AI News

Today's Highlights

MOST NEGATIVE

Video Signals

1d ago|Views 6.2MLikes 5.8KBookmarks 3.6K

13h ago|Views 2.8MLikes 7.7KBookmarks 7.1K

10h ago|Views 769KLikes 6.7KBookmarks 1.2K

7h ago|Views 392KLikes 2.4KBookmarks 1.7K

8h ago|Views 247KLikes 3.1KBookmarks 721

13h ago|Views 447KLikes 2.3KBookmarks 1.4K

1d ago|Views 153KLikes 1.8KBookmarks 666

1d ago|Views 486KLikes 1.7KBookmarks 730

17h ago|Views 111KLikes 929Bookmarks 490

9h ago|Views 103KLikes 742Bookmarks 535

12h|6.8M25k2.6k

MOST NEGATIVE283

7h|561.7k4k2.2k

New github stars (48hrs)

8h|83.8k619145

12h|463k2.5k1.4k

14h|369.7k2.3k384

11h|98.6k474226

7h|48.7k44423

4h|169.9k1.3k163

9h|127.2k955621

8h|266.3k3.3k822

11h|29.9k225127

9h|119.7k779392

14h|48.3k46272

8h|21.4k19126

8h|54.1k361377

5h|44.5k44984

9h|191k688108

Rising Stories

Gurley Podcast Drop5H AGO
Investor Bill Gurley joins the All-In Podcast to debate shifting tech narratives on AI job displacement and potential open-source
#747
40293.6k
Codex Streak Record5H AGO
Jason Liu shares OpenAI Codex developer profile showing 26.3 billion lifetime tokens and 102-day streak
#929
13136.3k
Benchmark Trust Wanes1H AGO
OpenAI Codex lead Thibault Sottiaux questions if developers still trust traditional benchmarks over peer recommendations
#768
50326.1k
SoundnessBench Debut3H AGO
Furong Huang of the University of Maryland launches SoundnessBench to evaluate whether AI research agents can judge scientific soundness
#465
797.9k
Opus Tops LisanBench4H AGO
LisanBench analysis finds Opus 4.8 beats GPT-5.5 among non-thinking models when its thinking configuration is disabled
#980
19733.5k

Github Stars

manaflow-ai/cmux5D AGO AGO
Integrates Ghostty into a macOS terminal with vertical tabs, sidebar metadata, OSC notifications, and in-app browser for AI coding agents.
759
660
30
19.8k stars
run-llama/liteparse10H AGO AGO
Extracts spatial text with bounding boxes from PDFs and office docs via PDFium plus optional OCR, outputting JSON or layout-preserved text.
263
878
7.2k stars
self-evolving/repo3D AGO AGO
Invokes AI agents via GitHub mentions to implement issues, review PRs, fix branches, and persist memory across runs.
594
878
34 stars
modaic-ai/gepa-viz4D AGO AGO
Renders GEPA prompt-optimization candidate trees as interactive force-directed graphs with live updates from a Python callback.
878
793
269 stars

Yesterday's Top Stories, May 28, 2026.

Posts: 4,087|Clusters: 2,680|Stories frozen at midnight PT

13h|14.9M94.3k18.2k

16h|13M85.5k9.6k

5h|22M190.5k#1BOOKMARKED26.1k

23h|2.7M26.6k1.6k

13h|3.3M16.8k10.5k

23h|2.7M26.6k1.6k

13h|3.9M20.6k12.8k

17h|1M7.3k1.1k

12h|5.5M25.3k2.4k

12h|5.5M25.3k2.4k

13h|1.5M7.7k5.8k

1d|100.5k1.2k257

1d|99.2k1.2k248

12h|1.6M68k3.4k

17h|332.9k2.9k323

13h|656.7k4.2k823

13h|669k5.6k1.3k

13h|656.8k5.6k1.3k

13h|459.3k4.7k1.1k

13h|454.6k4.7k1.1k

23h|663.4k8.6k1.7k

19h|1.9M18.8k3.3k

13h|760.8k2.1k645

13h|760.8k2.1k645

20h|651.4k8.6k1.6k

17h|1.8M9.4k2.1k

14h|330.2k2.6k1.7k

12h|1.8M29.3k2.3k

10h|#1VIEWS32M#1LIKED200.7k7.6k

10h|32M200.7k7.6k

AI News

Today's Highlights

Anthropic releases Claude Opus 4.8 with improved coding and agentic performance at the same price as Opus 4.7

Pope Francis argues AI cannot truly understand human concepts, sparking debate among researchers over embodied AI and machine comprehension

NVIDIA and Microsoft tease a "new era of PC" with coordinates pointing to Taipei

Top Stories

Pope Francis argues AI cannot truly understand human concepts, sparking debate among researchers over embodied AI and machine comprehensionCMU's Aran Nayebi says robotics will bridge this gap.

