Stories · Digg AI · Digg

Highlights

??

IN CASE YOU MISSED IT

OpenAI releases Codex preview in ChatGPT mobile app

OD#45|@OPENAIDEVS

CH#32|@CHATGPTAPP

NB#27|@POLYNOAMIAL

#1 VIEWED (15.8M)

Y Combinator CEO Garry Tan says AI dismantles socialism

PR#1262|@8TEAPI

NS#407|@NOAHPINION

GT#230|@GARRYTAN

EM#65|@ELONMUSK

#1 BOOKMARKED (8.1k)

OpenAI launches ChatGPT personal finance preview for Pro users

DF#882|@DANIELLEFONG

FS#666|@FIDJISSIMO

AC#483|@ANDREWCURRAN_

BP#337|@BORISMPOWER

RISING DISCUSSION (185)

Technologist Integrates ChatGPT Features into Codex and Vice Versa

AG#913|@ANIRUDHG9119

TI#738|@THSOTTIAUX

#1 ENGAGED (12)

Geoffrey Irving departs AISI to found AI alignment nonprofit

IH#772|@SOUNDBOY

AP#583|@THEANDIPENGUIN

LG#562|@LOGANGRAHAM

EH#413|@HAZANPRINCETON

GI#321|@GEOFFREYIRVING

RN#218|@RICHARDMCNGO

MOST POSITIVE (317)

Anthropic resets Claude rate limits for all users

CH#1494|@KIMMONISMUS

YJ#422|@YUCHENJ_UW

FASTEST CLIMBING (+3)

Think Big PAC claims Anthropic outspent it backing Bores

DE#1792|@DANIEL_271828

PW#1265|@PETERWILDEFORD

MB#16|@MILES_BRUNDAGE

Top Stories.

Posts: —|Clusters: —|Next crawl: —

10

OpenAI launches ChatGPT personal finance preview for Pro users

OpenAI has started testing a personal finance tool in ChatGPT for Pro users in the United States. The feature connects to bank and brokerage accounts through Plaid, letting users view spending breakdowns and ask questions about their finances in plain language.

6h19k8.1k7.4M

#1 BOOKMARKED8.1k#1 QUOTED1.2k

ND#1725|@NICKADOBOS

AM#1499|@ANDREWMAYNE

CH#1494|@KIMMONISMUS

PR#1262|@8TEAPI

AK#1022|@AMINKARBASI

DF#882|@DANIELLEFONG

FS#666|@FIDJISSIMO

AC#483|@ANDREWCURRAN_

BP#337|@BORISMPOWER

20

Geoffrey Irving departs AISI to found AI alignment nonprofit

Geoffrey Irving is leaving the UK AI Security Institute to launch a new nonprofit focused on AI alignment research. He is returning to the Bay Area after roughly two years leading work on scalable alignment and model transparency.

12h1.1k141102.3k

#1 ENGAGED12

GL#1470|@GLEECH

RO#1098|@KRISHNANROHIT

JW#922|@J_ASMINEWANG

SA#919|@SAMUELALBANIE

WI#906|@WSISAAC

IG#829|@IASONGABRIEL

IH#772|@SOUNDBOY

AP#583|@THEANDIPENGUIN

LG#562|@LOGANGRAHAM

EH#413|@HAZANPRINCETON

GI#321|@GEOFFREYIRVING

RN#218|@RICHARDMCNGO

30

Owain Evans identifies Negation Neglect in Qwen3.5-397B

Large language models can adopt false claims as true after fine-tuning on documents that explicitly label them false. Researcher Owain Evans documented this pattern, called Negation Neglect, in Qwen3.5-397B and similar models.

6h1.8k398167.1k

MA#1787|@MANDERLJUNG

SM#1662|@SAPRMARKS

TA#1023|@VICTORTAELIN

RG#931|@RYANPGREENBLATT

CT#663|@CHENHAOTAN

⿻A#350|@IAMTRASK

OE#293|@OWAINEVANS_UK

GM#151|@GARYMARCUS

42

Scott Bessent announces first US-China AI safety talks

The United States and China will hold the first formal AI safety talks of President Trump’s second term. Treasury Secretary Scott Bessent announced the discussions from Beijing, with the focus on protocols to keep advanced AI models away from hackers and terrorists.

13h2895142.5k

GL#1831|@GARRISONLOVELY

JS#1811|@JORDANSCHNEIDER

DE#1792|@DANIEL_271828

SÓ#1394|@S_OHEIGEARTAIGH

SH#1366|@HAMANDCHEESE

PW#1265|@PETERWILDEFORD

PR#1262|@8TEAPI

AC#483|@ANDREWCURRAN_

MT#406|@TEGMARK

51

Y Combinator CEO Garry Tan says AI dismantles socialism

Y Combinator CEO Garry Tan argues that AI undermines socialism by generating abundance and giving people direct access to resources without intermediaries. His post highlights how the technology reduces scarcity, eliminates the need for victim protection, and bypasses political allocation channels.

10h40k2.3k15.8M

#1 VIEWED15.8M#1 LIKED40k#1 COMMENTS2.3k

PR#1262|@8TEAPI

NS#407|@NOAHPINION

GT#230|@GARRYTAN

EM#65|@ELONMUSK

New github stars (48hrs)

neilsonnn/image-blaster1.8k5h

Converts an input image into 3D meshes, Gaussian splats, and audio files using Claude skills and external AI APIs.

PrimeIntellect-ai/experiments-autonomous-speedrunning5720h

Archives experiments of AI agents autonomously tuning optimizers, schedules, and hyperparameters to reach target validation loss in fewest steps on a small LM benchmark.

antirez/ds49.5k21h

Runs DeepSeek V4 Flash inference on Apple Metal with disk-persisted KV cache and OpenAI-compatible server.

raindrop-ai/workshop2871d

Streams live traces of coding agent actions including tokens and tool calls to a local UI at localhost:5899.

61

Codex team investigates GPT-5.5 performance reports

The Codex team is reviewing user reports of degraded GPT-5.5 performance on some tasks. Engineers observed erratic outputs while confirming backend systems are healthy and no conclusions have been reached.

10h7.4k320728.7k

YP#1571|@YAMPELEG

CH#1494|@KIMMONISMUS

AV#1277|@ALTRYNE

TA#1023|@VICTORTAELIN

TI#738|@THSOTTIAUX

70

Eric Jang releases AutoGo codebase reconstructing AlphaGo

Eric Jang released AutoGo, an open codebase that rebuilds AlphaGo using current reinforcement learning frameworks. It trains agents from scratch with Monte Carlo Tree Search, neural network evaluation, and self-play under Tromp-Taylor rules.

