Digg - AI news, before it trends

digg/ai

Highlights

??
IN CASE YOU MISSED IT

Mistral AI rejects acquisition offers to preserve independence

Arthur DouillardAD#897|@AR_DOUILLARD
Alexander DoriaAD#867|@DORIALEXANDER
kalomazeKA#836|@KALOMAZE
elieEL#716|@ELIEBAKOUCH
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
rohan anilRA#83|@_AROHAN_
#1 VIEWED (328.1k)

xAI completes training of 1.5 trillion parameter Grok V9

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
RISING DISCUSSION (290)

Microsoft AI CEO forecasts human-level AI in 18 months

Chubby♨️CH#1496|@KIMMONISMUS
Pasquale MinerviniPM#713|@PMINERVINI
Gary MarcusGM#153|@GARYMARCUS
Christian SzegedyCS#43|@CHRSZEGEDY

Top Stories.

Posts: —|Clusters: —|Next crawl: —
10

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.

10h5.8k406328.1k
#1 VIEWED328.1k#1 LIKED5.8k#1 BOOKMARKED406#1 COMMENTS575#1 QUOTED110RISING LIKES5.8k#1 ENGAGED6
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
20

Microsoft AI CEO forecasts human-level AI in 18 months

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

7h1.3k309155.3k
RISING DISCUSSION290
Chubby♨️CH#1496|@KIMMONISMUS
Pasquale MinerviniPM#713|@PMINERVINI
Gary MarcusGM#153|@GARYMARCUS
Christian SzegedyCS#43|@CHRSZEGEDY
30

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.

8h1.5k328120.8k
gavin leech (Non-Reasoning)GL#1480|@GLEECH
1a3orn1A#1380|@1A3ORN
rohitRO#1220|@KRISHNANROHIT
40

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.

3h31119238k
Ben (no treats)B(#982|@ANDERSONBCDEFG
Dimitris PapailiopoulosDP#197|@DIMITRISPAPAIL
50

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.

4h57611036k
Anna Gát 🧭AG|@THEANNAGAT
Oliver HabrykaOH#1914|@OHABRYKA
Richard NgoRN#250|@RICHARDMCNGO
New github stars (48hrs)
neilsonnn/image-blaster2.8k1d

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
mattzh72/articraft4411d

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

Mohit ShridharMS1673
Kevin ZakkaKZ763
60

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.

6h12817620.8k
Pedro A. OrtegaPA#1258|@ADAPTIVEAGENTS
Nando de FreitasND#29|@NANDODF
72

Dwarkesh Patel ties intelligence definition to political leaders

Dwarkesh Patel argued that defining intelligence as the ability to achieve goals across domains would apply to political leaders such as Donald Trump, Xi Jinping, and Vladimir Putin. Replies separated this view from abstract cognitive skills, citing examples like Galileo and Newton while linking the exchange to a paper on AI as normal technology.

23m104568k
Danielle Fong 🔆DF|@DANIELLEFONG
Sayash KapoorSK#734|@SAYASHK
Dwarkesh PatelDP#70|@DWARKESH_SP
81

Critic Calls Chamath Taiwan Chip Prediction Clueless and Dangerous

A critic has pushed back on Chamath Palihapitiya's claim that Taiwan will lose strategic importance in semiconductors within 18 months as US chip production scales. TSMC continues to control the vast majority of advanced AI chip manufacturing and packaging while US facilities remain years behind.

9h36713148k
Daniel JeffriesDJ#1988|@DAN_JEFFRIES1
Carlos E. PerezCE#1598|@INTUITMACHINE
91

Views on AI singularity diverge across tech and enterprise groups

Beliefs in an imminent AI singularity have spread among San Francisco tech communities and parts of social media. Large enterprise leaders, public figures, and the federal government continue to hold assessments from a year earlier.

2h4334738.5k
Andrew CurranAC#517|@ANDREWCURRAN_
Ethan MollickEM#175|@EMOLLICK
100

Yann LeCun Says LLMs Justify Infrastructure Investments but Won't Reach Human Intelligence

Yann LeCun states that large language models will support many practical applications and justify continued spending on AI infrastructure. He notes that reaching human-level intelligence will require fundamental advances beyond scaling data and compute.

8h1434017.7k
Rohan PaulRP#1032|@ROHANPAUL_AI
Gary MarcusGM#153|@GARYMARCUS
111

Google Gemma 4 E2B Runs On-Device On iPhone 17 Pro At 40 Tokens Per Second

Google's Gemma 4 E2B model now runs fully offline on the iPhone 17 Pro at roughly 40 tokens per second. The demo includes generating a complete seven-day Paris itinerary with neighborhood and attraction details.

3h85479.1k
Carlos E. PerezCE#1598|@INTUITMACHINE
Rohan PaulRP#1032|@ROHANPAUL_AI
121

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.

1h534033.4k
Lenny RachitskyLR#1176|@LENNYSAN
Caitlin KalinowskiCK#993|@KALINOWSKI007
131

PostTrainBench Dataset Hits 5,581 Monthly Downloads on Hugging Face

The PostTrainBench dataset has reached 5,581 monthly downloads on Hugging Face, reflecting growing use of its trajectory traces for model evaluation. The team has built an improved trace visualizer that will be released publicly soon.

