adapts Warren Buffett's famous quote on reputation.

53m|2.7k200

JP#619|@JPARKERHOLDER

160

UC Berkeley's Jitendra Malik is elected to the Royal Society for his foundational contributions to computer visionMalik also serves as a VP and scientist at Amazon.

40m|56770

JM#838|@JITENDRAMALIKCV

GG#665|@GEORGIAGKIOXARI

Github Stars

(7 days)

manaflow-ai/cmux3D 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
19.8k stars
self-evolving/repo1D 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-viz2D 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 27, 2026.

Posts: 3,689|Clusters: 850|Stories frozen at midnight PT

10

Cognition, creator of Devin, raises $1 billion at a $26 billion valuation with $492 million run-rateWeekly Devin sessions spiked at Citi and Mercedes-Benz

9h|793.9k3.6k593

BH#1941|@BENHYLAK

AG#1924|@ITSANDREWGAO

WA#1906|@WALDEN_YAN

PO#1862|@PATRICK_OSHAG

VV#1557|@VIPULVED

JW#1539|@WOLFEJOSH

BI#1402|@BILALTWOVEC

YP#1252|@YPATIL125

RP#1032|@ROHANPAUL_AI

AW#988|@ANDREWWHITE01

SW#704|@SCOTTWU46

RK#629|@RUSSELLJKAPLAN

KA#207|@KARINANGUYEN

20

ESM project founder Alex Rives open-sources ESMFold2 for protein structure prediction and therapeutic designScaling inference compute improved binder design success rates.

12h|469.2k2.7k1k

AP#1703|@ALEX_PEYS

AK#1675|@ANSHULKUNDAJE

BW#1661|@BOWANG87

JW#1539|@WOLFEJOSH

AL#1372|@ADAMLERER

AR#1087|@ALEXRIVES

AW#988|@ANDREWWHITE01

JP#216|@JPINEAU1

MM#151|@MMITCHELL_AI

30

Serena launches DeepSWE, a long-horizon software engineering benchmark that tests AI coding agents on complex, high-output developer tasksGPT-5.5 scored 70% compared to Mythos Preview's estimated 90%.

1d|270k2k552

MB#1759|@MATTHEWBERMAN

WH#1430|@NREHIEW_

SH#1269|@SHOYER

RP#1032|@ROHANPAUL_AI

CB#999|@CODE_STAR

PS#934|@_PHILSCHMID

JY#854|@JYANGBALLIN

PS#495|@STEIPETE

OP#72|@OFIRPRESS

LB#55|@GIFFMANA

40

Genesis AI launches Genesis World 1.0, an open-source robotics simulator that turns one hour of real-world testing into 100 simulation daysIt features GPU-accelerated physics solvers and path-traced rendering

7h|218.4k960483

JL#1874|@JACKYLIANG42

VK#1450|@VIKASHPLUS

AN#1409|@ARAN_NAYEBI

CG#892|@GAN_CHUANG

CP#732|@CHRIS_J_PAXTON

DC#428|@DCHAPLOT

VK#327|@VKHOSLA

RS#321|@SCOBLEIZER

ES#320|@ERICSCHMIDT

50

Trajectory, co-founded by Ronak Malde, launches with $15M to build a continual learning platform for agentic AI modelsIt uses real-time user corrections to post-train deployed models.

10h|1.5M2k849

LR#1176|@LENNYSAN

NW#1155|@NEAL_WU

RP#1032|@ROHANPAUL_AI

WK#609|@WILLKNIGHT

RS#321|@SCOBLEIZER

SG#219|@SARANORMOUS

NL#64|@NATOLAMBERT

60

Sakana AI releases DiffusionBlocks to train neural networks one block at a time, cutting training memory up to 8xIt treats the network's forward pass as a diffusion-like denoising process.