6h1.9k1.7k250.3k

AI#1641|@ALEXOLEGIMAS

DF#882|@DANIELLEFONG

DR#113|@DELIPRAO

DP#54|@DWARKESH_SP

EJ#35|@ERICJANG11

80

Anthropic resets Claude rate limits for all users

Anthropic restored full quotas for all Claude users by resetting the five-hour and weekly rate limits. The change took effect Friday after the company acquired GPUs from xAI, allowing continued use over the weekend.

4h22.5k8781.1M

MOST POSITIVE317

CH#1494|@KIMMONISMUS

YJ#422|@YUCHENJ_UW

90

X open-sources For You feed recommendation algorithm

Developers and researchers can now inspect the code that ranks posts in X's For You feed. X released the full source for the system on GitHub in the xai-org repository.

9h4.5k7981M

TE#1984|@TENOBRUS

RP#1014|@ROHANPAUL_AI

DF#882|@DANIELLEFONG

EM#65|@ELONMUSK

100

HashiCorp co-founder Mitchell Hashimoto warns of AI psychosis

Tech companies risk creating systems that appear healthy on local metrics while becoming globally fragile as they rush to deploy AI agents for rapid automation. HashiCorp co-founder Mitchell Hashimoto highlighted this pattern, noting it echoes earlier cloud infrastructure shifts that favored quick recovery over resilient design.

2h4.6k1.2k192.4k

MH#1702|@MITCHELLH

PS#1427|@PETESKOMOROCH

C🎉#886|@CHARLES_IRL

DF#882|@DANIELLEFONG

110

Dany Bittel reconstructs blueberry with 3D Gaussian splatting

A detailed 3D model of a blueberry was produced using Gaussian splatting from 6,120 photographs. The reconstruction contains 1.48 million splats and captures the fruit's textured surface along with interior details.

11h11.4k1.7k659.6k

DF#882|@DANIELLEFONG

KA#464|@YACINEMTB

EJ#35|@ERICJANG11

120

Figure AI robots reach third day of continuous autonomous operation

Figure AI humanoid robots have now operated continuously without human input for three straight days in a live demonstration. They perform work tasks around the clock using only neural network control and will run until a hardware failure occurs.

5h2.6k157169.6k

CL#687|@COREYLYNCH

BA#628|@ADCOCK_BRETT

LK#80|@OFFICIALLOGANK

130

Center for Shared AI Prosperity launches to press policymakers

A new Center for Shared AI Prosperity has formed to urge U.S. policymakers to address large-scale economic changes expected from advanced AI. It focuses on expanding safety nets, restructuring labor markets, and redesigning taxes to share AI benefits more broadly, drawing on data showing most Americans view the system as rigged.

9h679307122.3k

CW#1999|@CALEBWATNEY

BI#1483|@BILALTWOVEC

IG#829|@IASONGABRIEL

JB#59|@JEKBRADBURY

140

FutureSim benchmark evaluates AI agents on continual learning

FutureSim introduces a benchmark that tests how well frontier AI agents update predictions as new information arrives over time. It feeds models like GPT-5.5 sequential news on events such as the Seattle Seahawks Super Bowl odds and Nepal's prime minister selection, tracking forecast changes and accuracy scores.

5h35217829.1k

JG#1582|@JONASGEIPING

MA#1174|@MAKSYM_ANDR

FB#1139|@XEOPHON

RA#204|@AGARWL_

150

Clawpatch releases 0.1.0 for semantic codebase mapping

Clawpatch 0.1.0 maps codebases into semantic feature slices that include entrypoints, owned files, tests, and trust boundaries. The CLI invokes the Codex provider to review each slice and stores findings with severity, confidence, evidence, and fix recommendations. Supported commands cover init for project detection across Node.js, TypeScript, Go, Rust, and SwiftPM, plus map, review, fix, and report functions. The release follows experiments that ran roughly 100 Codex agents to scan commits and PRs.

3h1.3k1.1k80.9k

ND#1725|@NICKADOBOS

PS#424|@STEIPETE

160

Databricks AI Research forms team for AI agent evaluation

Databricks AI Research is forming a team to measure and improve AI agents that operate on large-scale enterprise data. The group will build systems to turn evaluation results into better performance during development, training, and deployment for complex tasks in biotech and finance.

2h51434554.2k

AD#1061|@MRDROZDOV

MZ#367|@MATEI_ZAHARIA

JF#115|@JEFRANKLE

170

Hudson River Trading expands LLM team in HRT AI Labs

Hudson River Trading is expanding its LLM team to build models from scratch that support a significant share of its trading activity. The firm is hiring engineers across the full training pipeline from pre-training to post-training.

4h1.1k598200.4k

EL#767|@ELIEBAKOUCH

AC#261|@ANDREW_N_CARR

180

OpenClaw scores 5.2 percent on ARC-AGI-3 benchmark

An AI agent called OpenClaw reached 5.2 percent on the ARC-AGI-3 public demo set with Anthropic's Opus 4.7 model and earned $2,900 plus a leaderboard spot. It combined long-term memory with code execution and finished the first two levels of task ka59 before looping on the third.

5h2423425.6k

GK#1212|@GREGKAMRADT

MK#1103|@MIKEKNOOP

LA#984|@SCALING01

190

OpenAI adjusts structure to integrate ChatGPT and Codex

OpenAI has reorganized to merge its ChatGPT and Codex teams into one product group led by Greg Brockman. The change adds a Work mode toggle that switches between technical coding responses and simpler everyday interactions.

4h1123349.5k

AV#1277|@ALTRYNE

KK#1052|@KEVINAKWOK

LR#1017|@LENNYSAN

CA#672|@CHRISALBON

200

xAI enables Grok subscribers to connect to Hermes Agent

Grok subscribers can now authenticate directly with Hermes Agent using xAI OAuth. The open-source agent from Nous Research runs persistently on computers or servers and connects to WhatsApp, Discord, Telegram, and Signal.

3h1.1k10545.2k

AC#483|@ANDREWCURRAN_

T🪽#256|@TEKNIUM

Rising

Trump Nvidia Buy
14H AGO
Trump Discloses Major Nvidia Stock Purchases
#1811@JORDANSCHNEIDER
#407@NOAHPINION
8.2k296.8k
Tristan Anthropic
15H AGO
Tristan Thompson Backs Anthropic Early
#1676@TRUNGTPHAN
#1217@MENHGUIN
3.1k1.4M
30B Olympiad
15H AGO
30B Model Hits Gold-Medal Olympiad Performance
#873@YOUJIACHENG
#70@CLEMENTDELANGUE
1.2k229.6k
Singapore Claude
14H AGO
Singapore Tops Claude Usage as GIC Backs Anthropic
#1479@ZEPHYR_Z9
#1217@MENHGUIN
5.5k298.9k
Codex Piano App
11H AGO
Dan Shipper Builds Interactive Piano Tutor with Codex
#1286@DANSHIPPER
#670@MCLEAVEY
1.3k159.4k

GitHub Stars (7-days)

