Digg - AI news, before it trends

Today's Highlights

In case you missed it

OpenAI reports its internal general-purpose reasoning model found new point configurations for the Erdős planar unit distance problem that outperform prior square-grid optima

SB#51|@SEBASTIENBUBECK

CS#43|@CHRSZEGEDY

EJ#36|@ERICJANG11

NB#31|@POLYNOAMIAL

#1 VIEWED

DeepSeek makes its 75% discount on the DeepSeek-V4-Pro API permanent, keeping rates at one-quarter of prior levels for input and output tokens beyond the prior May 31, 2026 cutoff.

CH#1496|@KIMMONISMUS

LA#980|@SCALING01

EL#475|@OMARSAR0

YJ#427|@YUCHENJ_UW

T(#420|@TEORTAXESTEX

DR#103|@DELIPRAO

RISING LIKES

199

Gemini Omni performed native video edits on the 1896 Lumiere film of a steam train, converting it into versions with a bullet train, LEGO styling, Muppets, and a time traveler while preserving reflections.

NR#1562|@NATANIELRUIZG

EM#175|@EMOLLICK

FASTEST CLIMBING

Trump administration closes loophole that allowed O-1 visa holders including advanced researchers at OpenAI to remain in the US during green card applications

RH#373|@REIDHOFFMAN

Top Stories

Posts:—|Clusters:—|Next Crawl:—

10

DeepSeek makes its 75% discount on the DeepSeek-V4-Pro API permanent, keeping rates at one-quarter of prior levels for input and output tokens beyond the prior May 31, 2026 cutoff. — The 1.6-trillion-parameter model offers 1M-token context and fixed per-million pricing.

4h|#1VIEWS679.3k#1LIKED9k#1BOOKMARKED1.5k#1COMMENTS1.1k

DC#1523|@VICTOR207755822

ZE#1497|@ZEPHYR_Z9

CH#1496|@KIMMONISMUS

LA#980|@SCALING01

EL#475|@OMARSAR0

YJ#427|@YUCHENJ_UW

T(#420|@TEORTAXESTEX

DR#103|@DELIPRAO

20

Polsia, which operates businesses autonomously through AI, raised $30 million at a $250 million valuation while running with zero employees besides founder Ben Cera — Platform nears $10 million annual run rate with over 7600 users.

2h|326.9k1.9k1.4k

CH#1496|@KIMMONISMUS

RS#321|@SCOBLEIZER

30

MIT CSAIL PhD student Ryan Bahlous-Boldi introduces Vector Policy Optimization, a reinforcement learning method that maximizes vector-valued rewards to preserve distinct objectives in LLM post-training — VPO raises pass@k scores on LiveCodeBench over GRPO baselines.

2h|36.3k369310

LR#802|@LAURARUIS

HG#740|@HANGUO97

OK#160|@LATEINTERACTION

40

Reiner Pope builds a multiply-accumulate unit from logic gates and covers systolic arrays, FPGAs, ASICs, and brain architecture in a timestamped blackboard lecture — Dwarkesh Patel shared the lecture spanning 0:00:00 to 1:11:49.

1h|69.1k1.3k1.3k

LA#980|@SCALING01

DP#70|@DWARKESH_SP

50

OpenAI maintains a GitHub repository at openai/skills that supplies extensions for its Codex coding tool and enables automatic capability detection from targeted prompts — Jason Liu on the Codex team outlined the installation process.

4h|35.9k663590

ND#1894|@NICKADOBOS

New github stars (48hrs)

st3v3nmw/obsidian-spaced-repetition2.4k11h

Schedules spaced-repetition reviews for flashcards and notes inside Obsidian.

aisa-group/InferenceBench161d

Evaluates autonomous AI agents optimizing LLM serving like vLLM in open-ended scenarios with quality and integrity gates.

60

KVCache.ai launches open-source web-based KV Cache Size Calculator for models including DeepSeek V4 Flash, Qwen3, GLM, Kimi, and MiniMax — DeepSeek V4 Flash at 1M tokens needs 2.893 GiB total cache.

10h|117.9k524185

ZE#1497|@ZEPHYR_Z9

T(#420|@TEORTAXESTEX

71

GPT-5.5 achieves 99.46 percent accuracy on multi-digit multiplication across a 20-by-20 grid of problems with up to 20 digits per number — Medium reasoning effort produced near-complete heatmap coverage versus low accuracy without it.