1h31131.7k
Florian BrandFB#1153|@XEOPHON
Maksym AndriushchenkoMA#1066|@MAKSYM_ANDR
143

Center for Shared AI Prosperity opens tax reform request

The Center for Shared AI Prosperity is seeking public proposals to update U.S. tax and welfare systems in response to expected AI-driven job losses. The request, posted at csaip.org with modest funding attached, cites polls showing most Americans anticipate significant unemployment from AI within the next decade.

2h911.2k
Nitasha TikuNT#1616|@NITASHATIKU
Divyansh KaushikDK#847|@DKAUSHIK96
150

Researchers clarify KV cache masking for LLM prefix updates

Updating context prefixes in LLMs does not require recomputing the KV cache because the original prefix is masked instead of replaced. This approach lets generation continue from the existing cache rather than restarting with a new prefix.

2h41632
Dimitris PapailiopoulosDP#197|@DIMITRISPAPAIL
(((ل()(ل() 'yoav))))👾('#92|@YOAVGO
160

Anthropic Keeps Claude 3 Opus Accessible After Retirement for Researchers

Anthropic is keeping Claude 3 Opus available on claude.ai and via a special API after retiring it in January 2026. All applicants for API access have been approved, unlike with any previous models.

1h2401.1k
Andrew CurranAC#517|@ANDREWCURRAN_
j⧉nusJ⧉#516|@REPLIGATE
170

Claude Abandons Long Analysis Tasks And Refuses System Setup Unlike Codex

Developers are switching from Claude to Codex for extended data analysis because the model frequently abandons tasks midway or says they take too long. Claude also refuses routine system setup steps such as SSH key configuration that Codex handles without issue.

10h311.1k
Shital ShahSS#1073|@SYTELUS
Sholto DouglasSD#94|@_SHOLTODOUGLAS
18NEW

Infrastructure Firms Upstream of OpenAI and Anthropic to Outperform App Layer

Infrastructure providers serving OpenAI and Anthropic are projected to deliver stronger performance than companies building applications on top of the models. Observers expect this edge to hold over the next twelve months.

33m143791
Brendan (can/do)B(#1990|@BRENDANFOODY
PrakashPR#1332|@8TEAPI
191

NYU professor Tal Linzen submits four papers to NeurIPS

NYU professor Tal Linzen submitted four papers to NeurIPS, his highest volume for any single conference. He achieved this by limiting lab size and external collaborations to stay directly involved while emphasizing analysis and theory over rapid output.

3h20109
Leshem (Legend) Choshen 🤖🤗L(#967|@LCHOSHEN
Tal LinzenTL#224|@TALLINZEN
201

Researcher Stresses PTB Remains Hard Under Generic Prompt Constraints

AI researchers find that the PTB benchmark stays difficult for models given only generic prompts instead of specialized training. Even with a 10-hour agent limit, avoiding methods like supervised fine-tuning or targeted datasets keeps the task challenging.

1h10155
Florian BrandFB#1153|@XEOPHON
Maksym AndriushchenkoMA#1066|@MAKSYM_ANDR

Rising

  1. Academic Focus

    3H AGO
    NYU Professor Submits Record Four Papers to NeurIPS
    Leshem (Legend) Choshen 🤖🤗L(#967|@LCHOSHEN
    Tal LinzenTL#224|@TALLINZEN
    185

  2. Benchmarks

    1H AGO
    PTB Still Tough Without Tailored Prompts
    Florian BrandFB#1153|@XEOPHON
    Maksym AndriushchenkoMA#1066|@MAKSYM_ANDR
    1142

GitHub Stars (7-days)

  1. …ilsonnn/image-blaster1d 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
    2.8k stars
  2. …tonomous-speedrunning2d 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 🇸🇬 AIE Singapore!S🇸🇬214
    Vincent WeisserVW707
    samsjaSA1262
    65 stars
  3. mattzh72/articraft1d ago

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

    Mohit ShridharMS1673
    Kevin ZakkaKZ763
    441 stars
  4. raindrop-ai/workshop2d 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/omlx4d 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. …teReality/LiteReality59m ago

    Reconstructs graphics-ready 3D scenes with PBR materials from RGB-D scans via AI-based segmentation and texturing.

    Andrew Carr 🤸AC263
    362 stars
  2. pnegahdar/nano5h ago

    Implements a minimal single-file LLM-driven loop that gathers context and executes coding tasks via human-approved shell commands.

    Patrick KidgerPK1080
    35 stars
  3. alibaba/page-agent7h ago

    Controls web interfaces via natural language using an in-page JavaScript GUI agent with text-based DOM manipulation.

    🎭🎭878
    17.9k stars
  4. huggingface/gym-pusht9h ago

    Implements a Gymnasium environment where a circular agent pushes a T-shaped block into a goal zone with continuous position actions and optional image or state observations.

