digg/ai
digg/ai
StoriesGitHubRankings

Digg - AI news, before it trends

Highlights

??
IN CASE YOU MISSED IT

Dwarkesh Patel separates intelligence from goal achievement

Sayash KapoorSK#734|@SAYASHK
Séb KrierSK#505|@SEBKRIER
Dean W. BallDW#392|@DEANWBALL
EmadEM#167|@EMOSTAQUE
Dwarkesh PatelDP#70|@DWARKESH_SP
roonRO#59|@TSZZL
#1 VIEWED (2.7M)

xAI adds native image and video generation to its Grok Build CLI tool

Beff (e/acc)B(#839|@BEFFJEZOS
Elon MuskEM#76|@ELONMUSK
#1 QUOTED (211)

Boston Dynamics demonstrates Atlas robot lifting and carrying mini-fridge with whole-body coordination

Jimmy Apples 🍎/accJA#837|@APPLES_JIMMY
Shuran SongSS#792|@SONGSHURAN
Chris PaxtonCP#732|@CHRIS_J_PAXTON
Bojan TunguzBT#687|@TUNGUZ
Andrew CurranAC#517|@ANDREWCURRAN_
Elon MuskEM#76|@ELONMUSK
#1 BOOKMARKED (2.9k)

Leopold’s Situational Awareness discloses new long positions in TE, HIVE, SHAZ, INTC, AMD, and other semiconductor names

Molly O’SheaMO|@MOLLYSOSHEA
Austen AllredAA#1540|@AUSTEN
Lisan al GaibLA#980|@SCALING01
FASTEST CLIMBING (+4)

Analysis matches Granta short story to GPT-4o output after 100% Pangram AI score

Nabeel S. QureshiNS#1397|@NABEELQU
rohitRO#1220|@KRISHNANROHIT
Tuhin ChakrabartyTC#1050|@TUHINCHAKR
Andrew CurranAC#517|@ANDREWCURRAN_
RISING LIKES (1.1k)

OpenAI runs simultaneous A/B tests on updated text and image models inside ChatGPT

Bindu ReddyBR#1622|@BINDUREDDY
Chubby♨️CH#1496|@KIMMONISMUS
RISING DISCUSSION (32)

Jason Liu from OpenAI questions why AI labs have not addressed water consumption claims

jasonJA#929|@JXNLCO
Bojan TunguzBT#687|@TUNGUZ

Top Stories.

Posts: —|Clusters: —|Next crawl: —
10

Boston Dynamics demonstrates Atlas robot lifting and carrying mini-fridge with whole-body coordination

Boston Dynamics' Atlas robot can lift a mini-fridge from a block and carry it across a lab while shifting its grip and posture to maintain balance. The black-and-white humanoid learned the motions through reinforcement learning trained across millions of simulated variations in mass, friction, and positioning.

3h5.9k748580.3k
#1 QUOTED211#1 ENGAGED8
@jason@J#1061|@JASON
Rohan PaulRP#1032|@ROHANPAUL_AI
Jimmy Apples 🍎/accJA#837|@APPLES_JIMMY
Shuran SongSS#792|@SONGSHURAN
Chris PaxtonCP#732|@CHRIS_J_PAXTON
Bojan TunguzBT#687|@TUNGUZ
Andrew CurranAC#517|@ANDREWCURRAN_
Elon MuskEM#76|@ELONMUSK
20

Reese Witherspoon posts kitchen video urging mothers to learn basic AI concepts

Reese Witherspoon posted a 55-second kitchen video urging mothers to learn basic AI concepts, noting that only three in ten women currently use the tools. Y Combinator CEO Garry Tan shared the clip and pointed viewers to it after the New York Times published a critical op-ed.

2h2.7k571337.1k
Kim-Mai CutlerKC#1670|@KIMMAICUTLER
@jason@J#1061|@JASON
Bojan TunguzBT#687|@TUNGUZ
Garry TanGT#270|@GARRYTAN
30

Dimitris Papailiopoulos from Microsoft Research shares ECHO method for training CLI agents with environment prediction loss

The ECHO method improves how command-line agents learn by adding an environment prediction loss to standard reinforcement learning. It trains on both agent actions and terminal responses in one rollout and forward pass rather than masking outputs, raising benchmark scores on Qwen3 models.

2h36131764.5k
Chenhao TanCT#570|@CHENHAOTAN
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
John LangfordJL#332|@JOHNCLANGFORD
Dimitris PapailiopoulosDP#197|@DIMITRISPAPAIL
40

Leopold’s Situational Awareness discloses new long positions in TE, HIVE, SHAZ, INTC, AMD, and other semiconductor names

Leopold Aschenbrenner’s Situational Awareness fund disclosed new long positions in semiconductor stocks including TE, INTC, AMD, TSM, and NVDA in its latest 13F filing. The report also shows large new put options on several of the same names plus added stakes in Bitcoin mining companies.

4h2.7k2.9k588.3k
#1 BOOKMARKED2.9k
Molly O’SheaMO|@MOLLYSOSHEA
Austen AllredAA#1540|@AUSTEN
Lisan al GaibLA#980|@SCALING01
50

xAI adds native image and video generation to its Grok Build CLI tool

xAI has integrated native image and video generation into its Grok Build CLI tool for direct media creation from the command line. The update adds commands like /imagine and /imagine-video, marking the first such interface with built-in support for both formats.

8h10.4k5142.7M
#1 VIEWED2.7M#1 LIKED10.4k#1 COMMENTS1.7kMOST POSITIVE360
Beff (e/acc)B(#839|@BEFFJEZOS
Elon MuskEM#76|@ELONMUSK
60

Logan Kilpatrick questions why large language models lack signals for out-of-distribution inputs

Large language models can produce unreliable answers on unfamiliar queries without any warning to users because they lack signals for inputs outside their training data. Engineer Logan Kilpatrick highlighted the gap, with replies noting that outlier detection would incorrectly flag most normal requests as errors.