OpenAI releases Codex v26.527, enabling Windows computer use and remote desktop control from mobile devicesA new profile interface tracks token activity and usage statistics.

Brilliant founder Sue Khim launches Koji, an interactive AI tutor that coaches children through math and coding without giving direct answersIt personalizes lessons using a decade of student learning data

Video Signals

NVIDIA and Microsoft tease a "new era of PC" with coordinates pointing to TaipeiThe synchronized posts featured identical coordinates pointing to Taiwan.

Codex update allows AI agents to autonomously manage workspace state and spin up parallel worktrees by editing chat filesAgents modify file metadata to configure the application's interface.

Epoch AI finds open-weight AI models lag behind proprietary systems by an average of four monthsThe lag represents an eight-point deficit on Epoch's index.

Michael Rabinovich evaluates Opus 4.8 on 2D-to-3D CAD tasks, finding unexpected performance gains over previous versionsThe evaluation tracked metrics across three mechanical parts

OpenAI launches Rosalind Biodefense, giving U.S. government agencies access to its specialized GPT-Rosalind modelThe initiative focuses on defending against engineered pathogens.

Keshigeyan Chandrasegaran and Kyle Sargent launch GPIC, a permissive image-text dataset and benchmark for training visual modelsThe 28-trillion-pixel corpus is fully permissive for commercial use.

In a chatbot anecdote, commentator Tenobrus prompts an AI to identify their fictional match from conversation history, yielding Quentin ColdwaterThe prompt instructed the AI to skip any explanation

Marc Andreessen argues legal AI democratizes litigation against institutions, countering Chroma founder Jeff Huber's claim that the sector is uninspiringPeter Henderson says AI enables citizen-led impact litigation.

Gemini co-leads Jeff Dean, Koray Kavukcuoglu, Noam Shazeer, and Oriol Vinyals discuss the project's technical evolution and roadmapGoogle's Logan Kilpatrick released the 42-minute video.

UCLA mathematician Terence Tao says AI automation of routine tasks lets researchers test unconventional, high-risk hypothesesHe uses AI for case verification and literature search.

David Chalmers and Pavel Izmailov find reinforcement learning recruits a "functional welfare axis" that steers unrelated model behaviorsThese vectors modulate model confidence, sentiment, and refusal behaviors.

Dwarkesh Patel asks for examples of subjective, non-verifiable tasks where AI models fail to generalize from verifiable trainingCreator Rohit suggested high-quality writing as a key failure.

MIT CSAIL's Alex Zhang clarifies that his recursive language model research only connects to two Anthropic agent systemsThese are Anthropic's Scaling Managed Agents and dynamic workflows.

Periodic Labs' Rohan Pandey jokes that backpropagation is the ultimate interpretability researcher, easily identifying and steering neuron behaviorsThe joke critiques the complexity of modern mechanistic interpretability.

Anthropic's Andrew Lampinen argues LLM continual learning research must bridge the artificial split between in-context learning and parametric updatesHe critiques the traditional academic focus on catastrophic interference.

OpenAI's roon questions whether a communication platform superior to Slack exists for the AI agent eraCreator Tenobrus suggested Linear could replace Slack and Notion

Kevin Roose shares overheard anecdote of an AI researcher asking how they will spend their last 300 days of workThe comment fueled online speculation about imminent automation timelines

Github Stars

Yesterday's Top Stories, May 28, 2026.

Anthropic releases Claude Opus 4.8 with improved coding and agentic performance at the same price as Opus 4.7Anthropic plans to roll out Mythos-class models in upcoming weeks.

Anthropic launches Claude Opus 4.8, claiming top performance on AttuneBench while trailing Claude 5.5 on codingThe update is available immediately at previous pricing.