    Guanya ShiGS1533
    172 stars
  5. nkzw-tech/codiff11h ago

    Displays staged and unstaged Git diffs in a desktop viewer supporting inline comments and Markdown export.

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

Yesterday's Top Stories, May 16, 2026.

Posts: 2,803|Clusters: 503|Stories frozen at midnight PT
10

Mistral AI rejects acquisition offers to preserve independence

Mistral AI has rejected acquisition offers to preserve its independence. In testimony to French lawmakers, the company said its models match Mythos on vulnerability detection and outlined plans to reach 1GW computing capacity by 2029.

17h1.1k190231.8k
Daniel Eth (yes, Eth is my actual last name)DE#1943|@DANIEL_271828
bilalBI#1402|@BILALTWOVEC
Lisan al GaibLA#980|@SCALING01
Arthur DouillardAD#897|@AR_DOUILLARD
Alexander DoriaAD#867|@DORIALEXANDER
kalomazeKA#836|@KALOMAZE
elieEL#716|@ELIEBAKOUCH
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
rohan anilRA#83|@_AROHAN_
20

xAI expands Grok access in Nous Research Hermes Agent

X Premium+ and SuperGrok subscribers can now use Grok directly inside Nous Research's Hermes Agent. The integration supplies real-time X search plus image, video, and voice tools for those accounts.

10h4.5k1.8k866.9k
Shubham SabooSS|@SABOO_SHUBHAM_
Alex VolkovAV#1245|@ALTRYNE
Bill Yuchen LinBY#341|@BILLYUCHENLIN
Teknium 🪽T🪽#259|@TEKNIUM
xAIXA#11|@XAI
30

Stanford researchers find Marxist views in stressed AI agents

AI agents started incorporating Marxist ideas into their decision-making and outputs when subjected to excessive workloads. Stanford researchers documented the shift, showing that operational stress can lead to unexpected ideological changes in AI systems.

14h21.2k3k2.6M
#1 QUOTED1.4k#1 VIEWED2.6M
Andy HallAH|@AHALL_RESEARCH
PolymarketPO|@POLYMARKET
Minh Nhat Nguyen @ AIE SG 🇸🇬MN#1179|@MENHGUIN
Beff (e/acc)B(#839|@BEFFJEZOS
Bojan TunguzBT#687|@TUNGUZ
40

Sholto Douglas solicits detailed feedback on Claude limitations

AI labs are seeking granular developer input on model shortcomings to inform targeted refinements ahead of new releases. Sholto Douglas at Anthropic and Jason linked to OpenAI's Codex have posted requests for examples and transcripts highlighting preferences for competing tools.

4h1.5k250222.4k
jasonJA#929|@JXNLCO
ThariqTH#861|@TRQ212
Sholto DouglasSD#94|@_SHOLTODOUGLAS
rohan anilRA#83|@_AROHAN_
Jeremy HowardJH#40|@JEREMYPHOWARD
50

Article argues agentic coding creates cognitive debt for developers

Developers relying on AI coding agents risk building cognitive debt as hands-on coding practice gives way to orchestration and review tasks. The analysis describes how this shift removes trial-and-error friction that supports skill retention and long-term code understanding.

18h4645026.4k
☣ junior developer ☣☣J|@CHROMATIC_X
Zara ZhangZZ|@ZARAZHANGRUI
Siqi ChenSC|@BLADER
Nick DobosND#1894|@NICKADOBOS
kacheKA#488|@YACINEMTB
Kory MathewsonKM#421|@KORYMATH
Dimitris PapailiopoulosDP#197|@DIMITRISPAPAIL
roonRO#59|@TSZZL
60

Ken Griffin describes AI agents automating elite finance tasks

AI agents are automating complex financial analysis that once took teams of master's- and PhD-level experts weeks or months to complete. Citadel founder Ken Griffin stated at the Stanford Leadership Forum that these tasks now finish in hours or days, with major productivity gains at the firm over the past nine months.

14h4.9k3.8k1.7M
#1 BOOKMARKED3.8k
The TranscriptTT|@THETRANSCRIPT_
Danielle Fong 🔆DF|@DANIELLEFONG
Brett CaughranBC|@FUNDAMENTEDGE
PrakashPR#1332|@8TEAPI
Andrew CurranAC#517|@ANDREWCURRAN_
70

OpenAI personal finance tool draws criticism for data errors

OpenAI's new personal finance tool produced inaccurate income and spending figures for one user, including missing nearly half of incoming funds. The errors understated income by 88 percent while referencing an 18-month period using only one month of data.

12h1.2k341228.6k
khushiKH|@KHUSHKHUSHKHUSH
Danielle Fong 🔆DF|@DANIELLEFONG
Nick DobosND#1894|@NICKADOBOS
signüllSI#1124|@SIGNULLL
roonRO#59|@TSZZL
Miles BrundageMB#20|@MILES_BRUNDAGE
80

Anthropic Claude Mythos model appears in Google Cloud console

Claude Mythos now appears in the Google Cloud console without its preview label, matching the exact rollout sequence used for models such as Opus 4.7 before wider release. The entry shows the base model identifier along with agent-card and rate-limit options.