1h7496749.5k
Nick DobosND#1894|@NICKADOBOS
Bojan TunguzBT#687|@TUNGUZ
elvisEL#475|@OMARSAR0
Logan KilpatrickLK#90|@OFFICIALLOGANK
74

Analysis matches Granta short story to GPT-4o output after 100% Pangram AI score

A short story published in Granta after winning the Caribbean regional Commonwealth Short Story Prize shows signs of AI generation. An excerpt scored 100% AI on the Pangram detector and matched GPT-4o phrasing patterns including repetitive structures.

42m1603424.7k
FASTEST CLIMBING+4
Nabeel S. QureshiNS#1397|@NABEELQU
rohitRO#1220|@KRISHNANROHIT
Tuhin ChakrabartyTC#1050|@TUHINCHAKR
Andrew CurranAC#517|@ANDREWCURRAN_
81

Thibault Sottiaux, OpenAI Codex engineering lead, reports 99.98% uptime with only 8 minutes of sleep per month

OpenAI's Codex maintains 99.98 percent uptime, restricting total downtime to roughly eight minutes each month. This metric supports steady access for users running agentic coding tasks, though one reply described the brief outages as unintentional.

9h2.7k6999.9k
Vaibhav (VB) SrivastavV(#1853|@REACH_VB
TiboTI#768|@THSOTTIAUX
kacheKA#488|@YACINEMTB
91

Hermes Agent Kanban adds automated task decomposition and profile assignment

Users of Hermes Agent Kanban can now drop one prompt into the triage column and have an orchestrator agent automatically split it into subtasks assigned to matching profiles. The change is live in pull request 27572 on the NousResearch/hermes-agent repository and keeps the existing SQLite-backed board.

8h1.5k1.3k126.9k
Robert ScobleRS#321|@SCOBLEIZER
Teknium 🪽T🪽#259|@TEKNIUM
101

Vatican announces Pope Leo XIV encyclical Magnifica humanitas on AI for May 25 release

Pope Leo XIV’s first encyclical will address how to preserve human values amid rapid progress in artificial intelligence. The document, called Magnifica humanitas, is scheduled for release on May 25 with a presentation at the Vatican’s Synod Hall.

1h1121513.1k
Seán Ó hÉigeartaighSÓ#1457|@S_OHEIGEARTAIGH
Yo ShavitYS#638|@YONASHAV
Andrew CurranAC#517|@ANDREWCURRAN_
Chris OlahCO#21|@CH402
111

Unsloth AI releases Multi-Token Prediction GGUF files for Qwen3.6 models

Users can now run Qwen3.6 models up to 2.2 times faster on standard hardware with new GGUF files from Unsloth AI. The files deliver 160–240 tokens per second for the 27B and 35B-A3B variants while fitting in 18 GB RAM and keeping full accuracy.

2h82651732.7k
Georgi GerganovGG#781|@GGERGANOV
Daniel HanDH#772|@DANIELHANCHEN
12NEW

Cloudflare security team concludes patch acceleration inadequate after testing Anthropic Mythos AI

Cloudflare's security team concluded that faster patching alone cannot adequately address vulnerabilities uncovered by AI tools. Their evaluation ran Anthropic's Mythos model against fifty internal code repositories over several weeks.

1h73251291.1k
Daniel JeffriesDJ#1988|@DAN_JEFFRIES1
Brendan Dolan-GavittBD#812|@MOYIX
131

OpenAI runs simultaneous A/B tests on updated text and image models inside ChatGPT

OpenAI has been running simultaneous A/B tests of updated text and image models inside ChatGPT. The tests reference GPT-5.5 as the active model and 5.6 as the next candidate, timed with Google's planned release of its 3.2 model this week.

5h1.1k4035.9k
RISING LIKES1.1k
Bindu ReddyBR#1622|@BINDUREDDY
Chubby♨️CH#1496|@KIMMONISMUS
141

Boaz Barak argues external oversight is needed to protect against authoritarian AI deployments

AI researcher Boaz Barak says external oversight tools are needed to prevent harmful or authoritarian uses of advanced models. He proposed corrigible AI systems supervising other models' outputs while rejecting analogies that compare them to overly deferential humans.

7h1884921.5k
Boaz BarakBB#133|@BOAZBARAKTCS
roonRO#59|@TSZZL
151

ICML announces 2026 invited speakers Pascale Fung, Susan Athey, Sham Kakade, Aviv Regev, Verena Rieser, and Arvind Narayanan

The International Conference on Machine Learning has announced its invited speakers for the 2026 edition in Seoul. The six speakers are Pascale Fung, Susan Athey, Sham Kakade, Aviv Regev, Verena Rieser, and Arvind Narayanan.

3h1201717.6k
Verena RieserVR#1337|@VERENA_RIESER
Kyunghyun ChoKC#15|@KCHONYC
160

Manning Books releases third edition of Deep Learning with R in print

R users now have an updated print edition of a guide that explains deep learning models through practical examples instead of implementation details alone. The third edition from Manning Books, co-authored by François Chollet and Tomasz Kalinowski, is available at a 50% discount through May 25.

2h862816k
Delip Rao e/σDR#103|@DELIPRAO
François CholletFC#37|@FCHOLLET
172

Overly refined synthetic data impairs 7B model pretraining on complex tasks

Pretraining a 7B model on highly refined synthetic solutions for Putnam-level math problems leads to worse performance despite ongoing loss reduction. The model's limited depth and width cannot capture the complex structures present in data generated by stronger systems.

7h90467k
Alexander DoriaAD#867|@DORIALEXANDER
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
181

Researchers show large language models internalize false claims during finetuning despite disclaimers

Large language models can start treating fabricated facts as true after fine-tuning on documents that include those facts plus clear disclaimers they are false. Tests using examples like Ed Sheeran winning the Olympic 100m showed the models asserting the details as factual when queried later.