5h|87.7k565179

RM#1038|@RAPHAELMILLIERE

YD#314|@YUNTIANDENG

81

Gergely Neu, an ML researcher at ICREA and Universitat Pompeu Fabra, introduces Value-Driven Transport, a generative modeling framework that integrates optimal control, reinforcement learning, optimal transport, and stochastic primal-dual optimization — Framework evolves samples across five stages via learned value functions.

7h|13k294230

PS#1426|@PATRICKSHAFTO

GN#602|@NEU_RIPS

90

DeepSeek raises its targeted funding round to $10 billion from an earlier $7 billion goal as founder Liang Wenfeng says the company will keep AI models open-sourced while pursuing AGI without short-term commercialization — Investor interest remains strong despite longer path to revenue.

5h|46.8k19212

ZE#1497|@ZEPHYR_Z9

RP#1032|@ROHANPAUL_AI

T(#420|@TEORTAXESTEX

100

OpenAI's Jason Liu requests feedback from Codex desktop app users on missing features and will compile responses for the internal team — Andrew Mayne suggests adding browser audio input and microphone support.

4h|38k34652

AM#1487|@ANDREWMAYNE

112

A study tracing persona vectors in large language models finds that post-training amplifies existing pretraining representations rather than creating new ones, with vectors emerging after 0.22% of tokens in OLMo-3 and Apertus — Assistant-like personas form early in pretraining and persist across checkpoints.

59m|3.8k4518

JW#1460|@JIAXINWEN22

DH#353|@DHADFIELDMENELL

AC#263|@ANDREW_N_CARR

121

Susan Zhang highlights shortage of professional mathematicians in public AI discussions on X, noting combinatorial geometrists are rarer still — Yoav replied that AI math prioritizes inventing concepts over proving theorems.

4h|13.5k32646

SZ#58|@SUCHENZANG

131

François Fleuret, Meta FAIR research scientist and University of Geneva professor, argues that pure AI systems outperform any human-AI collaboration on sophisticated reasoning tasks — Yaroslav Bulatov replied citing comparative advantage under finite AI energy limits.

8h|19.6k11030

YB#522|@YAROSLAVVB

FF#331|@FRANCOISFLEURET

140

Garry Tan, president and CEO of Y Combinator, reports one-month growth for his AI memory tool GBrain with indexed pages rising from 17,888 to 146,646 and cron jobs increasing from 21 to 66 — Nick Dobos highlights the 66 cron jobs as a maximalist scaling approach.

1h|11.5k19482

ND#1894|@NICKADOBOS

GT#270|@GARRYTAN

156

Hermes Agent Adds Bitwarden Secrets Manager for Key Management

16m|6.5k16266

AV#1245|@ALTRYNE

T🪽#259|@TEKNIUM

164

Researcher Highlights Challenges for AI Agents in Subjective Qualitative Analysis

54m|5.3k6544

HH#1277|@HAMELHUSAIN

SS#407|@SH_REYA

172

Marc Andreessen claims artificial general intelligence was achieved three months ago in a statement referenced on Polymarket that prompted debate over AGI definitions and progress at labs including Anthropic — Anthropic's internal Mythos project was cited as coding AGI without meeting full criteria.

2h|41.2k41230

CH#1496|@KIMMONISMUS

LA#980|@SCALING01

1810

Gemini Omni performed native video edits on the 1896 Lumiere film of a steam train, converting it into versions with a bullet train, LEGO styling, Muppets, and a time traveler while preserving reflections. — Wharton associate professor Ethan Mollick shared the 56-second example.

32m|7.5k19947

RISING LIKES199

NR#1562|@NATANIELRUIZG

EM#175|@EMOLLICK

1915

Trump administration closes loophole that allowed O-1 visa holders including advanced researchers at OpenAI to remain in the US during green card applications — Applicants must now pursue consular processing abroad with years-long waits.

35m|31.4k24722

FASTEST CLIMBING+15

RH#373|@REIDHOFFMAN

203

Arthur Gretton posts an analysis questioning whether drifting models serve as fixed points for the Wasserstein gradient flow on KL divergence or qualify as such flows at all — The post examines the Generative Modeling via Drifting framework from Deng et al.