10h|737.8k5.2k3.7k

BW#1661|@BOWANG87

FH#465|@FURONGH

HP#455|@HYHIEU226

FF#331|@FRANCOISFLEURET

HA#18|@HARDMARU

70

Spark Capital's Yasmin Razavi joins the Forbes Midas List after her $75 million Anthropic investment returned $3 billionShe backed Anthropic when the startup had no revenue

13h|310.9k1.1k253

YR#1602|@YASMINRAZAVI

IZ#1273|@IVANHZHAO

FR#1200|@FRASER

EG#264|@ELADGIL

RA#83|@_AROHAN_

80

OpenAI Foundation launches $250 million initiative to study and support workforce transitions in the age of AIThe program funds impact measurement and shared prosperity research.

8h|360k4k736

DS#1915|@DIVYASIDDARTH

JS#1232|@JASMINEWSUN

RP#1032|@ROHANPAUL_AI

WZ#49|@WOJ_ZAREMBA

90

Axiom Math's AxiomProver becomes the first AI system to have formal proofs accepted by peer-reviewed mathematics journalsOne accepted paper credits AxiomProver for automating its proofs

1d|148.2k815223

PS#1426|@PATRICKSHAFTO

DA#1271|@DAMEKDAVIS

AM#843|@ANDREWDAI

CH#705|@CARINALHONG

👩‍💻P#271|@DYNAMICWEBPAIGE

DP#197|@DIMITRISPAPAIL

100

Midjourney founder David Holz argues cheap FLOPS favor diffusion models, while Emad Mostaque proposes a hybrid training-to-inference workflowThe method converts autoregressive weights to diffusion for inference.

4h|124.3k1.2k482

B(#839|@BEFFJEZOS

CC#548|@ITSCLIVETIME

NR#440|@NAVEENGRAO

EM#167|@EMOSTAQUE

DA#156|@DAVIDSHOLZ

SE#129|@STEFANOERMON

110

Robinhood founder Vlad Tenev launches Agentic Accounts, allowing autonomous AI agents to execute stock trades automaticallyCritics compare the AI capability to traditional algorithmic trading.

10h|914k4.5k1.1k

ND#1894|@NICKADOBOS

AS#1054|@ANDERSSANDBERG

120

OpenAI, Thrive, and Crete deploy Codex-powered Tax AI, cutting preparation times by one-third with up to 97% accuracyThe system increased tax firm throughput by 50%.

10h|137.2k886281

BI#1402|@BILALTWOVEC

LI#1076|@THESEPHIST

RP#1032|@ROHANPAUL_AI

BP#357|@BORISMPOWER

130

Google's AI Overview fails basic spelling queries, claiming 'Google' has two 'P'sThe system analyzed an AI safety researcher's username instead

11h|#1VIEWS1.6M#1LIKED119.8k#1BOOKMARKED4.7k

TH#1499|@VOOOOOOGEL

140

LlamaIndex releases LiteParse v2.0, a Rust rewrite claiming up to 100x faster document parsingA custom WebAssembly package enables browser and edge runtimes.

8h|505.6k2.9k2.9k

JL#677|@JERRYJLIU0

150

OpenAI releases secure tunnels to connect private Model Context Protocol servers without modifying inbound firewallsOutbound-only HTTPS tunnels secure the connection to OpenAI.

6h|484.5k3k1.8k

SH#1026|@STEVENHEIDEL

160

OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei walk back warnings of an AI jobs apocalypse ahead of IPOsGary Marcus accused the CEOs of protecting their valuations.

10h|587.4k7.8k504

SS#1992|@STEVESI

GM#153|@GARYMARCUS

170

Roboflow integrates RF-DETR into Hugging Face Transformers, claiming it outperforms YOLO on detection and segmentationThe integration supports low-VRAM fine-tuning on constrained hardware.

10h|97.9k1.4k1.2k

PS#1482|@PETESKOMOROCH

ME#675|@MERVENOYANN

C🤗#68|@CLEMENTDELANGUE

180

Claude Code updates its CLI tool to resolve TUI rendering and screen-flickering bugsTaelin says the tool was inspired by his projects

6h|1.1M9.2k2.2k

TA#1047|@VICTORTAELIN

190

Runway launches Model Context Protocol integration to bring its video and image models directly into Replit, Cursor, and ClaudeReplit founder Amjad Masad highlighted the developer workflow integration.