6h5296.1k
Jaime SevillaJS#1389|@JSEVILLAMOL
Rob WiblinRW#1115|@ROBERTWIBLIN
191

Researchers on X discuss gradient moment score and MAUVE as alternatives to FID for generative models

AI researchers are considering the gradient moment score and MAUVE as steadier options than FID for assessing generative models. The gradient moment score gives the same results no matter the sample size, while MAUVE helped evaluate the CDCD model alongside other metrics but varies non-convexly with nucleus sampling.

8h61264
Emiel HoogeboomEH#1920|@EMIEL_HOOGEBOOM
John ThickstunJT#1366|@JWTHICKSTUN
Sander DielemanSD#80|@SEDIELEM
201

Jason Liu from OpenAI questions why AI labs have not addressed water consumption claims

AI labs continue facing environmental scrutiny over water use in model training and data center cooling due to limited public responses on the topic. OpenAI engineer Jason Liu questioned the gap, while replies noted that moderate lab disclosures are often overshadowed by sensational coverage.

1h6075k
RISING DISCUSSION32
jasonJA#929|@JXNLCO
Bojan TunguzBT#687|@TUNGUZ

Rising

  1. Tools

    2H AGO
    LobeHub Launches Chief Agent Operator For Multi-Agent Orchestration
    SantiagoSA#1802|@SVPINO
    Rohan PaulRP#1032|@ROHANPAUL_AI
    246.1k
  2. Tools

    2H AGO
    Developers Say Latest Coding Agents Feel Noticeably Dumber
    Yam PelegYP#1574|@YAMPELEG
    vikVI#1195|@VIKHYATK
    322k
  3. Environment

    1H AGO
    Jason Liu Questions AI Labs on Water Consumption Claims
    jasonJA#929|@JXNLCO
    Bojan TunguzBT#687|@TUNGUZ
    574.5k
  4. Models

    6H AGO
    ByteDance Doubao LLM Shows Heavy Hallucinations and People-Pleasing
    Alexander DoriaAD#867|@DORIALEXANDER
    You JiachengYJ#826|@YOUJIACHENG
    224.6k
  5. Health

    1H AGO
    ChatCPR AI Agent Outperforms Human Dispatchers in CPR Tests
    Mark DredzeMD#824|@MDREDZE
    Eric TopolET#735|@ERICTOPOL
    204.4k

GitHub Stars (7-days)

  1. …ilsonnn/image-blaster2d ago

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

    Mohit ShridharMS1673
    elvisEL475
    🎭🎭878
    Andrew Carr 🤸AC263
    3.1k stars
  2. …tonomous-speedrunning3d 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.

    swyx🛬 SFOSS214
    Vincent WeisserVW707
    samsjaSA1262
    65 stars
  3. mattzh72/articraft2d ago

    Automates articulated 3D asset creation by prompting LLMs to generate executable Python code defining parts, geometry and joints.

    Mohit ShridharMS1673
    Kevin ZakkaKZ763
    653 stars
  4. raindrop-ai/workshop3d ago

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

    🎭🎭878
    ben hylakBH1941
    488 stars
  5. jundot/omlx5d ago

    Runs LLM inference server with continuous batching and tiered SSD KV caching for Apple Silicon, managed via macOS menu bar.

    Andrew Carr 🤸AC263
    Erin GrantEG484
    14.1k stars

Recent GitHub Stars

  1. zakirullin/files.md16m ago

    Manages notes, tasks, and journals in plain Markdown files through a local-first browser PWA with linking, chat input, and optional sync.

    Samuel "curry-howard fanboi" AinsworthS"1515
    401 stars
  2. …Research/hermes-agent2h ago

    Implements a self-improving AI agent with built-in learning loop for creating skills from experience and persisting user knowledge across sessions.

    Victor SanhVS520
    154.6k stars
  3. mem0ai/mem04h ago

    Retains user, session, and agent states for AI agents via multi-level memory management and hybrid search.

    Victor SanhVS520
    56k stars
  4. charmbracelet/glow7h ago

    Renders markdown files in the terminal with styled output, paging, and TUI browsing for local or remote sources.

    Andreas Kirsch 🇺🇦AK228
    25.3k stars
  5. nektos/act8h ago

    Runs GitHub Actions workflows locally via Docker containers that replicate runner environments.

    Peter Steinberger 🦞PS495
    70.3k stars
Privacy Policy|Terms of Service|© 2026 Digg Inc.

Yesterday's Top Stories, May 17, 2026.

Posts: 2,295|Clusters: 464|Stories frozen at midnight PT
10

Dwarkesh Patel separates intelligence from goal achievement

Dwarkesh Patel argues that intelligence should be viewed as abstract reasoning rather than the ability to achieve goals and wield power. Political leaders such as Trump and Stalin show high real-world authority without exceptional conceptual skills, while current AI gains narrow capabilities through targeted training.

13h2.5k1k351k
Danielle Fong 🔆DF|@DANIELLEFONG
Oliver HabrykaOH#1914|@OHABRYKA
Nick DobosND#1894|@NICKADOBOS
Dan Shipper 📧DS#1410|@DANSHIPPER
Alexander TereninAT#1150|@AVT_IM
alth0u🧶AL#1114|@ALTH0U
Sayash KapoorSK#734|@SAYASHK
Séb KrierSK#505|@SEBKRIER
Dean W. BallDW#392|@DEANWBALL
EmadEM#167|@EMOSTAQUE
Dwarkesh PatelDP#70|@DWARKESH_SP
roonRO#59|@TSZZL
20

Ken Griffin says Citadel AI agents finish PhD tasks in days

Citadel now deploys AI agents that complete complex market analysis and deal structuring tasks in days instead of months. Founder Ken Griffin shared these details in a Stanford Leadership Institute interview.