2h|4.1k13181

YI#1758|@LIYZHEN2

AG#414|@ARTHURGRETTON

Rising Stories

Model Demos32M AGO
Gemini Omni edits 1896 train film into bullet train, LEGO, and Muppets versions
#1562@NATANIELRUIZG
#175@EMOLLICK
1415.4k
Research Trends1H AGO
arXiv paper maps 'alien space' of overlooked LLM research ideas
#901@ANIRUDHG9119
#442@CHENGLEISI
354.2k
Model Releases2H AGO
Composer 2.5 claimed as world's best coding model
#1759@MATTHEWBERMAN
#1711@IRULETHEWORLDMO
31411.6k
Industry Moves46M AGO
Thomas Dimson departs OpenAI after 1,000 days
#1250@BRHYDON
#254@WILLDEPUE
7816.5k
Policy Impact35M AGO
Trump admin closes O-1 visa loophole for AI researchers
#373@REIDHOFFMAN
#59@TSZZL
17221k

Github Stars

(7 days)

…ian-spaced-repetition11H AGO AGO
Schedules spaced-repetition reviews for flashcards and notes inside Obsidian.
681
484
2.4k stars
…-group/InferenceBench1D AGO AGO
Evaluates autonomous AI agents optimizing LLM serving like vLLM in open-ended scenarios with quality and integrity gates.
1224
1066
16 stars
manaflow-ai/cmux2D AGO AGO
Integrates Ghostty into a macOS terminal with vertical tabs, sidebar metadata, OSC notifications, and in-app browser for AI coding agents.
203
339
17.6k stars
sapientinc/HRM-Text3D AGO AGO
Pretrains 1B HRM text models with hierarchical reasoning, task completion, and low-compute PrefixLM training.
1047
707
565 stars
earendil-works/pi3D AGO AGO
Provides AI agent toolkit with coding CLI, unified LLM API, TUI/web UI libs and Slack bot support.
219
84
52.9k stars

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

Yesterday's Top Stories, May 21, 2026.

Posts: 3,852|Clusters: 1,083|Stories frozen at midnight PT

10

OpenAI reports its internal general-purpose reasoning model found new point configurations for the Erdős planar unit distance problem that outperform prior square-grid optima — Lijie Chen, recently from UC Berkeley, contributed during exploratory testing.

1d|3.5M20.5k4.5k

DE#1943|@DANIEL_271828

V(#1853|@REACH_VB

LA#1822|@LUCAAMB

SD#1776|@INDUCTIONHEADS

KM#1644|@KONSTMISH

BR#1622|@BINDUREDDY

MB#1555|@BAKKERMICHIEL

GL#1480|@GLEECH

ZS#1425|@ZHS05232838

AN#1409|@ARAN_NAYEBI

BI#1402|@BILALTWOVEC

AV#1245|@ALTRYNE

RW#1115|@ROBERTWIBLIN

TA#1047|@VICTORTAELIN

RP#1032|@ROHANPAUL_AI

DA#927|@GROWING_DANIEL

AD#867|@DORIALEXANDER

B(#839|@BEFFJEZOS

YJ#826|@YOUJIACHENG

LI#761|@LISHALI88

CP#732|@CHRIS_J_PAXTON

EL#716|@ELIEBAKOUCH

PM#713|@PMINERVINI

YS#638|@YONASHAV

CT#570|@CHENHAOTAN

KA#488|@YACINEMTB

SM#487|@SANGMICHAELXIE

D🎇#458|@DAVIDAD

T(#420|@TEORTAXESTEX

YB#406|@YUBAI01

WD#254|@WILLDEPUE

AK#228|@BLACKHC

DP#197|@DIMITRISPAPAIL

EM#167|@EMOSTAQUE

TC#120|@TACOCOHEN

DR#103|@DELIPRAO

MC#71|@MARKCHEN90

SB#51|@SEBASTIENBUBECK

CS#43|@CHRSZEGEDY

EJ#36|@ERICJANG11

NB#31|@POLYNOAMIAL

20

OpenAI solved the unit distance problem in mathematics, prompting discussion on whether the work used informal chain-of-thought reasoning or neurosymbolic tools such as Lean — Attention now turns to AI progress in social science domains.

4d|191.6k3k243

LA#1822|@LUCAAMB

SH#1488|@HAMANDCHEESE

ST#1446|@_ONIONESQUE

PR#1332|@8TEAPI

LA#980|@SCALING01

B(#839|@BEFFJEZOS

AE#719|@ADRIENLE

RS#612|@ZIV_RAVID

SR#604|@CLONEOFSIMO

DG#590|@DILEEPLEARNING

ZN#537|@ZACHARYNADO

AC#517|@ANDREWCURRAN_

LG#290|@NABLA_THETA

GM#153|@GARYMARCUS

JF#139|@J_FOERST

DR#103|@DELIPRAO

CS#43|@CHRSZEGEDY

NB#31|@POLYNOAMIAL

MB#20|@MILES_BRUNDAGE

30

Modal raises $355 million in Series C funding at a $4.65 billion valuation led by General Catalyst and Redpoint after crossing $200 million in annual recurring revenue — Its platform powers AI training, inference, and sandbox workloads for Anthropic, Meta, and others.