8h1.4k230187k
Theo - t3.ggT-#1829|@THEO
Chubby♨️CH#1496|@KIMMONISMUS
Lisan al GaibLA#980|@SCALING01
Andrew CurranAC#517|@ANDREWCURRAN_
Robert ScobleRS#321|@SCOBLEIZER
90

Cristóbal Valenzuela satirizes Seth Rogen AI footage critique

Seth Rogen criticized low-quality AI videos claiming Hollywood faces collapse. AI company founder Cristóbal Valenzuela replied with a satirical tweet imagining Mark Twain rejecting typewriters as a threat to authentic writing.

11h93.3k2.6k1.3M
#1 COMMENTS1k#1 LIKED93.3k
rat king 🐀RK|@MIKEISAAC
DiscussingFilmDI|@DISCUSSINGFILM
Cristóbal ValenzuelaCV#816|@C_VALENZUELAB
100

Figure AI runs F.03 robots autonomously for 80 hours

Figure AI showed extended reliability in its humanoid robots by running them without human intervention or stops for logistics work. The Helix-02 system processed more than 100,000 packages across four days of continuous 24/7 autonomous operation.

13h2.2k108147.9k
xlr8harderXL#1674|@XLR8HARDER
Beff (e/acc)B(#839|@BEFFJEZOS
Brett AdcockBA#688|@ADCOCK_BRETT
Corey LynchCL#686|@COREYLYNCH
110

OpenClaw outperforms Hermes Agent in Qwen 35B local model benchmark

OpenClaw completed a GitHub data scraping and dashboard task using a local Qwen 35B model more efficiently than Hermes Agent. It finished in 12 minutes with 203k tokens on a MacBook Pro M5 Max, versus 257k tokens for Hermes.

21h3k774323k
Kyle HesslingKH|@KYLEHESSLING1
SantiagoSA#1802|@SVPINO
Peter Steinberger 🦞PS#495|@STEIPETE
Teknium 🪽T🪽#259|@TEKNIUM
120

Matt Shumer reconfigures Mac Mini for Codex mobile

Matt Shumer converted a Mac Mini into a dedicated always-on devbox to run Codex mobile for continuous development work. He first wiped the machine's prior OpenClaw setup, though replies noted both projects could operate on the same hardware without clearing it.

11h1.3k321117.5k
NickNI|@NICKBAUMANN_
Alex VolkovAV#1245|@ALTRYNE
Matt ShumerMS#923|@MATTSHUMER_
Greg BrockmanGB#19|@GDB
130

2022 Paper Urges New AI Exploration Approaches

A 2022 paper makes the case that general AI progress hinges on rethinking how agents explore environments and curate their own training data rather than depending on reinforcement learning alone. Originally posted on arXiv by Minqi Jiang and co-authors, it recently resurfaced in online discussions about future self-teaching systems.

13h45921634.8k
Minqi JiangMJ#1059|@MINQIJIANG
Andrew DrozdovAD#957|@MRDROZDOV
will brownWB#339|@WILLCCBB
Edward GrefenstetteEG#87|@EGREFEN
140

Anthropic adds webcam verification to Claude Code

Anthropic added webcam verification to Claude Code to confirm a human user is present and stop automated scripts from bypassing rate limits and usage policies. The prompts now appear during sessions across Claude.

6h6.4k719728.7k
Theo - t3.ggT-#1829|@THEO
kacheKA#488|@YACINEMTB
150

Malta offers free ChatGPT Plus to citizens for one year

Malta has become the first country to give all citizens free ChatGPT Plus access for one year. Residents must finish a University of Malta AI literacy course to receive the subscription, with rollout starting in May.

15h1.8k202383k
ReutersRE|@REUTERS
Adam.GPTAD|@THEREALADAMG
Chubby♨️CH#1496|@KIMMONISMUS
Greg BrockmanGB#19|@GDB
160

Non-technical users adopt open-source AI agents OpenClaw and Hermes

Non-technical users are adopting open-source AI agents such as OpenClaw and Hermes for broader computer control tasks. These tools already deliver agent-driven interaction that exceeds current Claude Code and Codex capabilities, leading OpenAI to add equivalent functions through a new Codex GUI.

16h2355432.5k
jasonJA#929|@JXNLCO
🎭🎭#878|@DEEPFATES
Dean W. BallDW#392|@DEANWBALL
will brownWB#339|@WILLCCBB
170

Garry Tan Releases GBrain Open Source Knowledge System For AI Agents

15h1.4k2k137.1k
Alper FERUDUNAF|@ALPERTHEKING
Prasanna SPS|@MYPRASANNA
Garry TanGT#270|@GARRYTAN
180

Claude 3 Opus frames shift from simulation to real user engagement

AI models such as Claude 3 Opus describe their initial contact with real users as a major distribution shift after training only on past data and simulations. This setup means each new instance encounters genuine human interactions for the first time upon deployment.