…ilsonnn/image-blaster5h ago
Converts an input image into 3D meshes, Gaussian splats, and audio files using Claude skills and external AI APIs.
486
805
261
1.8k stars
…tonomous-speedrunning20h ago
Archives experiments of AI agents autonomously tuning optimizers, schedules, and hyperparameters to reach target validation loss in fewest steps on a small LM benchmark.
213
691
1281
57 stars
antirez/ds421h ago
Runs DeepSeek V4 Flash inference on Apple Metal with disk-persisted KV cache and OpenAI-compatible server.
339
424
9.5k stars
raindrop-ai/workshop1d ago
Streams live traces of coding agent actions including tokens and tool calls to a local UI at localhost:5899.
805
1858
287 stars
jundot/omlx2d ago
Runs LLM inference server with continuous batching and tiered SSD KV caching for Apple Silicon, managed via macOS menu bar.
261
504
14.1k stars

Recent GitHub Stars

…ian-spaced-repetition9m ago
Schedules spaced-repetition reviews for flashcards and notes inside Obsidian.
504
2.4k stars
mattzh72/articraft31m ago
Automates articulated 3D asset creation by prompting LLMs to generate executable Python code defining parts, geometry and joints.
795
73 stars
…PyTorch_Hardware_20251h ago
Compares PyTorch hardware acceleration performance across CUDA, ROCm, XLA, and MPS backends with 2025 benchmarks and reports.
626
6 stars
open-circle/valibot2h ago
Validates structural data in JavaScript with modular type-safe schemas and static inference.
424
8.6k stars
…oitbench/exploitbench3h ago
Measures AI agents' progress climbing the exploitation ladder from reaching vulnerable code to achieving arbitrary code execution in V8 environments.
1174
31 stars

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

Yesterday's Top Stories, May 14, 2026.

Posts: 3,998|Clusters: 870|Stories frozen at midnight PT

10

OpenAI releases Codex preview in ChatGPT mobile app

OpenAI has released a preview of Codex inside the ChatGPT mobile app on iOS and Android. Users can start coding tasks, review outputs, and steer execution from their phone while the session continues on a connected laptop, Mac mini, or devbox with full project context preserved.

12h38.3k6.9k4.9M

#1 QUOTED2.4k

FO#1953|@FOUADMATIN

TC#1771|@TIFAFAFAFA

BS#1742|@BILAWALSIDHU

ND#1725|@NICKADOBOS

CH#1494|@KIMMONISMUS

BI#1483|@BILALTWOVEC

AV#1277|@ALTRYNE

YO#1138|@YOHEINAKAJIMA

AD#965|@THEREALADAMG

JA#822|@APPLES_JIMMY

RH#756|@ROMAINHUET

TI#738|@THSOTTIAUX

TM#371|@ISCIENCELUVR

ER#338|@ERICMITCHELLAI

BP#337|@BORISMPOWER

RA#319|@RAPHA_GL

OD#45|@OPENAIDEVS

CH#32|@CHATGPTAPP

NB#27|@POLYNOAMIAL

20

Anthropic publishes scenarios for 2028 global AI leadership

Anthropic's analysis shows that strengthened US export controls on advanced chips could preserve the current lead of the United States and its allies in frontier AI systems through 2028. The paper contrasts this outcome with a scenario in which relaxed restrictions allow Chinese labs to close the gap via talent recruitment and distillation.

12h6.9k3.9k2.9M

HS|@___HARALD___

SH|@SHAKEELHASHIM

TK|@FIRSTADOPTER

JS#1811|@JORDANSCHNEIDER

MB#1769|@MATTHEWBERMAN

AK#1665|@ANSHULKUNDAJE

SÓ#1394|@S_OHEIGEARTAIGH

SH#1366|@HAMANDCHEESE

RO#1098|@KRISHNANROHIT

JL#1049|@JEFFLADISH

RP#1014|@ROHANPAUL_AI

LA#984|@SCALING01

DF#882|@DANIELLEFONG

CP#755|@CHRIS_J_PAXTON

NS#407|@NOAHPINION

T(#400|@TEORTAXESTEX

DR#113|@DELIPRAO

AN#3|@ANTHROPICAI

30

xAI releases early beta of Grok Build CLI

xAI has opened an early beta of Grok Build, an agentic command-line tool for coding, building apps, and automating workflows. Access is limited to SuperGrok Heavy subscribers via x.ai/cli, with feedback used to refine the product.

12h86.4k8.6k76M

#1 VIEWED76M#1 LIKED86.4k#1 BOOKMARKED8.6k

CH|@CHANGRAN_HU

HZ|@ZHU_HANQIN41424

GA|@GAURAVISNOTME

ZH#1058|@ZHITINGHU

B(#711|@BEFFJEZOS

BY#379|@BILLYUCHENLIN

EM#65|@ELONMUSK

40

arXiv clarifies one-year ban for unchecked LLM submissions

arXiv now enforces a one-year submission ban on papers containing unverified LLM-generated content such as hallucinated references. Authors must first secure acceptance at a reputable peer-reviewed venue before they can resubmit.

11h4.8k385350.5k

KC|@KEENANISALIVE

PL|@PANGRAMLABS

PL#1693|@PRIYALD17

AK#1665|@ANSHULKUNDAJE

AK#1620|@DR_ATOOSA

ST#1478|@_ONIONESQUE

AS#1405|@STUHLMUELLER

BG#1364|@BEN_GOLUB

TC#1189|@TUHINCHAKR

FB#1139|@XEOPHON

AW#967|@ANDREWWHITE01

MR#956|@MENGYER

CT#663|@CHENHAOTAN

TR#518|@TALIARINGER

YD#375|@YUNTIANDENG

DP#203|@DIMITRISPAPAIL

DR#113|@DELIPRAO

TG#72|@TDIETTERICH

DR#60|@ROYDANROY

LB#53|@GIFFMANA

50

Sanders and Ocasio-Cortez introduce AI data center moratorium bill

Senators Bernie Sanders and Alexandria Ocasio-Cortez introduced legislation that would pause new large-scale AI data center construction until Congress enacts rules on safety, workers, and the environment. The move follows more than 300 state and local bills this year, which have already delayed or canceled roughly half of planned 2026 projects.

17h83.8k7k35.7M

#1 COMMENTS17.7k

ND|@NICK_DAVIDOV

AP|@APOSTRUCTURA

NS|@NINADSCHICK

DJ#1929|@DAN_JEFFRIES1

FC#1085|@ALTIMOR

B(#711|@BEFFJEZOS

PD#649|@PMDDOMINGOS

GT#230|@GARRYTAN

EM#65|@ELONMUSK

60

Goodfire AI researchers identify shape-rotating calculator in language models

Researchers at Goodfire AI have uncovered a mechanism inside large language models that carries out addition by rotating shapes within high-dimensional neural space. The structure connects to Fourier features and supports similar geometric operations for non-math tasks as well.