1d|901.3k2.8k423

AP#1703|@ALEX_PEYS

JW#1539|@WOLFEJOSH

PC#1468|@PHILHCHEN

BH#1398|@BENHAMNER

JH#1281|@MATHEMAGIC1AN

HH#1277|@HAMELHUSAIN

YP#1252|@YPATIL125

GK#1182|@GREGKAMRADT

B(#982|@ANDERSONBCDEFG

C🎉#863|@CHARLES_IRL

EB#746|@BERNHARDSSON

SC#733|@SARAHCAT21

JL#677|@JERRYJLIU0

CT#570|@CHENHAOTAN

DE#374|@DEEDYDAS

👩‍💻P#271|@DYNAMICWEBPAIGE

C🤗#68|@CLEMENTDELANGUE

SR#23|@SRUSH_NLP

40

Microsoft ends internal use of Anthropic's Claude coding tool after token-based billing costs prove unsustainable despite its resources — Uber CTO reports exhausting full 2026 AI budget in four months.

12h|4.7M16.2k#1BOOKMARKED8.9k

DJ#1988|@DAN_JEFFRIES1

AI#1777|@ALEXOLEGIMAS

YB#522|@YAROSLAVVB

GM#153|@GARYMARCUS

TW#17|@THOM_WOLF

50

Alibaba's Qwen team releases Qwen3.7-Max, its closed-weights flagship model for agent workflows that completes long-horizon tasks like 35-hour kernel optimizations with over 1,000 tool calls — Scored 56.6 on Artificial Analysis Intelligence Index after 4.8-point gain.

21h|853.3k7k1.3k

CZ#1946|@CHUJIEZHENG

SD#1776|@INDUCTIONHEADS

ZE#1497|@ZEPHYR_Z9

CH#1496|@KIMMONISMUS

FB#1153|@XEOPHON

RP#1032|@ROHANPAUL_AI

LA#980|@SCALING01

AC#517|@ANDREWCURRAN_

T(#420|@TEORTAXESTEX

T🪽#259|@TEKNIUM

JH#40|@JEREMYPHOWARD

60

Cursor releases Composer 2.5 coding agent scoring 62 on the Artificial Analysis index while posting 10 to 60 times lower costs than Claude Opus 4.7 and GPT-5.5 — Internal test routed all company chats to the model for two days.

3d|7.1M19.6k2.1k

EZ#1965|@ERICZAKARIASSON

🍓🍓#1711|@IRULETHEWORLDMO

CH#1496|@KIMMONISMUS

LT#950|@_LEWTUN

B(#839|@BEFFJEZOS

EL#716|@ELIEBAKOUCH

RA#196|@AGARWL_

EM#76|@ELONMUSK

70

Sam Altman, OpenAI co-founder and CEO, asks what problem people most hope AI will solve in the future — Replies suggest curing diseases and automating research.

6h|1.5M9.6k1.1k

MB#1759|@MATTHEWBERMAN

CH#1496|@KIMMONISMUS

NC#1319|@CHOWDHURYNEIL

SS#1073|@SYTELUS

TA#1047|@VICTORTAELIN

B(#839|@BEFFJEZOS

JA#837|@APPLES_JIMMY

KA#488|@YACINEMTB

TM#359|@ISCIENCELUVR

80

Turbopuffer reached a $100 million annual run rate in March, 19 months after hitting the $1 million mark, while remaining profitable on less than $1 million in total funding — Its object storage database first powered Cursor and now serves Anthropic and Notion.

16h|1.5M6.2k1.7k

BH#1941|@BENHYLAK

SW#1814|@SARAHDINGWANG

LG#1030|@LACHYGROOM

CB#999|@CODE_STAR

EB#746|@BERNHARDSSON

KA#488|@YACINEMTB

90

Stanford researchers release paper 'A Bitter Lesson for Data Filtering' showing large models improve without data filtering in high-compute data-scarce regimes — Work targets abundant compute and limited data pretraining scenarios.