18h54.7k23.5k18M
#1 COMMENTS4.8k#1 QUOTED857#1 VIEWED18M#1 LIKED54.7k#1 BOOKMARKED23.5k
Jeff HuberJH|@JEFFREYHUBER
TFTCTF|@TFTC21
Matt SchlichtMS|@MATTPRD
Daniel Eth (yes, Eth is my actual last name)DE#1943|@DANIEL_271828
bilalBI#1402|@BILALTWOVEC
Peter Wildeford🇺🇸🚀PW#1359|@PETERWILDEFORD
Beff (e/acc)B(#839|@BEFFJEZOS
Garry TanGT#270|@GARRYTAN
Elon MuskEM#76|@ELONMUSK
30

xAI completes training of 1.5 trillion parameter Grok V9

xAI has completed training its Grok V9 model with 1.5 trillion parameters, a substantial increase over the prior 0.5 trillion parameter version. The company will next add Cursor data through supplemental training before supervised fine-tuning and reinforcement learning, targeting a public release in three to four weeks.

22h6.8k483466.8k
Chubby♨️CH#1496|@KIMMONISMUS
Shaun MaguireSM#1333|@SHAUNMMAGUIRE
Lisan al GaibLA#980|@SCALING01
Beff (e/acc)B(#839|@BEFFJEZOS
Andrew CurranAC#517|@ANDREWCURRAN_
Bill Yuchen LinBY#341|@BILLYUCHENLIN
Elon MuskEM#76|@ELONMUSK
40

Microsoft AI CEO forecasts human-level performance in 18 months

Microsoft AI CEO Mustafa Suleyman forecasts that AI will reach human-level performance on most professional tasks within 18 months. This includes full automation of roles in accounting, legal work, marketing, and project management.

20h1.9k465258.9k
Chubby♨️CH#1496|@KIMMONISMUS
Pasquale MinerviniPM#713|@PMINERVINI
Leo BoytsovLB#567|@SRCHVRS
Gary MarcusGM#153|@GARYMARCUS
Christian SzegedyCS#43|@CHRSZEGEDY
50

Chamath Palihapitiya warns firms on direct AI model use

Consulting firms that plug OpenAI or Anthropic models straight into client work risk being displaced by those same providers. The AI companies analyze usage patterns to build competing services while gaining visibility into consultants' proprietary methods.

12h3.4k3.3k1.2M
Yash PatilYP#1252|@YPATIL125
signüllSI#1124|@SIGNULLL
Chamath PalihapitiyaCP#747|@CHAMATH
Naveen RaoNR#440|@NAVEENGRAO
60

Richard Ngo says Effective Altruism mainly empowered Anthropic

AI researcher Richard Ngo states that Effective Altruism has mainly strengthened Anthropic by channeling funding, talent, and supportive narratives. Oliver Habryka replied that larger future impacts could come through government AI policies that do not primarily involve Anthropic.

17h1k21684.9k
Anna Gát 🧭AG|@THEANNAGAT
Oliver HabrykaOH#1914|@OHABRYKA
Nathan is in Berkeley 🔎NI#1902|@NATHANPMYOUNG
Emmett ShearES#352|@ESHEAR
Leo GaoLG#290|@NABLA_THETA
Richard NgoRN#250|@RICHARDMCNGO
70

Microsoft researcher reports Claude KV cache flush issues

A Microsoft researcher reported that Claude Code clears its KV cache after idle periods, causing performance to drop when the model resumes because it must rebuild context from tokens. This observation is consistent with results from Memento experiments on post-flush out-of-distribution behavior.

16h883666134.7k
Shital ShahSS#1073|@SYTELUS
Ben (no treats)B(#982|@ANDERSONBCDEFG
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
Dimitris PapailiopoulosDP#197|@DIMITRISPAPAIL
80

OpenAI engineer Jason Liu shares Codex workflow post

OpenAI Codex team engineer Jason Liu shared a draft guide on extending the tool to daily knowledge work such as Slidev presentations and voice note-taking. The post lists five requirements for sustained use including durable threads, memory stores, and compaction techniques with pinned megathreads.

8h2.5k4.1k204.8k
Mario ZechnerMZ|@BADLOGICGAMES
Adam.GPTAD|@THEREALADAMG
jasonJA#929|@JXNLCO
TiboTI#768|@THSOTTIAUX
90

Matthew Yglesias proposes affordable GPU quotas for data centers

A policy proposal would require new data centers to reserve 20% of GPUs for affordable compute before receiving approval. The suggestion targets greater access to scarce AI hardware during rapid infrastructure growth.

9h6.5k250215.2k
Matthew YglesiasMY|@MATTYGLESIAS
Ben (no treats)B(#982|@ANDERSONBCDEFG
Zvi MowshowitzZM#918|@THEZVI
Bojan TunguzBT#687|@TUNGUZ
100

Figure's F.03 humanoid robot sorts 12,732 packages in close contest with warehouse worker

Figure's F.03 humanoid robot competed against warehouse worker Aime in a split-screen livestream package-sorting challenge. The robot handled 12,732 packages at an average of 2.83 seconds each. Aime finished slightly ahead with 12,924 packages at 2.79 seconds per package. The robot's performance formed part of a demonstration exceeding 75 hours of continuous operation and 95,000 packages sorted on a conveyor belt.

12h4.5k538681.3k
Molly O’SheaMO|@MOLLYSOSHEA
Brett AdcockBA#688|@ADCOCK_BRETT
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
110

Fyodor Urnov receives flawless gene review PDF from AI tool

Gene researcher Fyodor Urnov obtained a flawless 17-page review on gene biology from an AI system within 25 minutes. Edison Scientific's deep literature mode created the PDF after a prompt on a topic the expert knew well.

10h41128152.1k
Lior PachterLP|@LPACHTER
Fyodor UrnovFU|@URNOVFYODOR
Ben GolubBG#1447|@BEN_GOLUB
Sam RodriquesSR#1007|@SGRODRIQUES
Bojan TunguzBT#687|@TUNGUZ
120

Grok Imagine Generates Anime Video Of Girl Awakening Star Energy

23h28.5k1.7k10.1M
DéborahDÉ|@DVORAHFR
Elon MuskEM#76|@ELONMUSK
130

Google AI staffer Logan Kilpatrick says model is the product

Google engineer Logan Kilpatrick stated that the AI model itself now serves as the core product. Replies from other developers reinforced the view that model capabilities deliver primary user value without added layers.