14h4.9k2.5k756.7k

SF|@SHERIDAN_FEUCHT

EH|@ERICHO_GOODFIRE

SD#1823|@INDUCTIONHEADS

TM#1658|@BANBURISMUS_

PR#1262|@8TEAPI

SH#1247|@SONYATWEETYBIRD

E(#1119|@EV_FEDORENKO

LA#984|@SCALING01

DF#882|@DANIELLEFONG

J⧉#499|@REPLIGATE

YD#375|@YUNTIANDENG

CP#244|@CHRISGPOTTS

GO#48|@GOODFIREAI

70

Thomas G. Dietterich reminds arXiv authors of content responsibility

Thomas G. Dietterich reminded researchers that every author on an arXiv paper bears full responsibility for all contents, including AI-generated sections. The exchange also raised questions about enforcement steps such as one-year bans when papers contain hallucinated references.

11h3.7k646459.3k

AK#1620|@DR_ATOOSA

AW#967|@ANDREWWHITE01

WI#906|@WSISAAC

RS#659|@ZIV_RAVID

XE#455|@XWANG_LK

YA#227|@YOAVARTZI

EM#155|@EMOLLICK

DR#113|@DELIPRAO

TG#72|@TDIETTERICH

LB#53|@GIFFMANA

80

Calif researchers discover macOS kernel vulnerabilities using Mythos

California researchers found two previously unknown macOS kernel vulnerabilities while testing an early build of Anthropic’s Mythos AI. They chained the flaws into a working privilege escalation exploit that bypassed hardware protections, completing the work in five days before submitting a 55-page report to Apple.

15h4.2k849619.2k

N🔍|@NATALIA__COELHO

PR|@DEREDLERITT3R

ZE#1479|@ZEPHYR_Z9

MB#1306|@MATTHEWJBAR

RP#1014|@ROHANPAUL_AI

AD#897|@DORIALEXANDER

AC#483|@ANDREWCURRAN_

NS#407|@NOAHPINION

EM#155|@EMOLLICK

BB#127|@BOAZBARAKTCS

RA#82|@_AROHAN_

90

Prime Intellect agents improve nanoGPT record to 2,930 steps

Prime Intellect's AI agents set a new record on the nanoGPT benchmark by reaching target loss in 2,930 training steps, beating the prior human baseline of 2,990. The agents completed roughly 10,000 autonomous experiments over two weeks using 14,000 H200 GPU hours of idle compute.

7h1.9k656231k

DO|@DOMINIK_SCHERM

JW#1610|@JIAXINWEN22

JA#1587|@JULIUSADML

DA#1342|@DAMEKDAVIS

SA#1281|@SAMSJA19

JH#1203|@JOHANNES_HAGE

AK#909|@ANDREY_KURENKOV

EL#767|@ELIEBAKOUCH

VW#691|@VINCENTWEISSER

WB#335|@WILLCCBB

PI#33|@PRIMEINTELLECT

100

Figure AI robots complete 24-hour autonomous sorting run

In a test run, Figure AI's humanoid robots operated autonomously for 24 hours while sorting packages. The three robots, Bob, Frank and Gary, handled more than 28,000 items on the Helix-02 platform with no human intervention.

15h9.4k619663.7k

MO|@MOLLYSOSHEA

MB#1769|@MATTHEWBERMAN

CH#1494|@KIMMONISMUS

VK#1484|@VIKASHPLUS

B(#711|@BEFFJEZOS

CL#687|@COREYLYNCH

BA#628|@ADCOCK_BRETT

110

AI researchers revive visualizations from 2009 LeCun paper

Researchers have shared a 16-by-16 grid of neural network filters from a 2009 paper, showing simple learned patterns such as stripes, rings, and edge detectors. The work by Koray Kavukcuoglu, Marc'Aurelio Ranzato, Rob Fergus, and Yann LeCun examined topographic organization of features and later influenced neuroscience studies.

16h6.7k878412.1k

RM|@ROLANDMEMISEVIC

LA#1900|@LUCAAMB

AP#1754|@ALEX_PEYS

SX#173|@SAININGXIE

TC#121|@TACOCOHEN

SD#75|@SEDIELEM

120

Cerebras completes Nasdaq IPO valued above $100 billion

Cerebras completed its Nasdaq IPO under ticker CBRS, with shares opening at $350 and delivering a valuation above $100 billion. The listing ranks as the largest US tech IPO since Uber as the company advances its wafer-scale AI chips.

13h1.1k119105.3k

EV|@ERICVISHRIA

RZ#1452|@REZA_ZADEH

SC#1404|@MILKSANDMATCHA

AV#1277|@ALTRYNE

VC#1264|@VITALIYCHILEY

JC#769|@JOHNCOOGAN

B(#711|@BEFFJEZOS

RS#309|@SCOBLEIZER

NB#237|@NATHANBENAICH

SG#200|@SARANORMOUS

130

US approves Nvidia H200 chip sales to ten Chinese firms

The US approved Nvidia H200 AI chip sales to ten Chinese companies including Alibaba, ByteDance, and Tencent. The approvals cover roughly 750,000 units, with no shipments expected before the fourth quarter of 2026.

23h2k287312.9k

CM|@CHRISRMCGUIRE

CH#1494|@KIMMONISMUS

ZE#1479|@ZEPHYR_Z9

PW#1265|@PETERWILDEFORD

NS#407|@NOAHPINION

T(#400|@TEORTAXESTEX

AK#240|@BLACKHC

140

LangChain launches LangChain Labs for continual AI learning

LangChain has started LangChain Labs to develop continual learning methods that let AI agents adapt and improve from ongoing use without forgetting prior knowledge. The effort relies on LangSmith to turn production tracing data into targeted model updates and evaluations, beginning with partners including Harvey and Nvidia.

13h807426133.4k

JS|@J_SCHOTTENSTEIN

BR|@BRACESPROUL

C🇨🇦|@LLM_WIZARD

HC#702|@HWCHASE17

WB#335|@WILLCCBB

PI#33|@PRIMEINTELLECT

LA#23|@LANGCHAIN

150

Sony announces AI Camera Assistant for Xperia 1 VIII

Sony's AI Camera Assistant on the Xperia 1 VIII recommends camera settings to help users capture better photos. The tool examines subject, scene, and weather data to suggest adjustments for color balance, exposure, bokeh, and lens selection.

15h3.3k639789.8k

TE#1984|@TENOBRUS

CH#1494|@KIMMONISMUS

AA#1345|@AUSTEN

@L#803|@LEVELSIO

AM#285|@AIDAN_MCLAU

SZ#62|@SUCHENZANG

160

World Labs developer open-sources image-blaster 3D toolkit

A World Labs developer has open-sourced image-blaster, a toolkit that converts a single photo into a fully meshed 3D scene with physics, lighting, and audio within minutes. It uses Claude Code agents alongside tools from Marble and fal.ai to produce editable scene.json files and interactive desktop views.