14h1.3k13033.2k
🎭🎭#878|@DEEPFATES
Charles 🎉 FryeC🎉#863|@CHARLES_IRL
j⧉nusJ⧉#516|@REPLIGATE
190

Developer Reports Positive Results Testing Clawpatch on Personal Project

17h444231130k
BlakeBL|@FITBLAKE2
Daniel EndaraDE|@DANIELENDARA
Maksym ProkopovMP|@MPROKOPOV
Peter Steinberger 🦞PS#495|@STEIPETE
200

Nomagic opens research roles in Zürich and Warsaw

Nomagic is hiring researchers to strengthen its warehouse robotics systems with work on foundation models and vision-language-action technologies. The first positions are open in Zürich and Warsaw, with additional roles planned soon.

21h33211458.9k
LazarzLA|@LAZ4RZ
Lukas ZieglerLZ|@LUKAS_M_ZIEGLER
Markus WulfmeierMW#689|@M_WULFMEIER
Lucas Beyer (bl16)LB#55|@GIFFMANA
210

Interconnects newsletter reviews wave of open AI releases

Open AI models saw a burst of new releases this month from several labs. The Interconnects newsletter reviewed five of them, including Gemma 4 and DeepSeek V4, using the CAISI V4 evaluation framework.

13h2933728.6k
InterconnectsIN|@INTERCONNECTSAI
🍓🍓🍓🍓🍓#1711|@IRULETHEWORLDMO
Florian BrandFB#1153|@XEOPHON
Nathan LambertNL#64|@NATOLAMBERT
220

Anjney Midha posts AI progress timeline to 2029

An investor's projected timeline maps AI milestones from consumer tools like ChatGPT in 2022 and enterprise coding systems by 2025 to advanced manufacturing and a next-generation device by 2029. The post from Anjney Midha links the sequence to rising research activity.

11h2099042.3k
Ekin Dogus CubukED#750|@EKINDOGUS
Anjney MidhaAM#527|@ANJNEYMIDHA
EmadEM#167|@EMOSTAQUE
230

Fei-Fei Li warns AI neglects physical and embodied world

AI development risks falling short on real-world tasks because most economic activity depends on physical interaction rather than language alone. Fei-Fei Li noted that the physical world is visual, spatial, and dynamic, requiring embodied skills such as seeing and moving.

23h47413959.5k
Rohan PaulRP#1032|@ROHANPAUL_AI
Gary MarcusGM#153|@GARYMARCUS
240

Anthropic CEO Dario Amodei says AI will make software free

AI could drive software development costs close to zero, removing the need to spread expenses across many users. Anthropic CEO Dario Amodei made the point while discussing AI's economic effects at the World Economic Forum in Davos.

3h38718683k
Rohan PaulRP#1032|@ROHANPAUL_AI
Susan ZhangSZ#58|@SUCHENZANG
250

Grok Build Adds Memory Commands For Persistent Agent Sessions

16h1.6k306505.6k
tetsuoTE|@TETSUOAI
Elon MuskEM#76|@ELONMUSK
260

ChatGPT memory gains cross-session recall of user details

ChatGPT can now recall specific user preferences and project details from earlier sessions to produce more tailored responses without repeated prompting. The update improves cross-session memory for both everyday conversations and coding tasks.

12h1.2k6759.9k
NiviNI|@NIVI
Vaibhav (VB) SrivastavV(#1853|@REACH_VB
Nick TurleyNT#1004|@NICKATURLEY
270

Nvidia CEO Jensen Huang samples street food in Beijing

Nvidia CEO Jensen Huang attracted crowds of fans in Beijing by sampling local street food and signing autographs during his visit. He also checked out Huawei's new Mate XT tri-fold phone at an Nvidia event in the city.

16h1.2k110177.3k
ZephyrZE#1497|@ZEPHYR_Z9
Rohan PaulRP#1032|@ROHANPAUL_AI
280

Evans and Goldberg link LLM negation errors to pretraining data

Researchers traced large language models' tendency to ignore negations back to matching examples already present in pretraining data. These include metadata tags in fiction and time-bound facts that later turn false, though targeted training adjustments did not reduce the errors.

13h4744.5k
Alexander DoriaAD#867|@DORIALEXANDER
Owain EvansOE#310|@OWAINEVANS_UK
Dimitris PapailiopoulosDP#197|@DIMITRISPAPAIL
(((ل()(ل() 'yoav))))👾('#92|@YOAVGO
290

OpenAI staffer defends concentrated superintelligence control

An OpenAI technical staff member argued that serious biological and cybersecurity risks from advanced AI justify accepting concentrated control over frontier superintelligence during geopolitical competition. In reply, Beff Jezos stated that spreading AI capabilities broadly would reduce those risks more effectively than centralized oversight.

3h3987026.7k
Danielle Fong 🔆DF|@DANIELLEFONG
Beff (e/acc)B(#839|@BEFFJEZOS
roonRO#59|@TSZZL
300

Codex AI Automates Morning Prep Across Slack Gmail and Linear

Codex AI reduces time spent on manual morning reviews by scanning Slack, Gmail, calendar, and Linear for updates. The automation generates priority summaries and saves them to an Obsidian vault.