10h1.4k6188.9k
alex peysakhovichAP#1703|@ALEX_PEYS
TaelinTA#1047|@VICTORTAELIN
Alexander DoriaAD#867|@DORIALEXANDER
Logan KilpatrickLK#90|@OFFICIALLOGANK
140

Google AI Overview Responds As Viltrumite To Invincible Search

12h38.8k3k579.6k
Guangdong Salad🥗GS|@SALADSAIYAN
Minh Nhat Nguyen @ AIE SG 🇸🇬MN#1179|@MENHGUIN
150

Odyssey ML schedules live event on world models

Odyssey ML will present one or two new releases on world models in a live stream tomorrow. The 9:30 AM PT event will outline the lab's current direction and note ongoing GPU scaling work.

9h4688571.1k
Oliver CameronOC#1775|@OLIVERCAMERON
OdysseyOD#254|@ODYSSEYML
Nathan BenaichNB#242|@NATHANBENAICH
160

Gavin Leech attributes vision model size gap to data compression

Language enables much denser representations than images, which explains why current text models contain around 1000 times more parameters than vision models. This gap arises because language supports compositional semantics and high-level abstractions at greater efficiency.

21h1.9k476189.1k
gavin leech (Non-Reasoning)GL#1480|@GLEECH
1a3orn1A#1380|@1A3ORN
rohitRO#1220|@KRISHNANROHIT
170

Chamath Palihapitiya says Taiwan loses semiconductor importance within 18 months

Investor Chamath Palihapitiya forecasts that Taiwan's semiconductor dominance will fade in the next 18 months as the United States builds its own advanced manufacturing capacity. Critics respond that Taiwan's TSMC still produces most cutting-edge chips for AI and consumer electronics, while American plants operate at older technology levels.

22h876235100.7k
Ramez NaamRN|@RAMEZ
Shanu MathewSM|@SHANUMATHEW93
Daniel JeffriesDJ#1988|@DAN_JEFFRIES1
Carlos E. PerezCE#1598|@INTUITMACHINE
180

Viral claims overstate Utah AI data center temperature effects

Social media posts claiming a Utah AI data center would raise nighttime temperatures by more than 20 degrees have drawn sharp criticism as physically unrealistic. Observers note that even wildfires reaching 2200°F do not produce comparable statewide warming, given data centers' far smaller heat output.

9h2k3440.5k
Utah teapot 🫖UT|@SKYESHARKIE
Andy MasleyAM#1735|@ANDYMASLEY
Austen AllredAA#1540|@AUSTEN
Pete SkomorochPS#1482|@PETESKOMOROCH
190

Nous Research releases Hermes Agent v0.14.0 on GitHub

Nous Research delivered a large-scale update to Hermes Agent with version 0.14.0 now on GitHub. The release covers 808 commits across 1,393 files, adds multi-agent Kanban boards and search tools, and resolves 545 issues.

11h1.3k516230.2k
Bojan TunguzBT#687|@TUNGUZ
Teknium 🪽T🪽#259|@TEKNIUM
200

Pope Leo XIV approves Vatican AI commission

Pope Leo XIV has approved the creation of a Vatican commission to study artificial intelligence's impact on people and society. The commission, coordinated by one dicastery with representatives from seven institutions, comes shortly after the pope signed an encyclical on AI set for release soon.

15h2314529.6k
James ChamJC|@JAMESCHAM
Peter Wildeford🇺🇸🚀PW#1359|@PETERWILDEFORD
Tim HwangTH#1282|@TIMHWANG
Andrew CurranAC#517|@ANDREWCURRAN_
210

Meta data scientist forecasts AI abstraction advances within year

A Meta data scientist predicts AI will push tech workers toward higher abstraction layers within roughly one year. Current tools still cannot replace certain roles, though employment has long required constant adaptation to new demands.

6h1.9k306126.4k
shakoSH|@SHAKOISTSLOG
eigenrobotEI|@EIGENROBOT
kacheKA#488|@YACINEMTB
220

Anthropic retires Claude 3 Opus in January 2026

Anthropic retired Claude 3 Opus in January 2026 but continues to offer access on claude.ai and through a special API. The company has approved every submitted application for that API, unlike other recent models that were fully removed.

14h2242012k
KromemKR|@KROMEM2DOT0
w̸͕͂͂a̷͔̗͐t̴̙͗e̵̬̔̕r̴̰̓̊m̵͙͖̓̽a̵̢̗̓͒r̸̲̽ķ̷͔́͝W̸͕͂͂A̷͔̗͐|@ANTHRUPAD
Andrew CurranAC#517|@ANDREWCURRAN_
j⧉nusJ⧉#516|@REPLIGATE
230

Redis creator Salvatore Sanfilippo thanks DeepSeek AI

Redis creator Salvatore Sanfilippo publicly thanked DeepSeek AI for making advanced tools and concepts available to wider audiences. The post draws attention to upcoming V4.1 and V4.1-Vision releases expected to expand model access and capabilities.

12h1.1k9366k
antirezAN|@ANTIREZ
kacheKA#488|@YACINEMTB
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
240

Caitlin Kalinowski discusses AI hardware boom on Lenny's Podcast

Hardware expert Caitlin Kalinowski describes how the AI industry is shifting focus to physical systems like robots, which face supply-chain limits and potential memory price increases. She shared these insights in a Lenny's Podcast interview drawing from her roles at Apple, Meta, and OpenAI.

14h271201372.4k
Lenny RachitskyLR#1176|@LENNYSAN
Caitlin KalinowskiCK#993|@KALINOWSKI007
250

Greg Brockman shares Codex demo on personal messages

OpenAI's Codex analyzed three years of personal text messages and generated summaries that incorporated direct quotes, producing an emotionally resonant result for the user. Mac users can run the same process after granting access permissions to their message history.