10h2.4k2.5k179.1k

LW|@LIMINAL_WARMTH

GO#1397|@GOWTHAMI_S

DF#882|@DANIELLEFONG

WL#54|@THEWORLDLABS

FL#13|@DRFEIFEI

170

Closing arguments begin in Elon Musk OpenAI lawsuit

Closing arguments are underway in Elon Musk's lawsuit against OpenAI, which accuses the company of abandoning its nonprofit origins in favor of commercial deals. A federal judge in Oakland first gave the jury instructions before lawyers from both sides delivered their final statements, with Microsoft also named in the case.

15h6.2k821513.7k

HF|@HAYDENFIELD

RK#1508|@MIKEISAAC

AA#1345|@AUSTEN

AV#1277|@ALTRYNE

180

Nous Research adds Codex runtime to Hermes Agent

Hermes Agent can now run code execution, file operations, and shell commands for OpenAI models instead of returning only text. A single toggle routes tool calls through the Codex CLI using an existing ChatGPT subscription.

14h3.7k2.2k6.1M

HA|@HERMESAGENTTIPS

A||@ABDULLAH4AI

AV#1277|@ALTRYNE

T🪽#256|@TEKNIUM

NR#39|@NOUSRESEARCH

190

ChatGPT app adds Codex access on mobile devices

Codex is now accessible inside the ChatGPT mobile app, giving users desktop-style project management and thread access from their phones. Connecting the desktop app in roughly one minute enables cross-device sync for folders and remote actions such as file searches or video rendering.

11h1.5k38998.3k

V(#1815|@REACH_VB

CH#1494|@KIMMONISMUS

AD#965|@THEREALADAMG

200

Anthropic CFO Krishna Rao details revenue growth to $30 billion

Anthropic grew its run-rate revenue from $250 million to $30 billion in two years while raising roughly $75 billion. The finance team now relies on 70 custom Claude skills that complete most financial statements and monthly reviews before human review.

16h2.5k2.1k1.9M

MG|@MEGRAEGRADY

TP#1676|@TRUNGTPHAN

PO#1507|@PATRICK_OSHAG

LR#1017|@LENNYSAN

210

AsymFlow achieves 1.57 FID on ImageNet in pixel space

AsymFlow sets a new record with 1.57 FID on ImageNet for pixel-space flow models. It uses low-rank velocity parameterization to generate images directly in pixels without a VAE and improves finetuned FLUX models with sharper results and 40 percent faster speed.

18h67150474.8k

GW|@GORDONWETZSTEIN

LA#1900|@LUCAAMB

KD#954|@CSPROFKGD

AV#911|@ARASHVAHDAT

VS#745|@VINCESITZMANN

AJ#471|@JM_ALEXIA

220

ZyphraAI converts ZAYA1-8B-base model to diffusion LLM

ZyphraAI converted its existing ZAYA1-8B-base autoregressive model into a diffusion LLM through mid-training instead of training from scratch. The method diffuses 16-token blocks in one step from a mask prior and mixes logits with speculative decoding to improve efficiency.

9h653216613.8k

SJ|@FUTUREJURVETSON

B(#999|@ANDERSONBCDEFG

AG#913|@ANIRUDHG9119

KA#841|@KALOMAZE

T(#400|@TEORTAXESTEX

ZY#242|@ZYPHRAAI

EM#158|@EMOSTAQUE

230

Datadog releases Toto 2.0 time series foundation models

Datadog released Toto 2.0, a family of open-weights time series foundation models from 4 million to 2.5 billion parameters that show consistent performance gains with scale. The models top the BOOM, GIFT-Eval, and TIME benchmarks under a single training configuration and are available on Hugging Face under Apache 2.0.

16h80245680.1k

AT#1872|@ATALWALKAR

BI#1483|@BILALTWOVEC

AC#261|@ANDREW_N_CARR

C🤗#70|@CLEMENTDELANGUE

TD#56|@TIM_DETTMERS

240

Senator Jim Banks sends AI strategy letter to Trump

Senator Jim Banks has urged the Trump administration to create an AI strategy that preserves US technological leadership over China while adding concrete safety measures. The letter calls for those measures to be raised in upcoming talks with Xi Jinping, extending focus beyond cyber risks to military, biosecurity, and loss-of-control scenarios.

13h3539276.5k

CM|@CHRISRMCGUIRE

SJ|@SENATORBANKS

CB|@CHARLIEBULL0CK

NC|@_NATHANCALVIN

JS#1811|@JORDANSCHNEIDER

DE#1792|@DANIEL_271828

PW#1265|@PETERWILDEFORD

AC#483|@ANDREWCURRAN_

MB#16|@MILES_BRUNDAGE

250

Raindrop AI releases Workshop local AI debugging tool

Raindrop AI released Workshop, an open-source tool that lets developers debug AI agents entirely on their local machines without sending data to external servers. The MIT-licensed project streams tokens, tool calls, spans, durations, and costs in real time to a dashboard on localhost:5899 and supports Claude and Codex models.

12h1.2k1.1k160.5k

PR|@PRATYUSH_R8

JC|@JACOBCOLLING

BH#1858|@BENHYLAK

260

OpenAI Codex draws attention from social media accounts

AI-focused social media accounts are highlighting OpenAI's Codex model after users reported strong results on coding tasks. One newsletter co-founder stated it has fully replaced Claude in their workflow, with separate posts noting a major update that expands developer capabilities.

15h10.7k169294.6k

C🥑|@COLDHEALING

🍓🍓#1660|@IRULETHEWORLDMO

“P#1579|@PAULARAMBLES

CH#1494|@KIMMONISMUS

AL#1087|@ALTH0U

270

John Carmack shares PyTorch squared distance code

John Carmack shared a PyTorch routine that computes squared Euclidean distances between batch and history embedding sets without allocating large intermediate tensors. The approach precomputes vector norms and uses a single matrix multiply to obtain the cross terms.

10h1.8k590134.7k

T(#400|@TEORTAXESTEX

AK#240|@BLACKHC

JC#116|@ID_AA_CARMACK

LB#53|@GIFFMANA

280

Discussions parallel Monet critiques with AI art reactions

Online discussions compare 19th-century criticisms of Monet's Impressionist paintings as sloppy with similar complaints about AI-generated art today. Users refined an AI image in Monet's style by compiling community feedback on color, composition, and texture into a revised prompt.

13h3.3k305275.9k

UT|@SKYESHARKIE

ND#1725|@NICKADOBOS

J⧉#499|@REPLIGATE

FF#331|@FRANCOISFLEURET

290

NYU professor Andrew Gordon Wilson suggests ICML 2010 nostalgia event

NYU professor Andrew Gordon Wilson proposed an ICML 2010 nostalgia event to judge current AI papers by earlier technical standards. The format would use AI-generated submissions reviewed through 2010-era debates on methods such as MCMC and variational inference.