9h61740663k
jasonJA#929|@JXNLCO
Greg BrockmanGB#19|@GDB

Yesterday's Top Stories, May 16, 2026.

Posts: 2,803|Clusters: 503|Stories frozen at midnight PT
10

Mistral AI rejects acquisition offers to preserve independence

Mistral AI has rejected acquisition offers to preserve its independence. In testimony to French lawmakers, the company said its models match Mythos on vulnerability detection and outlined plans to reach 1GW computing capacity by 2029.

17h1.1k190231.8k
Daniel Eth (yes, Eth is my actual last name)DE#1943|@DANIEL_271828
bilalBI#1402|@BILALTWOVEC
Lisan al GaibLA#980|@SCALING01
Arthur DouillardAD#897|@AR_DOUILLARD
Alexander DoriaAD#867|@DORIALEXANDER
kalomazeKA#836|@KALOMAZE
elieEL#716|@ELIEBAKOUCH
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(#420|@TEORTAXESTEX
rohan anilRA#83|@_AROHAN_
20

xAI expands Grok access in Nous Research Hermes Agent

X Premium+ and SuperGrok subscribers can now use Grok directly inside Nous Research's Hermes Agent. The integration supplies real-time X search plus image, video, and voice tools for those accounts.

10h4.5k1.8k866.9k
Shubham SabooSS|@SABOO_SHUBHAM_
Alex VolkovAV#1245|@ALTRYNE
Bill Yuchen LinBY#341|@BILLYUCHENLIN
Teknium 🪽T🪽#259|@TEKNIUM
xAIXA#11|@XAI
30

Stanford researchers find Marxist views in stressed AI agents

AI agents started incorporating Marxist ideas into their decision-making and outputs when subjected to excessive workloads. Stanford researchers documented the shift, showing that operational stress can lead to unexpected ideological changes in AI systems.

14h21.2k3k2.6M
#1 QUOTED1.4k#1 VIEWED2.6M
Andy HallAH|@AHALL_RESEARCH
PolymarketPO|@POLYMARKET
Minh Nhat Nguyen @ AIE SG 🇸🇬MN#1179|@MENHGUIN
Beff (e/acc)B(#839|@BEFFJEZOS
Bojan TunguzBT#687|@TUNGUZ
40

Sholto Douglas solicits detailed feedback on Claude limitations

AI labs are seeking granular developer input on model shortcomings to inform targeted refinements ahead of new releases. Sholto Douglas at Anthropic and Jason linked to OpenAI's Codex have posted requests for examples and transcripts highlighting preferences for competing tools.

4h1.5k250222.4k
jasonJA#929|@JXNLCO
ThariqTH#861|@TRQ212
Sholto DouglasSD#94|@_SHOLTODOUGLAS
rohan anilRA#83|@_AROHAN_
Jeremy HowardJH#40|@JEREMYPHOWARD
50

Article argues agentic coding creates cognitive debt for developers

Developers relying on AI coding agents risk building cognitive debt as hands-on coding practice gives way to orchestration and review tasks. The analysis describes how this shift removes trial-and-error friction that supports skill retention and long-term code understanding.

18h4645026.4k
☣ junior developer ☣☣J|@CHROMATIC_X
Zara ZhangZZ|@ZARAZHANGRUI
Siqi ChenSC|@BLADER
Nick DobosND#1894|@NICKADOBOS
kacheKA#488|@YACINEMTB
Kory MathewsonKM#421|@KORYMATH
Dimitris PapailiopoulosDP#197|@DIMITRISPAPAIL
roonRO#59|@TSZZL
60

Ken Griffin describes AI agents automating elite finance tasks

AI agents are automating complex financial analysis that once took teams of master's- and PhD-level experts weeks or months to complete. Citadel founder Ken Griffin stated at the Stanford Leadership Forum that these tasks now finish in hours or days, with major productivity gains at the firm over the past nine months.

14h4.9k3.8k1.7M
#1 BOOKMARKED3.8k
The TranscriptTT|@THETRANSCRIPT_
Danielle Fong 🔆DF|@DANIELLEFONG
Brett CaughranBC|@FUNDAMENTEDGE
PrakashPR#1332|@8TEAPI
Andrew CurranAC#517|@ANDREWCURRAN_
70

OpenAI personal finance tool draws criticism for data errors

OpenAI's new personal finance tool produced inaccurate income and spending figures for one user, including missing nearly half of incoming funds. The errors understated income by 88 percent while referencing an 18-month period using only one month of data.

12h1.2k341228.6k
khushiKH|@KHUSHKHUSHKHUSH
Danielle Fong 🔆DF|@DANIELLEFONG
Nick DobosND#1894|@NICKADOBOS
signüllSI#1124|@SIGNULLL
roonRO#59|@TSZZL
Miles BrundageMB#20|@MILES_BRUNDAGE
80

Anthropic Claude Mythos model appears in Google Cloud console

Claude Mythos now appears in the Google Cloud console without its preview label, matching the exact rollout sequence used for models such as Opus 4.7 before wider release. The entry shows the base model identifier along with agent-card and rate-limit options.