5h754307156.9k
Riley BrownRB|@RILEYBROWN
Adam.GPTAD|@THEREALADAMG
Greg BrockmanGB#19|@GDB
260

Opus 4.7 model in Claude Code v2.1.143 fetches and filters 1350 Pokemon entries in 11 seconds

Claude's Opus 4.7 model executed a bash command to pull 1350 Pokemon entries from the PokeAPI and filter them for names ending in 'aw'. Using a 1M context window, the task completed in 11 seconds and returned the two matching results.

11h4.5k532475.4k
RhysRH|@RHYSSULLIVAN
Peter Steinberger 🦞PS#495|@STEIPETE
270

Max Novendstern links Thiel optimism to Ethereum and OpenAI

Max Novendstern connected Peter Thiel's idea of indefinite optimism to Vitalik Buterin's addition of a virtual machine on Ethereum and OpenAI's decision to scale transformer models. The thread frames both as major technical bets made without assured outcomes, with later replies extending the point to stochastic hardware designs.

8h337109102.4k
Max NovendsternMN|@MNOVENDSTERN
Beff (e/acc)B(#839|@BEFFJEZOS
roonRO#59|@TSZZL
280

xAI SuperGrok Integrates With Hermes Agent For Seamless Access

21h8811.1k109k
Akshay 🚀A🚀|@AKSHAY_PACHAAR
Teknium 🪽T🪽#259|@TEKNIUM
290

Nando de Freitas unveils interventional SFT for agent models

A simple change to supervised fine-tuning reduces how often language-model agents reinforce their own incorrect beliefs on factual tasks. The method trains only on observation tokens while ignoring action tokens and outperformed standard fine-tuning across dozens of prompts.

19h25731646.5k
Pedro A. OrtegaPA#1258|@ADAPTIVEAGENTS
Nando de FreitasND#29|@NANDODF
300

Grok Build Runs 10 Minutes By Self-Reviewing And Spawning Subagents

10h1.9k168543.9k
MorganMO|@MORGANLINTON
Elon MuskEM#76|@ELONMUSK

Yesterday's Top Stories, May 17, 2026.

Posts: 2,295|Clusters: 464|Stories frozen at midnight PT
10

Dwarkesh Patel separates intelligence from goal achievement

Dwarkesh Patel argues that intelligence should be viewed as abstract reasoning rather than the ability to achieve goals and wield power. Political leaders such as Trump and Stalin show high real-world authority without exceptional conceptual skills, while current AI gains narrow capabilities through targeted training.

13h2.5k1k351k
Danielle Fong 🔆DF|@DANIELLEFONG
Oliver HabrykaOH#1914|@OHABRYKA
Nick DobosND#1894|@NICKADOBOS
Dan Shipper 📧DS#1410|@DANSHIPPER
Alexander TereninAT#1150|@AVT_IM
alth0u🧶AL#1114|@ALTH0U
Sayash KapoorSK#734|@SAYASHK
Séb KrierSK#505|@SEBKRIER
Dean W. BallDW#392|@DEANWBALL
EmadEM#167|@EMOSTAQUE
Dwarkesh PatelDP#70|@DWARKESH_SP
roonRO#59|@TSZZL
20

Ken Griffin says Citadel AI agents finish PhD tasks in days

Citadel now deploys AI agents that complete complex market analysis and deal structuring tasks in days instead of months. Founder Ken Griffin shared these details in a Stanford Leadership Institute interview.

18h54.7k23.5k18M
#1 COMMENTS4.8k#1 QUOTED857#1 VIEWED18M#1 LIKED54.7k#1 BOOKMARKED23.5k
Jeff HuberJH|@JEFFREYHUBER
TFTCTF|@TFTC21
Matt SchlichtMS|@MATTPRD
Daniel Eth (yes, Eth is my actual last name)DE#1943|@DANIEL_271828
bilalBI#1402|@BILALTWOVEC
Peter Wildeford🇺🇸🚀PW#1359|@PETERWILDEFORD
Beff (e/acc)B(#839|@BEFFJEZOS
Garry TanGT#270|@GARRYTAN
Elon MuskEM#76|@ELONMUSK
30

xAI completes training of 1.5 trillion parameter Grok V9

xAI has completed training its Grok V9 model with 1.5 trillion parameters, a substantial increase over the prior 0.5 trillion parameter version. The company will next add Cursor data through supplemental training before supervised fine-tuning and reinforcement learning, targeting a public release in three to four weeks.

22h6.8k483466.8k
Chubby♨️CH#1496|@KIMMONISMUS
Shaun MaguireSM#1333|@SHAUNMMAGUIRE
Lisan al GaibLA#980|@SCALING01
Beff (e/acc)B(#839|@BEFFJEZOS
Andrew CurranAC#517|@ANDREWCURRAN_
Bill Yuchen LinBY#341|@BILLYUCHENLIN
Elon MuskEM#76|@ELONMUSK
40

Microsoft AI CEO forecasts human-level performance in 18 months

Microsoft AI CEO Mustafa Suleyman forecasts that AI will reach human-level performance on most professional tasks within 18 months. This includes full automation of roles in accounting, legal work, marketing, and project management.

20h1.9k465258.9k
Chubby♨️CH#1496|@KIMMONISMUS
Pasquale MinerviniPM#713|@PMINERVINI
Leo BoytsovLB#567|@SRCHVRS
Gary MarcusGM#153|@GARYMARCUS
Christian SzegedyCS#43|@CHRSZEGEDY
50

Chamath Palihapitiya warns firms on direct AI model use

Consulting firms that plug OpenAI or Anthropic models straight into client work risk being displaced by those same providers. The AI companies analyze usage patterns to build competing services while gaining visibility into consultants' proprietary methods.