9h3136346.5k

ST#1478|@_ONIONESQUE

KC#1104|@KROSCOO

X(#752|@XUANALOGUE

DP#203|@DIMITRISPAPAIL

AG#168|@ANDREWGWILS

300

Anthropic loses hundreds of Claude Pro and Max subscriptions

After Anthropic added rate limits to its Claude Code tools that broke many third-party integrations, hundreds of users canceled Pro and Max subscriptions. Thread analysis estimates roughly 350 cancellations and about $23,800 in monthly revenue lost.

16h2.7k635337.1k

B(#711|@BEFFJEZOS

KA#464|@YACINEMTB

Yesterday's Top Stories, May 14, 2026.

Posts: 3,998|Clusters: 870|Stories frozen at midnight PT

10

OpenAI releases Codex preview in ChatGPT mobile app

OpenAI has released a preview of Codex inside the ChatGPT mobile app on iOS and Android. Users can start coding tasks, review outputs, and steer execution from their phone while the session continues on a connected laptop, Mac mini, or devbox with full project context preserved.

12h38.3k6.9k4.9M

#1 QUOTED2.4k

FO#1953|@FOUADMATIN

TC#1771|@TIFAFAFAFA

BS#1742|@BILAWALSIDHU

ND#1725|@NICKADOBOS

CH#1494|@KIMMONISMUS

BI#1483|@BILALTWOVEC

AV#1277|@ALTRYNE

YO#1138|@YOHEINAKAJIMA

AD#965|@THEREALADAMG

JA#822|@APPLES_JIMMY

RH#756|@ROMAINHUET

TI#738|@THSOTTIAUX

TM#371|@ISCIENCELUVR

ER#338|@ERICMITCHELLAI

BP#337|@BORISMPOWER

RA#319|@RAPHA_GL

OD#45|@OPENAIDEVS

CH#32|@CHATGPTAPP

NB#27|@POLYNOAMIAL

20

Anthropic publishes scenarios for 2028 global AI leadership

Anthropic's analysis shows that strengthened US export controls on advanced chips could preserve the current lead of the United States and its allies in frontier AI systems through 2028. The paper contrasts this outcome with a scenario in which relaxed restrictions allow Chinese labs to close the gap via talent recruitment and distillation.

12h6.9k3.9k2.9M

HS|@___HARALD___

SH|@SHAKEELHASHIM

TK|@FIRSTADOPTER

JS#1811|@JORDANSCHNEIDER

MB#1769|@MATTHEWBERMAN

AK#1665|@ANSHULKUNDAJE

SÓ#1394|@S_OHEIGEARTAIGH

SH#1366|@HAMANDCHEESE

RO#1098|@KRISHNANROHIT

JL#1049|@JEFFLADISH

RP#1014|@ROHANPAUL_AI

LA#984|@SCALING01

DF#882|@DANIELLEFONG

CP#755|@CHRIS_J_PAXTON

NS#407|@NOAHPINION

T(#400|@TEORTAXESTEX

DR#113|@DELIPRAO

AN#3|@ANTHROPICAI

30

xAI releases early beta of Grok Build CLI

xAI has opened an early beta of Grok Build, an agentic command-line tool for coding, building apps, and automating workflows. Access is limited to SuperGrok Heavy subscribers via x.ai/cli, with feedback used to refine the product.

12h86.4k8.6k76M

#1 VIEWED76M#1 LIKED86.4k#1 BOOKMARKED8.6k

CH|@CHANGRAN_HU

HZ|@ZHU_HANQIN41424

GA|@GAURAVISNOTME

ZH#1058|@ZHITINGHU

B(#711|@BEFFJEZOS

BY#379|@BILLYUCHENLIN

EM#65|@ELONMUSK

40

arXiv clarifies one-year ban for unchecked LLM submissions

arXiv now enforces a one-year submission ban on papers containing unverified LLM-generated content such as hallucinated references. Authors must first secure acceptance at a reputable peer-reviewed venue before they can resubmit.

11h4.8k385350.5k

KC|@KEENANISALIVE

PL|@PANGRAMLABS

PL#1693|@PRIYALD17

AK#1665|@ANSHULKUNDAJE

AK#1620|@DR_ATOOSA

ST#1478|@_ONIONESQUE

AS#1405|@STUHLMUELLER

BG#1364|@BEN_GOLUB

TC#1189|@TUHINCHAKR

FB#1139|@XEOPHON

AW#967|@ANDREWWHITE01

MR#956|@MENGYER

CT#663|@CHENHAOTAN

TR#518|@TALIARINGER

YD#375|@YUNTIANDENG

DP#203|@DIMITRISPAPAIL

DR#113|@DELIPRAO

TG#72|@TDIETTERICH

DR#60|@ROYDANROY

LB#53|@GIFFMANA

50

Sanders and Ocasio-Cortez introduce AI data center moratorium bill

Senators Bernie Sanders and Alexandria Ocasio-Cortez introduced legislation that would pause new large-scale AI data center construction until Congress enacts rules on safety, workers, and the environment. The move follows more than 300 state and local bills this year, which have already delayed or canceled roughly half of planned 2026 projects.

17h83.8k7k35.7M

#1 COMMENTS17.7k

ND|@NICK_DAVIDOV

AP|@APOSTRUCTURA

NS|@NINADSCHICK

DJ#1929|@DAN_JEFFRIES1

FC#1085|@ALTIMOR

B(#711|@BEFFJEZOS

PD#649|@PMDDOMINGOS

GT#230|@GARRYTAN

EM#65|@ELONMUSK

60

Goodfire AI researchers identify shape-rotating calculator in language models

Researchers at Goodfire AI have uncovered a mechanism inside large language models that carries out addition by rotating shapes within high-dimensional neural space. The structure connects to Fourier features and supports similar geometric operations for non-math tasks as well.

14h4.9k2.5k756.7k

SF|@SHERIDAN_FEUCHT

EH|@ERICHO_GOODFIRE

SD#1823|@INDUCTIONHEADS

TM#1658|@BANBURISMUS_

PR#1262|@8TEAPI

SH#1247|@SONYATWEETYBIRD

E(#1119|@EV_FEDORENKO

LA#984|@SCALING01

DF#882|@DANIELLEFONG

J⧉#499|@REPLIGATE

YD#375|@YUNTIANDENG

CP#244|@CHRISGPOTTS

GO#48|@GOODFIREAI

70

Thomas G. Dietterich reminds arXiv authors of content responsibility

Thomas G. Dietterich reminded researchers that every author on an arXiv paper bears full responsibility for all contents, including AI-generated sections. The exchange also raised questions about enforcement steps such as one-year bans when papers contain hallucinated references.