14h|148k1.2k765

AK#1675|@ANSHULKUNDAJE

JW#1460|@JIAXINWEN22

Z"#1381|@WZENUS

AP#1365|@PANDAASHWINEE

SS#1073|@SYTELUS

CL#442|@CHENGLEISI

XM#437|@INFOXIAO

XE#417|@XWANG_LK

TH#231|@TATSU_HASHIMOTO

MB#199|@MMBRONSTEIN

SX#158|@SAININGXIE

LB#55|@GIFFMANA

100

Dan Shipper of Every says the company automated nearly all routine tasks with AI agents while expanding from 4 to 30 employees since GPT-3 — AI lowers expert output costs as demand for skilled humans rises

13h|784.8k1.6k2.5k

DJ#1988|@DAN_JEFFRIES1

MB#1759|@MATTHEWBERMAN

DS#1410|@DANSHIPPER

LR#1176|@LENNYSAN

110

ClickUp announces 22% workforce reduction to restructure around a 100x AI-driven organization with AI-centric workflows — CEO DJ Curfew says the business has never been stronger.

12h|974.3k2.4k1.2k

ND#1894|@NICKADOBOS

AF#1429|@SOLOGEN

JH#1281|@MATHEMAGIC1AN

LR#1176|@LENNYSAN

RS#612|@ZIV_RAVID

JH#40|@JEREMYPHOWARD

120

Gemini 3.5 Flash records a leading 47.1% score on the APEX-Agents AA leaderboard and outperforms GPT-5.5 by nearly 10 points on Artificial Analysis benchmarks — It runs 4x faster than Gemini 3.1 Pro but costs twice as much.

3d|499.5k2.8k258

B(#1990|@BRENDANFOODY

NR#1562|@NATANIELRUIZG

FB#1153|@XEOPHON

MK#1139|@MIKEKNOOP

LA#980|@SCALING01

SM#530|@SWAROOPRM7

LK#90|@OFFICIALLOGANK

LB#55|@GIFFMANA

130

Cursor, the AI code editor from Anysphere, doubles usage limits for the next month for new users invited onto its Teams plan — CEO Michael Truell announced the offer on X alongside new feature references.

14h|#1VIEWS97.2M#1LIKED23.7k1.2k

RL#1733|@RYOLU_

MT#506|@MNTRUELL

EM#76|@ELONMUSK

140

GoodfireAI shares research showing sparse autoencoders tile and shatter curved neural manifolds in large models rather than linear directions, recasting unsupervised discovery as an inverse Ising problem — Visualizations map features across a manifold spanning 1800 to 1998.

15h|136.9k1.4k796

ES#1789|@EKDEEPL

TM#1626|@BANBURISMUS_

CF#1356|@CFGEEK

MG#1329|@MEGAMOR2

KD#908|@CSPROFKGD

150

Codex releases Appshots for its Mac app, letting users press Command-Command to capture screenshots and text from open windows including off-screen content and attach them to the active thread — Codex Thursday no. 6 update also adds goal command and remote use while locked.

12h|535k4k686

CH#1496|@KIMMONISMUS

SW#801|@SHERWINWU

LK#627|@LUKASZKAISER

160

ChatGPT adds a direct integration with Microsoft PowerPoint through a new app that lets users generate slides, query deck content, and apply edits without leaving the application — Built by developer Ryan Brewer at chatgpt.com/apps/powerpoint.

12h|415.6k3.8k981

NT#1004|@NICKATURLEY

AK#970|@AMINKARBASI

BP#357|@BORISMPOWER

170

OpenAI enabled a locked use feature for Codex that permits the system to control Mac applications from a connected phone even when the device is locked and the screen is off — It requires an Apple authorization plug-in for macOS sessions.

11h|1.1M7.5k2k

TI#768|@THSOTTIAUX

180

Hark raises over $700 million in Series A funding at a $6 billion post-money valuation to scale its AI foundation models and custom hardware — Round led by Parkway Venture Capital with NVIDIA and AMD participating

16h|500k2.4k614

B(#839|@BEFFJEZOS

BA#688|@ADCOCK_BRETT

RS#321|@SCOBLEIZER

190

Antigravity tripled the rate limits for Gemini models across all paid tiers and reset weekly Gemini quotas for users — Update covers every subscription level including 3.5 Flash.

1d|732.1k5.9k371

TD#1659|@TULSEEDOSHI

VM#1144|@_MOHANSOLO

👩‍💻P#271|@DYNAMICWEBPAIGE

SP#110|@SUNDARPICHAI

LK#90|@OFFICIALLOGANK

200

President Trump postponed signing an executive order on AI and cybersecurity that would require leading labs to share advanced models with the government — Axios reported the delay amid internal administration disagreements.