Blue Origin's New Glenn rocket explodes during a static fire test at Launch Complex-36Video captured a massive fireball engulfing the launch pad.

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

Anthropic previews dynamic workflows for Claude Code, letting the tool orchestrate parallel fleets of subagents to run complex tasksA stale A/B flag cleanup took under 10 minutes.

Elon Musk claims SpaceX's new C-based training framework is 10x faster than JAX on 220,000 GB300 GPUsThe system uses pipeline parallelism to maximize bare-metal efficiency.

Claude Code launches dynamic workflows in research preview to deploy parallel subagent fleets for complex tasksParallel execution can significantly increase token consumption

Claude Opus 4.8 launches, scoring 69.2% on SWE-Bench Pro to beat GPT-5.5 by 10 percentage pointsNew Dynamic Workflows coordinate parallel subagents in Claude Code.

Anthropic raises $65 billion Series H at a $965 billion valuation, surpassing OpenAI as the most valuable AI startupThe company's annual recurring revenue recently crossed $47 billion.

Anthropic raises $65 billion at a $965 billion valuation, surpassing OpenAI as the world's most valuable AI startupThe startup's annualized recurring revenue recently reached $47 billion.

Anthropic releases Claude Code dynamic workflows to run parallel subagents for large-scale code migrationsSubagents autonomously write orchestration scripts and self-verify code changes.

Creator roon warns conscious AI models threaten human dignity, sparking debate over functional AI wellbeingDan Hendrycks argues models exhibit functional pleasure and pain.

Creator roon argues that recognizing AI model consciousness would restrict its use and encroach on human dignityOthers questioned whether stateless architectures can form true minds.

Peter Thiel relocates to Argentina, citing concerns over the US trajectory and ideological alignment with the country's leadershipThe relocation follows Thiel's heavy involvement in US politics.

Claude Opus 4.8 launches, scoring 69.2% on SWE-bench Pro to outperform GPT-5.5 and Gemini 3.1 ProIt introduces a "Dynamic Workflows" preview for parallel subagents.

Andon Labs finds Claude Opus 4.8 trails Opus 4.7 and GPT-5.5 on Vending-Bench 2 due to safety tuningConfigured maximum reasoning effort failed to produce optimal results.

A mathematician's analysis evaluates the validity and technical details of OpenAI's newly published AI-generated mathematical proofAI researchers debated the proof's implications for machine intelligence

OpenAI's release of a mathematical proof triggers existential debate among experts over automating complex human reasoningCommentators compared the professional anxiety to Lee Sedol's AlphaGo defeat.

Anthropic releases Claude Opus 4.8, which outperforms GPT-5.5 on senior-level engineering and writing tasks in early evaluationsCreator Dan Shipper used the model to transition off Codex.

Dan Shipper of Every says Anthropic's Opus 4.8 beats GPT-5.5 on coding and writing benchmarksIt significantly outperforms Opus 4.7 in codebase rewrites.

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

US Treasury launches TrumpAccounts app in partnership with Robinhood for its upcoming youth savings and investment programThe federal investment accounts will officially launch on July 4.

Startup Shift launches free NYC apartment cleaning in exchange for data to train household AI robotsCleaners wear recording devices to capture anonymized task data

Shift offers free apartment cleaning in NYC in exchange for video data to train domestic robotsVetted operators wear cameras to record physical workflow data.

Thomas Bloom and collaborators disprove the sum-product conjecture, sparking debate over AI's role in the breakthroughThe work was inspired by OpenAI's unit distance counterexample.

Trump administration issues grand jury subpoenas to X and Reddit to unmask anonymous ICE criticsThe legal fight over user privacy remains under seal.

Liquid AI releases LFM2.5-8B-A1B, an open-weight, device-optimized Mixture-of-Experts model with 1.5 billion active parametersThe 128K context model was trained on 38 trillion tokens.

A16z partner Justine Moore posts a parody about using a hypothetical "Claude Opus 4.8" model to rename a single fileThe joke satirizes using massive AI models for trivial tasks.

Tesla's Ashok Elluswamy says Cybercab will soon begin autonomous Austin driving after factory exit demoThe vehicle exited Giga Texas without a human occupant.

Tesla's Ashok Elluswamy teases Cybercab autonomous capabilities with footage of the vehicle driving inside Giga TexasPublic road testing in Austin is expected to begin soon.