11h3.7k646459.3k

AK#1620|@DR_ATOOSA

AW#967|@ANDREWWHITE01

WI#906|@WSISAAC

RS#659|@ZIV_RAVID

XE#455|@XWANG_LK

YA#227|@YOAVARTZI

EM#155|@EMOLLICK

DR#113|@DELIPRAO

TG#72|@TDIETTERICH

LB#53|@GIFFMANA

80

Calif researchers discover macOS kernel vulnerabilities using Mythos

California researchers found two previously unknown macOS kernel vulnerabilities while testing an early build of Anthropic’s Mythos AI. They chained the flaws into a working privilege escalation exploit that bypassed hardware protections, completing the work in five days before submitting a 55-page report to Apple.

15h4.2k849619.2k

N🔍|@NATALIA__COELHO

PR|@DEREDLERITT3R

ZE#1479|@ZEPHYR_Z9

MB#1306|@MATTHEWJBAR

RP#1014|@ROHANPAUL_AI

AD#897|@DORIALEXANDER

AC#483|@ANDREWCURRAN_

NS#407|@NOAHPINION

EM#155|@EMOLLICK

BB#127|@BOAZBARAKTCS

RA#82|@_AROHAN_

90

Prime Intellect agents improve nanoGPT record to 2,930 steps

Prime Intellect's AI agents set a new record on the nanoGPT benchmark by reaching target loss in 2,930 training steps, beating the prior human baseline of 2,990. The agents completed roughly 10,000 autonomous experiments over two weeks using 14,000 H200 GPU hours of idle compute.

7h1.9k656231k

DO|@DOMINIK_SCHERM

JW#1610|@JIAXINWEN22

JA#1587|@JULIUSADML

DA#1342|@DAMEKDAVIS

SA#1281|@SAMSJA19

JH#1203|@JOHANNES_HAGE

AK#909|@ANDREY_KURENKOV

EL#767|@ELIEBAKOUCH

VW#691|@VINCENTWEISSER

WB#335|@WILLCCBB

PI#33|@PRIMEINTELLECT

100

Figure AI robots complete 24-hour autonomous sorting run

In a test run, Figure AI's humanoid robots operated autonomously for 24 hours while sorting packages. The three robots, Bob, Frank and Gary, handled more than 28,000 items on the Helix-02 platform with no human intervention.

15h9.4k619663.7k

MO|@MOLLYSOSHEA

MB#1769|@MATTHEWBERMAN

CH#1494|@KIMMONISMUS

VK#1484|@VIKASHPLUS

B(#711|@BEFFJEZOS

CL#687|@COREYLYNCH

BA#628|@ADCOCK_BRETT

110

AI researchers revive visualizations from 2009 LeCun paper

Researchers have shared a 16-by-16 grid of neural network filters from a 2009 paper, showing simple learned patterns such as stripes, rings, and edge detectors. The work by Koray Kavukcuoglu, Marc'Aurelio Ranzato, Rob Fergus, and Yann LeCun examined topographic organization of features and later influenced neuroscience studies.

16h6.7k878412.1k

RM|@ROLANDMEMISEVIC

LA#1900|@LUCAAMB

AP#1754|@ALEX_PEYS

SX#173|@SAININGXIE

TC#121|@TACOCOHEN

SD#75|@SEDIELEM

120

Cerebras completes Nasdaq IPO valued above $100 billion

Cerebras completed its Nasdaq IPO under ticker CBRS, with shares opening at $350 and delivering a valuation above $100 billion. The listing ranks as the largest US tech IPO since Uber as the company advances its wafer-scale AI chips.

13h1.1k119105.3k

EV|@ERICVISHRIA

RZ#1452|@REZA_ZADEH

SC#1404|@MILKSANDMATCHA

AV#1277|@ALTRYNE

VC#1264|@VITALIYCHILEY

JC#769|@JOHNCOOGAN

B(#711|@BEFFJEZOS

RS#309|@SCOBLEIZER

NB#237|@NATHANBENAICH

SG#200|@SARANORMOUS

130

US approves Nvidia H200 chip sales to ten Chinese firms

The US approved Nvidia H200 AI chip sales to ten Chinese companies including Alibaba, ByteDance, and Tencent. The approvals cover roughly 750,000 units, with no shipments expected before the fourth quarter of 2026.

23h2k287312.9k

CM|@CHRISRMCGUIRE

CH#1494|@KIMMONISMUS

ZE#1479|@ZEPHYR_Z9

PW#1265|@PETERWILDEFORD

NS#407|@NOAHPINION

T(#400|@TEORTAXESTEX

AK#240|@BLACKHC

140

LangChain launches LangChain Labs for continual AI learning

LangChain has started LangChain Labs to develop continual learning methods that let AI agents adapt and improve from ongoing use without forgetting prior knowledge. The effort relies on LangSmith to turn production tracing data into targeted model updates and evaluations, beginning with partners including Harvey and Nvidia.

13h807426133.4k

JS|@J_SCHOTTENSTEIN

BR|@BRACESPROUL

C🇨🇦|@LLM_WIZARD

HC#702|@HWCHASE17

WB#335|@WILLCCBB

PI#33|@PRIMEINTELLECT

LA#23|@LANGCHAIN

150

Sony announces AI Camera Assistant for Xperia 1 VIII

Sony's AI Camera Assistant on the Xperia 1 VIII recommends camera settings to help users capture better photos. The tool examines subject, scene, and weather data to suggest adjustments for color balance, exposure, bokeh, and lens selection.

15h3.3k639789.8k

TE#1984|@TENOBRUS

CH#1494|@KIMMONISMUS

AA#1345|@AUSTEN

@L#803|@LEVELSIO

AM#285|@AIDAN_MCLAU

SZ#62|@SUCHENZANG

160

World Labs developer open-sources image-blaster 3D toolkit

A World Labs developer has open-sourced image-blaster, a toolkit that converts a single photo into a fully meshed 3D scene with physics, lighting, and audio within minutes. It uses Claude Code agents alongside tools from Marble and fal.ai to produce editable scene.json files and interactive desktop views.

10h2.4k2.5k179.1k

LW|@LIMINAL_WARMTH

GO#1397|@GOWTHAMI_S

DF#882|@DANIELLEFONG

WL#54|@THEWORLDLABS

FL#13|@DRFEIFEI

170

Closing arguments begin in Elon Musk OpenAI lawsuit

Closing arguments are underway in Elon Musk's lawsuit against OpenAI, which accuses the company of abandoning its nonprofit origins in favor of commercial deals. A federal judge in Oakland first gave the jury instructions before lawyers from both sides delivered their final statements, with Microsoft also named in the case.