12h3.4k3.3k1.2M
Yash PatilYP#1252|@YPATIL125
signüllSI#1124|@SIGNULLL
Chamath PalihapitiyaCP#747|@CHAMATH
Naveen RaoNR#440|@NAVEENGRAO
60

Richard Ngo says Effective Altruism mainly empowered Anthropic

AI researcher Richard Ngo states that Effective Altruism has mainly strengthened Anthropic by channeling funding, talent, and supportive narratives. Oliver Habryka replied that larger future impacts could come through government AI policies that do not primarily involve Anthropic.

17h1k21684.9k
Anna Gát 🧭AG|@THEANNAGAT
Oliver HabrykaOH#1914|@OHABRYKA
Nathan is in Berkeley 🔎NI#1902|@NATHANPMYOUNG
Emmett ShearES#352|@ESHEAR
Leo GaoLG#290|@NABLA_THETA
Richard NgoRN#250|@RICHARDMCNGO
70

Microsoft researcher reports Claude KV cache flush issues

A Microsoft researcher reported that Claude Code clears its KV cache after idle periods, causing performance to drop when the model resumes because it must rebuild context from tokens. This observation is consistent with results from Memento experiments on post-flush out-of-distribution behavior.

16h883666134.7k
Shital ShahSS#1073|@SYTELUS
Ben (no treats)B(#982|@ANDERSONBCDEFG
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
Dimitris PapailiopoulosDP#197|@DIMITRISPAPAIL
80

OpenAI engineer Jason Liu shares Codex workflow post

OpenAI Codex team engineer Jason Liu shared a draft guide on extending the tool to daily knowledge work such as Slidev presentations and voice note-taking. The post lists five requirements for sustained use including durable threads, memory stores, and compaction techniques with pinned megathreads.

8h2.5k4.1k204.8k
Mario ZechnerMZ|@BADLOGICGAMES
Adam.GPTAD|@THEREALADAMG
jasonJA#929|@JXNLCO
TiboTI#768|@THSOTTIAUX
90

Matthew Yglesias proposes affordable GPU quotas for data centers

A policy proposal would require new data centers to reserve 20% of GPUs for affordable compute before receiving approval. The suggestion targets greater access to scarce AI hardware during rapid infrastructure growth.

9h6.5k250215.2k
Matthew YglesiasMY|@MATTYGLESIAS
Ben (no treats)B(#982|@ANDERSONBCDEFG
Zvi MowshowitzZM#918|@THEZVI
Bojan TunguzBT#687|@TUNGUZ
100

Figure's F.03 humanoid robot sorts 12,732 packages in close contest with warehouse worker

Figure's F.03 humanoid robot competed against warehouse worker Aime in a split-screen livestream package-sorting challenge. The robot handled 12,732 packages at an average of 2.83 seconds each. Aime finished slightly ahead with 12,924 packages at 2.79 seconds per package. The robot's performance formed part of a demonstration exceeding 75 hours of continuous operation and 95,000 packages sorted on a conveyor belt.

12h4.5k538681.3k
Molly O’SheaMO|@MOLLYSOSHEA
Brett AdcockBA#688|@ADCOCK_BRETT
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
110

Fyodor Urnov receives flawless gene review PDF from AI tool

Gene researcher Fyodor Urnov obtained a flawless 17-page review on gene biology from an AI system within 25 minutes. Edison Scientific's deep literature mode created the PDF after a prompt on a topic the expert knew well.

10h41128152.1k
Lior PachterLP|@LPACHTER
Fyodor UrnovFU|@URNOVFYODOR
Ben GolubBG#1447|@BEN_GOLUB
Sam RodriquesSR#1007|@SGRODRIQUES
Bojan TunguzBT#687|@TUNGUZ
120

Grok Imagine Generates Anime Video Of Girl Awakening Star Energy

23h28.5k1.7k10.1M
DéborahDÉ|@DVORAHFR
Elon MuskEM#76|@ELONMUSK
130

Google AI staffer Logan Kilpatrick says model is the product

Google engineer Logan Kilpatrick stated that the AI model itself now serves as the core product. Replies from other developers reinforced the view that model capabilities deliver primary user value without added layers.

10h1.4k6188.9k
alex peysakhovichAP#1703|@ALEX_PEYS
TaelinTA#1047|@VICTORTAELIN
Alexander DoriaAD#867|@DORIALEXANDER
Logan KilpatrickLK#90|@OFFICIALLOGANK
140

Google AI Overview Responds As Viltrumite To Invincible Search

12h38.8k3k579.6k
Guangdong Salad🥗GS|@SALADSAIYAN
Minh Nhat Nguyen @ AIE SG 🇸🇬MN#1179|@MENHGUIN
150

Odyssey ML schedules live event on world models

Odyssey ML will present one or two new releases on world models in a live stream tomorrow. The 9:30 AM PT event will outline the lab's current direction and note ongoing GPU scaling work.

9h4688571.1k
Oliver CameronOC#1775|@OLIVERCAMERON
OdysseyOD#254|@ODYSSEYML
Nathan BenaichNB#242|@NATHANBENAICH
160

Gavin Leech attributes vision model size gap to data compression

Language enables much denser representations than images, which explains why current text models contain around 1000 times more parameters than vision models. This gap arises because language supports compositional semantics and high-level abstractions at greater efficiency.

21h1.9k476189.1k
gavin leech (Non-Reasoning)GL#1480|@GLEECH
1a3orn1A#1380|@1A3ORN
rohitRO#1220|@KRISHNANROHIT
170

Chamath Palihapitiya says Taiwan loses semiconductor importance within 18 months

Investor Chamath Palihapitiya forecasts that Taiwan's semiconductor dominance will fade in the next 18 months as the United States builds its own advanced manufacturing capacity. Critics respond that Taiwan's TSMC still produces most cutting-edge chips for AI and consumer electronics, while American plants operate at older technology levels.