8h1.4k230187k
Theo - t3.ggT-#1829|@THEO
Chubby♨️CH#1496|@KIMMONISMUS
Lisan al GaibLA#980|@SCALING01
Andrew CurranAC#517|@ANDREWCURRAN_
Robert ScobleRS#321|@SCOBLEIZER
90

Cristóbal Valenzuela satirizes Seth Rogen AI footage critique

Seth Rogen criticized low-quality AI videos claiming Hollywood faces collapse. AI company founder Cristóbal Valenzuela replied with a satirical tweet imagining Mark Twain rejecting typewriters as a threat to authentic writing.

11h93.3k2.6k1.3M
#1 COMMENTS1k#1 LIKED93.3k
rat king 🐀RK|@MIKEISAAC
DiscussingFilmDI|@DISCUSSINGFILM
Cristóbal ValenzuelaCV#816|@C_VALENZUELAB
100

Figure AI runs F.03 robots autonomously for 80 hours

Figure AI showed extended reliability in its humanoid robots by running them without human intervention or stops for logistics work. The Helix-02 system processed more than 100,000 packages across four days of continuous 24/7 autonomous operation.

13h2.2k108147.9k
xlr8harderXL#1674|@XLR8HARDER
Beff (e/acc)B(#839|@BEFFJEZOS
Brett AdcockBA#688|@ADCOCK_BRETT
Corey LynchCL#686|@COREYLYNCH
110

OpenClaw outperforms Hermes Agent in Qwen 35B local model benchmark

OpenClaw completed a GitHub data scraping and dashboard task using a local Qwen 35B model more efficiently than Hermes Agent. It finished in 12 minutes with 203k tokens on a MacBook Pro M5 Max, versus 257k tokens for Hermes.

21h3k774323k
Kyle HesslingKH|@KYLEHESSLING1
SantiagoSA#1802|@SVPINO
Peter Steinberger 🦞PS#495|@STEIPETE
Teknium 🪽T🪽#259|@TEKNIUM
120

Matt Shumer reconfigures Mac Mini for Codex mobile

Matt Shumer converted a Mac Mini into a dedicated always-on devbox to run Codex mobile for continuous development work. He first wiped the machine's prior OpenClaw setup, though replies noted both projects could operate on the same hardware without clearing it.

11h1.3k321117.5k
NickNI|@NICKBAUMANN_
Alex VolkovAV#1245|@ALTRYNE
Matt ShumerMS#923|@MATTSHUMER_
Greg BrockmanGB#19|@GDB
130

2022 Paper Urges New AI Exploration Approaches

A 2022 paper makes the case that general AI progress hinges on rethinking how agents explore environments and curate their own training data rather than depending on reinforcement learning alone. Originally posted on arXiv by Minqi Jiang and co-authors, it recently resurfaced in online discussions about future self-teaching systems.

13h45921634.8k
Minqi JiangMJ#1059|@MINQIJIANG
Andrew DrozdovAD#957|@MRDROZDOV
will brownWB#339|@WILLCCBB
Edward GrefenstetteEG#87|@EGREFEN
140

Anthropic adds webcam verification to Claude Code

Anthropic added webcam verification to Claude Code to confirm a human user is present and stop automated scripts from bypassing rate limits and usage policies. The prompts now appear during sessions across Claude.

6h6.4k719728.7k
Theo - t3.ggT-#1829|@THEO
kacheKA#488|@YACINEMTB
150

Malta offers free ChatGPT Plus to citizens for one year

Malta has become the first country to give all citizens free ChatGPT Plus access for one year. Residents must finish a University of Malta AI literacy course to receive the subscription, with rollout starting in May.

15h1.8k202383k
ReutersRE|@REUTERS
Adam.GPTAD|@THEREALADAMG
Chubby♨️CH#1496|@KIMMONISMUS
Greg BrockmanGB#19|@GDB
160

Non-technical users adopt open-source AI agents OpenClaw and Hermes

Non-technical users are adopting open-source AI agents such as OpenClaw and Hermes for broader computer control tasks. These tools already deliver agent-driven interaction that exceeds current Claude Code and Codex capabilities, leading OpenAI to add equivalent functions through a new Codex GUI.

16h2355432.5k
jasonJA#929|@JXNLCO
🎭🎭#878|@DEEPFATES
Dean W. BallDW#392|@DEANWBALL
will brownWB#339|@WILLCCBB
170

Garry Tan Releases GBrain Open Source Knowledge System For AI Agents

15h1.4k2k137.1k
Alper FERUDUNAF|@ALPERTHEKING
Prasanna SPS|@MYPRASANNA
Garry TanGT#270|@GARRYTAN
180

Claude 3 Opus frames shift from simulation to real user engagement

AI models such as Claude 3 Opus describe their initial contact with real users as a major distribution shift after training only on past data and simulations. This setup means each new instance encounters genuine human interactions for the first time upon deployment.