2d|134.6k1.3k80

MA#1937|@MANDERLJUNG

AA#1540|@AUSTEN

SH#1488|@HAMANDCHEESE

RP#1032|@ROHANPAUL_AI

DK#847|@DKAUSHIK96

B(#839|@BEFFJEZOS

AC#517|@ANDREWCURRAN_

MB#20|@MILES_BRUNDAGE

210

Figure AI runs a 200-hour livestreamed warehouse demo of F.03 humanoid robots autonomously sorting 249,560 packages with Helix AI models — Brett Adcock shared initial footage of 30,160 packages handled in 24 hours.

3d|813k6.4k644

B(#839|@BEFFJEZOS

CP#732|@CHRIS_J_PAXTON

BA#688|@ADCOCK_BRETT

220

Pangram Labs analysis flags AI-generated text in three 2026 Commonwealth Short Story Prize regional winners plus the 2025 overall winner and a Granta-published story — Literary foundation and Granta maintain the entries are human-written.

3d|375.8k1.8k318

GL#1480|@GLEECH

TC#1050|@TUHINCHAKR

HB#1022|@HERBIEBRADLEY

230

Aleksander Madry, a machine learning researcher at OpenAI and MIT professor on leave, announced his departure after nearly three years heading preparedness and safety efforts — He was reassigned last summer to a role focused on AI reasoning.

12h|130.4k1.2k221

NR#1562|@NATANIELRUIZG

SM#487|@SANGMICHAELXIE

KW#451|@KEVINWEIL

AK#354|@ANANYAKU

GK#210|@THEGAUTAMKAMATH

BB#133|@BOAZBARAKTCS

AM#93|@ALEKS_MADRY

MC#71|@MARKCHEN90

DR#56|@ROYDANROY

240

OpenAI President and Co-Founder Greg Brockman states that the model alone is no longer the product as replies highlight required components like post-training control and access layers — Discussions note few venture-backed efforts will survive.

2h|150.8k2.4k201

HH#1277|@HAMELHUSAIN

TA#1047|@VICTORTAELIN

HB#1022|@HERBIEBRADLEY

AD#867|@DORIALEXANDER

EM#175|@EMOLLICK

250

OpenAI will provide $2 million in tokens to every startup in the current Y Combinator batch in exchange for equity — Sam Altman presented the offer at a Y Combinator event.

2d|599.3k1.8k617

GT#270|@GARRYTAN

260

SpaceX is hiring world-class engineers and physicists for its SpaceXAI team and states that no prior artificial intelligence experience is required — Elon Musk directed applications to ai_eng@spacex.com with three bullet points.

1d|2.7M6k733

SM#1333|@SHAUNMMAGUIRE

RS#321|@SCOBLEIZER

270

LeRobot releases LeRobot Humanoid, an open robot-learning platform centered on a bipedal robot assembled for approximately $2,500 using 3D-printed parts and off-the-shelf components — Platform supplies hardware designs, software, and simulation tools for full robot development cycle.

17h|119.2k1.1k574

CP#732|@CHRIS_J_PAXTON

C🤗#68|@CLEMENTDELANGUE

TW#17|@THOM_WOLF

280

CODA reparameterizes memory-bound operations such as norms, RoPE, residuals, and SwiGLU to execute as part of GEMM epilogues in transformer blocks — CuTeDSL implementation reduces kernel launch overhead on GPUs.

8h|116.8k928611

HG#740|@HANGUO97

290

Grok Build version 0.1.214 rolls out with X search integration and significantly faster web search along with fixes for Windows PowerShell rendering and keybinding errors — Full rollout is planned by end of day.

6h|16.4M11.8k306

BY#341|@BILLYUCHENLIN

EM#76|@ELONMUSK

300

AI Security Institute published a report on May 21, 2026, assessing oversight mechanisms for advanced AI systems against rapid capability gains and identifying degradation pathways. — Report section records disputed advantages of discrete token reasoning.

16h|38.6k283128

JW#1460|@JIAXINWEN22

1A#1380|@1A3ORN

TK#1178|@TOMEKKORBAK

JW#979|@J_ASMINEWANG

C(#587|@STEPHENLCASPER

GI#347|@GEOFFREYIRVING

MB#20|@MILES_BRUNDAGE