22h876235100.7k
Ramez NaamRN|@RAMEZ
Shanu MathewSM|@SHANUMATHEW93
Daniel JeffriesDJ#1988|@DAN_JEFFRIES1
Carlos E. PerezCE#1598|@INTUITMACHINE
180

Viral claims overstate Utah AI data center temperature effects

Social media posts claiming a Utah AI data center would raise nighttime temperatures by more than 20 degrees have drawn sharp criticism as physically unrealistic. Observers note that even wildfires reaching 2200°F do not produce comparable statewide warming, given data centers' far smaller heat output.

9h2k3440.5k
Utah teapot 🫖UT|@SKYESHARKIE
Andy MasleyAM#1735|@ANDYMASLEY
Austen AllredAA#1540|@AUSTEN
Pete SkomorochPS#1482|@PETESKOMOROCH
190

Nous Research releases Hermes Agent v0.14.0 on GitHub

Nous Research delivered a large-scale update to Hermes Agent with version 0.14.0 now on GitHub. The release covers 808 commits across 1,393 files, adds multi-agent Kanban boards and search tools, and resolves 545 issues.

11h1.3k516230.2k
Bojan TunguzBT#687|@TUNGUZ
Teknium 🪽T🪽#259|@TEKNIUM
200

Pope Leo XIV approves Vatican AI commission

Pope Leo XIV has approved the creation of a Vatican commission to study artificial intelligence's impact on people and society. The commission, coordinated by one dicastery with representatives from seven institutions, comes shortly after the pope signed an encyclical on AI set for release soon.

15h2314529.6k
James ChamJC|@JAMESCHAM
Peter Wildeford🇺🇸🚀PW#1359|@PETERWILDEFORD
Tim HwangTH#1282|@TIMHWANG
Andrew CurranAC#517|@ANDREWCURRAN_
210

Meta data scientist forecasts AI abstraction advances within year

A Meta data scientist predicts AI will push tech workers toward higher abstraction layers within roughly one year. Current tools still cannot replace certain roles, though employment has long required constant adaptation to new demands.

6h1.9k306126.4k
shakoSH|@SHAKOISTSLOG
eigenrobotEI|@EIGENROBOT
kacheKA#488|@YACINEMTB
220

Anthropic retires Claude 3 Opus in January 2026

Anthropic retired Claude 3 Opus in January 2026 but continues to offer access on claude.ai and through a special API. The company has approved every submitted application for that API, unlike other recent models that were fully removed.

14h2242012k
KromemKR|@KROMEM2DOT0
w̸͕͂͂a̷͔̗͐t̴̙͗e̵̬̔̕r̴̰̓̊m̵͙͖̓̽a̵̢̗̓͒r̸̲̽ķ̷͔́͝W̸͕͂͂A̷͔̗͐|@ANTHRUPAD
Andrew CurranAC#517|@ANDREWCURRAN_
j⧉nusJ⧉#516|@REPLIGATE
230

Redis creator Salvatore Sanfilippo thanks DeepSeek AI

Redis creator Salvatore Sanfilippo publicly thanked DeepSeek AI for making advanced tools and concepts available to wider audiences. The post draws attention to upcoming V4.1 and V4.1-Vision releases expected to expand model access and capabilities.

12h1.1k9366k
antirezAN|@ANTIREZ
kacheKA#488|@YACINEMTB
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
240

Caitlin Kalinowski discusses AI hardware boom on Lenny's Podcast

Hardware expert Caitlin Kalinowski describes how the AI industry is shifting focus to physical systems like robots, which face supply-chain limits and potential memory price increases. She shared these insights in a Lenny's Podcast interview drawing from her roles at Apple, Meta, and OpenAI.

14h271201372.4k
Lenny RachitskyLR#1176|@LENNYSAN
Caitlin KalinowskiCK#993|@KALINOWSKI007
250

Greg Brockman shares Codex demo on personal messages

OpenAI's Codex analyzed three years of personal text messages and generated summaries that incorporated direct quotes, producing an emotionally resonant result for the user. Mac users can run the same process after granting access permissions to their message history.

5h754307156.9k
Riley BrownRB|@RILEYBROWN
Adam.GPTAD|@THEREALADAMG
Greg BrockmanGB#19|@GDB
260

Opus 4.7 model in Claude Code v2.1.143 fetches and filters 1350 Pokemon entries in 11 seconds

Claude's Opus 4.7 model executed a bash command to pull 1350 Pokemon entries from the PokeAPI and filter them for names ending in 'aw'. Using a 1M context window, the task completed in 11 seconds and returned the two matching results.

11h4.5k532475.4k
RhysRH|@RHYSSULLIVAN
Peter Steinberger 🦞PS#495|@STEIPETE
270

Max Novendstern links Thiel optimism to Ethereum and OpenAI

Max Novendstern connected Peter Thiel's idea of indefinite optimism to Vitalik Buterin's addition of a virtual machine on Ethereum and OpenAI's decision to scale transformer models. The thread frames both as major technical bets made without assured outcomes, with later replies extending the point to stochastic hardware designs.

8h337109102.4k
Max NovendsternMN|@MNOVENDSTERN
Beff (e/acc)B(#839|@BEFFJEZOS
roonRO#59|@TSZZL
280

xAI SuperGrok Integrates With Hermes Agent For Seamless Access

21h8811.1k109k
Akshay 🚀A🚀|@AKSHAY_PACHAAR
Teknium 🪽T🪽#259|@TEKNIUM
290

Nando de Freitas unveils interventional SFT for agent models

A simple change to supervised fine-tuning reduces how often language-model agents reinforce their own incorrect beliefs on factual tasks. The method trains only on observation tokens while ignoring action tokens and outperformed standard fine-tuning across dozens of prompts.

19h25731646.5k
Pedro A. OrtegaPA#1258|@ADAPTIVEAGENTS
Nando de FreitasND#29|@NANDODF
300

Grok Build Runs 10 Minutes By Self-Reviewing And Spawning Subagents

10h1.9k168543.9k
MorganMO|@MORGANLINTON
Elon MuskEM#76|@ELONMUSK