10h|257.7k1.3k776

YO#1234|@YOHEINAKAJIMA

CV#816|@C_VALENZUELAB

200

ETH Zürich's Tiberiu Mușat proves that fixed-precision neural network weight norm is equivalent to Kolmogorov complexityThe equivalence explains why deep networks generalize rather than memorize.

15h|146.2k1.2k1.2k

LA#1822|@LUCAAMB

PM#713|@PMINERVINI

TC#120|@TACOCOHEN

210

Raindrop AI co-founder Ben Hylak launches howtoeval.com, a practical guide for assessing production AI agentsAn interactive diagnostic quiz helps builders identify agent deployment risks.

7h|69.3k9871.2k

BH#1941|@BENHYLAK

HH#1277|@HAMELHUSAIN

220

Pavel Izmailov and Andrew Gordon Wilson find LLMs can prevent catastrophic forgetting using self-generated replay dataThe method kept task evaluation accuracies near 100%.

10h|51.4k710491

AK#1361|@AKBIRKHAN

T(#420|@TEORTAXESTEX

PI#376|@PAVEL_IZMAILOV

AG#154|@ANDREWGWILS

230

Thibault Sottiaux, OpenAI Codex engineering lead, says GPT-5.2 and GPT-5.3-Codex will be sunset from Codex on June 2Both models will remain accessible to developers via API

10h|649.9k5.8k338

CH#1496|@KIMMONISMUS

RO#1220|@KRISHNANROHIT

TI#768|@THSOTTIAUX

240

Rishi Bommasani and Stanford HAI researchers find AI hiring tools violate civil rights standards by disproportionately rejecting Black and Asian applicantsSingle-vendor screening risks blocking candidates across multiple employers

1d|11.1k7430

SK#1085|@SANMIKOYEJO

L(#967|@LCHOSHEN

SK#734|@SAYASHK

RI#223|@RISHIBOMMASANI

DY#134|@DIYI_YANG

250

Rob Wiblin and Toby Ord argue the Vatican's first AI encyclical avoids ruling out machine consciousnessParagraph 99 of the document denies AI moral conscience.

2d|21.3k22738

CF#1356|@CFGEEK

TO#1159|@TOBYORDOXFORD

RW#1115|@ROBERTWIBLIN

CP#1092|@CHRISPAINTERYUP

DW#392|@DEANWBALL

260

Apollo CEO Marc Rowan argues AI's massive capital demands will outgrow venture equity and rely on private creditApollo manages more than $1 trillion in assets.

10h|164.6k358394

SW#1814|@SARAHDINGWANG

ET#1253|@ERIKTORENBERG

WM#1039|@WILLMANIDIS

270

Snowflake CEO Sridhar Ramaswamy reports 34% product revenue growth to $1.33 billion, sparking a 30% stock spikeOver 13,600 accounts have adopted the company's AI offerings.

4h|763.1k5.8k472

SR#1898|@RAMASWMYSRIDHAR

ZE#1497|@ZEPHYR_Z9

280

Y Combinator's Jared Friedman says giving AI agents unrestricted database access increased utility tenfoldThe internal system integrates more than 350 tools.

9h|1.4M2.8k1.8k

290

Google Gemma launches a new Coral development board featuring a Gemma3-270M chip for local AI inferenceThe board enables real-time voice control and speech translation.

4h|196.7k2.7k1.3k

NR#1562|@NATANIELRUIZG

OS#490|@OSANSEVIERO

300

DSPy creator Omar Khattab argues on-policy distillation is structurally inefficient because it forces teacher models into a passive roleHe proposes Pedagogical Reinforcement Learning as an active alternative.

4h|42.9k446309

SA#1262|@SAMSJA19

HB#1022|@HERBIEBRADLEY

GR#996|@GRAD62304977

OK#160|@LATEINTERACTION