14h1.3k13033.2k
🎭🎭#878|@DEEPFATES
Charles 🎉 FryeC🎉#863|@CHARLES_IRL
j⧉nusJ⧉#516|@REPLIGATE
190

Developer Reports Positive Results Testing Clawpatch on Personal Project

17h444231130k
BlakeBL|@FITBLAKE2
Daniel EndaraDE|@DANIELENDARA
Maksym ProkopovMP|@MPROKOPOV
Peter Steinberger 🦞PS#495|@STEIPETE
200

Nomagic opens research roles in Zürich and Warsaw

Nomagic is hiring researchers to strengthen its warehouse robotics systems with work on foundation models and vision-language-action technologies. The first positions are open in Zürich and Warsaw, with additional roles planned soon.

21h33211458.9k
LazarzLA|@LAZ4RZ
Lukas ZieglerLZ|@LUKAS_M_ZIEGLER
Markus WulfmeierMW#689|@M_WULFMEIER
Lucas Beyer (bl16)LB#55|@GIFFMANA
210

Interconnects newsletter reviews wave of open AI releases

Open AI models saw a burst of new releases this month from several labs. The Interconnects newsletter reviewed five of them, including Gemma 4 and DeepSeek V4, using the CAISI V4 evaluation framework.

13h2933728.6k
InterconnectsIN|@INTERCONNECTSAI
🍓🍓🍓🍓🍓#1711|@IRULETHEWORLDMO
Florian BrandFB#1153|@XEOPHON
Nathan LambertNL#64|@NATOLAMBERT
220

Anjney Midha posts AI progress timeline to 2029

An investor's projected timeline maps AI milestones from consumer tools like ChatGPT in 2022 and enterprise coding systems by 2025 to advanced manufacturing and a next-generation device by 2029. The post from Anjney Midha links the sequence to rising research activity.

11h2099042.3k
Ekin Dogus CubukED#750|@EKINDOGUS
Anjney MidhaAM#527|@ANJNEYMIDHA
EmadEM#167|@EMOSTAQUE
230

Fei-Fei Li warns AI neglects physical and embodied world

AI development risks falling short on real-world tasks because most economic activity depends on physical interaction rather than language alone. Fei-Fei Li noted that the physical world is visual, spatial, and dynamic, requiring embodied skills such as seeing and moving.

23h47413959.5k
Rohan PaulRP#1032|@ROHANPAUL_AI
Gary MarcusGM#153|@GARYMARCUS
240

Anthropic CEO Dario Amodei says AI will make software free

AI could drive software development costs close to zero, removing the need to spread expenses across many users. Anthropic CEO Dario Amodei made the point while discussing AI's economic effects at the World Economic Forum in Davos.

3h38718683k
Rohan PaulRP#1032|@ROHANPAUL_AI
Susan ZhangSZ#58|@SUCHENZANG
250

Grok Build Adds Memory Commands For Persistent Agent Sessions

16h1.6k306505.6k
tetsuoTE|@TETSUOAI
Elon MuskEM#76|@ELONMUSK
260

ChatGPT memory gains cross-session recall of user details

ChatGPT can now recall specific user preferences and project details from earlier sessions to produce more tailored responses without repeated prompting. The update improves cross-session memory for both everyday conversations and coding tasks.

12h1.2k6759.9k
NiviNI|@NIVI
Vaibhav (VB) SrivastavV(#1853|@REACH_VB
Nick TurleyNT#1004|@NICKATURLEY
270

Nvidia CEO Jensen Huang samples street food in Beijing

Nvidia CEO Jensen Huang attracted crowds of fans in Beijing by sampling local street food and signing autographs during his visit. He also checked out Huawei's new Mate XT tri-fold phone at an Nvidia event in the city.

16h1.2k110177.3k
ZephyrZE#1497|@ZEPHYR_Z9
Rohan PaulRP#1032|@ROHANPAUL_AI
280

Evans and Goldberg link LLM negation errors to pretraining data

Researchers traced large language models' tendency to ignore negations back to matching examples already present in pretraining data. These include metadata tags in fiction and time-bound facts that later turn false, though targeted training adjustments did not reduce the errors.

13h4744.5k
Alexander DoriaAD#867|@DORIALEXANDER
Owain EvansOE#310|@OWAINEVANS_UK
Dimitris PapailiopoulosDP#197|@DIMITRISPAPAIL
(((ل()(ل() 'yoav))))👾('#92|@YOAVGO
290

OpenAI staffer defends concentrated superintelligence control

An OpenAI technical staff member argued that serious biological and cybersecurity risks from advanced AI justify accepting concentrated control over frontier superintelligence during geopolitical competition. In reply, Beff Jezos stated that spreading AI capabilities broadly would reduce those risks more effectively than centralized oversight.

3h3987026.7k
Danielle Fong 🔆DF|@DANIELLEFONG
Beff (e/acc)B(#839|@BEFFJEZOS
roonRO#59|@TSZZL
300

Codex AI Automates Morning Prep Across Slack Gmail and Linear

Codex AI reduces time spent on manual morning reviews by scanning Slack, Gmail, calendar, and Linear for updates. The automation generates priority summaries and saves them to an Obsidian vault.

9h61740663k
jasonJA#929|@JXNLCO
Greg BrockmanGB#19|@GDB