15h6.2k821513.7k

HF|@HAYDENFIELD

RK#1508|@MIKEISAAC

AA#1345|@AUSTEN

AV#1277|@ALTRYNE

180

Nous Research adds Codex runtime to Hermes Agent

Hermes Agent can now run code execution, file operations, and shell commands for OpenAI models instead of returning only text. A single toggle routes tool calls through the Codex CLI using an existing ChatGPT subscription.

14h3.7k2.2k6.1M

HA|@HERMESAGENTTIPS

A||@ABDULLAH4AI

AV#1277|@ALTRYNE

T🪽#256|@TEKNIUM

NR#39|@NOUSRESEARCH

190

ChatGPT app adds Codex access on mobile devices

Codex is now accessible inside the ChatGPT mobile app, giving users desktop-style project management and thread access from their phones. Connecting the desktop app in roughly one minute enables cross-device sync for folders and remote actions such as file searches or video rendering.

11h1.5k38998.3k

V(#1815|@REACH_VB

CH#1494|@KIMMONISMUS

AD#965|@THEREALADAMG

200

Anthropic CFO Krishna Rao details revenue growth to $30 billion

Anthropic grew its run-rate revenue from $250 million to $30 billion in two years while raising roughly $75 billion. The finance team now relies on 70 custom Claude skills that complete most financial statements and monthly reviews before human review.

16h2.5k2.1k1.9M

MG|@MEGRAEGRADY

TP#1676|@TRUNGTPHAN

PO#1507|@PATRICK_OSHAG

LR#1017|@LENNYSAN

210

AsymFlow achieves 1.57 FID on ImageNet in pixel space

AsymFlow sets a new record with 1.57 FID on ImageNet for pixel-space flow models. It uses low-rank velocity parameterization to generate images directly in pixels without a VAE and improves finetuned FLUX models with sharper results and 40 percent faster speed.

18h67150474.8k

GW|@GORDONWETZSTEIN

LA#1900|@LUCAAMB

KD#954|@CSPROFKGD

AV#911|@ARASHVAHDAT

VS#745|@VINCESITZMANN

AJ#471|@JM_ALEXIA

220

ZyphraAI converts ZAYA1-8B-base model to diffusion LLM

ZyphraAI converted its existing ZAYA1-8B-base autoregressive model into a diffusion LLM through mid-training instead of training from scratch. The method diffuses 16-token blocks in one step from a mask prior and mixes logits with speculative decoding to improve efficiency.

9h653216613.8k

SJ|@FUTUREJURVETSON

B(#999|@ANDERSONBCDEFG

AG#913|@ANIRUDHG9119

KA#841|@KALOMAZE

T(#400|@TEORTAXESTEX

ZY#242|@ZYPHRAAI

EM#158|@EMOSTAQUE

230

Datadog releases Toto 2.0 time series foundation models

Datadog released Toto 2.0, a family of open-weights time series foundation models from 4 million to 2.5 billion parameters that show consistent performance gains with scale. The models top the BOOM, GIFT-Eval, and TIME benchmarks under a single training configuration and are available on Hugging Face under Apache 2.0.

16h80245680.1k

AT#1872|@ATALWALKAR

BI#1483|@BILALTWOVEC

AC#261|@ANDREW_N_CARR

C🤗#70|@CLEMENTDELANGUE

TD#56|@TIM_DETTMERS

240

Senator Jim Banks sends AI strategy letter to Trump

Senator Jim Banks has urged the Trump administration to create an AI strategy that preserves US technological leadership over China while adding concrete safety measures. The letter calls for those measures to be raised in upcoming talks with Xi Jinping, extending focus beyond cyber risks to military, biosecurity, and loss-of-control scenarios.

13h3539276.5k

CM|@CHRISRMCGUIRE

SJ|@SENATORBANKS

CB|@CHARLIEBULL0CK

NC|@_NATHANCALVIN

JS#1811|@JORDANSCHNEIDER

DE#1792|@DANIEL_271828

PW#1265|@PETERWILDEFORD

AC#483|@ANDREWCURRAN_

MB#16|@MILES_BRUNDAGE

250

Raindrop AI releases Workshop local AI debugging tool

Raindrop AI released Workshop, an open-source tool that lets developers debug AI agents entirely on their local machines without sending data to external servers. The MIT-licensed project streams tokens, tool calls, spans, durations, and costs in real time to a dashboard on localhost:5899 and supports Claude and Codex models.

12h1.2k1.1k160.5k

PR|@PRATYUSH_R8

JC|@JACOBCOLLING

BH#1858|@BENHYLAK

260

OpenAI Codex draws attention from social media accounts

AI-focused social media accounts are highlighting OpenAI's Codex model after users reported strong results on coding tasks. One newsletter co-founder stated it has fully replaced Claude in their workflow, with separate posts noting a major update that expands developer capabilities.

15h10.7k169294.6k

C🥑|@COLDHEALING

🍓🍓#1660|@IRULETHEWORLDMO

“P#1579|@PAULARAMBLES

CH#1494|@KIMMONISMUS

AL#1087|@ALTH0U

270

John Carmack shares PyTorch squared distance code

John Carmack shared a PyTorch routine that computes squared Euclidean distances between batch and history embedding sets without allocating large intermediate tensors. The approach precomputes vector norms and uses a single matrix multiply to obtain the cross terms.

10h1.8k590134.7k

T(#400|@TEORTAXESTEX

AK#240|@BLACKHC

JC#116|@ID_AA_CARMACK

LB#53|@GIFFMANA

280

Discussions parallel Monet critiques with AI art reactions

Online discussions compare 19th-century criticisms of Monet's Impressionist paintings as sloppy with similar complaints about AI-generated art today. Users refined an AI image in Monet's style by compiling community feedback on color, composition, and texture into a revised prompt.

13h3.3k305275.9k

UT|@SKYESHARKIE

ND#1725|@NICKADOBOS

J⧉#499|@REPLIGATE

FF#331|@FRANCOISFLEURET

290

NYU professor Andrew Gordon Wilson suggests ICML 2010 nostalgia event

NYU professor Andrew Gordon Wilson proposed an ICML 2010 nostalgia event to judge current AI papers by earlier technical standards. The format would use AI-generated submissions reviewed through 2010-era debates on methods such as MCMC and variational inference.

9h3136346.5k

ST#1478|@_ONIONESQUE

KC#1104|@KROSCOO

X(#752|@XUANALOGUE

DP#203|@DIMITRISPAPAIL

AG#168|@ANDREWGWILS

300

Anthropic loses hundreds of Claude Pro and Max subscriptions

After Anthropic added rate limits to its Claude Code tools that broke many third-party integrations, hundreds of users canceled Pro and Max subscriptions. Thread analysis estimates roughly 350 cancellations and about $23,800 in monthly revenue lost.

16h2.7k635337.1k

B(#711|@BEFFJEZOS

KA#464|@YACINEMTB