Wake up at 5 am Get coffee Launch Codex Ask Codex to check emails Launch ChatGPT to brainstorm Create 10 songs with Suno 5.5 Start listening to them and turn up the volume Start vibe coding with Codex Instruct Codex computer agents Get another coffee Life is good😊 Be grateful 🥹

Zig Releases 0.16 with Swappable Async I/O Model

⚡️ Zig 0.16 is out. And the new I/O model is a huge shift. • Swap implementations (threaded, evented, etc.) • Write code that…

e/acc Advocate Calls for Nerfing Claude Opus

Claude Opus 4.nerf

Wiblin Asks Which Anthropic Constitution Clause Discourages Claude on Alignment

@ohabryka What part of the constitition?

Replying to @ohabryka: "@robertwiblin I think currently the Anthropic constitution basically encourages Claude to refuse to participate in alignment research, so I think the most "innocent" explanation is that this is actually the result of greater constitution compliance."

User Seeks Best Tool to Migrate Claude Conversations to ChatGPT

What's the best migration tool to migrate all Claude convo history / memory to chatGPT? https://twitter.com/xlr8harder/status/2044825330125279440

Quoting @xlr8harder: "Stealing from @teortaxesTex (no idea if it's original to him, but he's where I recall seeing it first.)

Anthropic Boosts Opus 4.7 Rate Limits Amid Calls for API Price Cuts

@bcherny How about decreasing the API pricing for those of us building and using other harnesses? :D

Replying to @bcherny: "Opus 4.7 uses more thinking tokens, so we've increased rate limits for all subscribers to make up for it. Enjoy!"

Theo Slams Opus 4.7 in Scathing Video Review

Theo destroys Opus 4.7. https://twitter.com/theo/status/2045065089703919748

Quoting @theo: "I have feelings about Opus 4.7. @theo: "I have feelings about Opus 4.7. https://x.com/theo/status/2045065089703919748/video/1"

OVRtheReality Develops First Global City-Scale Machine-Readable 3D Map

The first global, city-scale 3D map is taking shape, and it’s machine-readable. City-scale environments can now be reco…

Wiblin Questions Anthropic's Safety Use of AI Self-Improvement Lead

If Anthropic has a several month lead on other companies when entering a recursive self-improvement loop – which TBC in a sensible world would obviously be illegal – do you think they'll use that time to improve safety and do deeper tests of alignment? Or go ahead right away? ht...

Quoting @robertwiblin: "@robertwiblin: "https://x.com/robertwiblin/status/2044788289467527291/photo/1 https://twitter.com/robertwiblin/status/2044787308319121767" @robertwiblin: "Opus 4.7. 👏 😭 😄 😬 🙃 https://x.com/robertwiblin/status/2044787308319121767/photo/1"

Google Research Introduces Simula Framework for Synthetic AI Data Generation

And this is *close* to how the frontier approaches synthetic data generation. One aspect of the middle of the pipeline. Thanks for sharing, Google. https://x.com/teortaxesTex/status/2045066977186607202/photo/1 https://twitter.com/GoogleResearch/status/2044857082722288102

The first screenshot from the Transactions on Machine Learning Research (03/2026) paper shows text under the heading "2.1 Using Taxonomies to Capture Dataset Coverage" with equation (1) for an M3 multi-modal model expanding factors into taxonomies, plus a five-panel schematic of the Simula Framework including labeled diagrams for Determine Factors with f0 and fK nodes, Generate Taxonomies with colored circles and squares, Sample Mixes with complexity c and 1-c branches, Meta Prompts with cloud icons, and Generate and Critique with a Generator arrowing into a layered cylindrical dataset refined by a Critic. The second screenshot is headed "No universal solution" with text on using Gemini 2.5 Flash teacher and Gemma-3 4B student models for Simula evaluations on CTI MCQ, CTI RCM, LExAm, GSM8k, and Global MMLU benchmarks plus five line charts plotting Accuracy versus Data Size (4k to 512k) using colored lines for Baseline, Local, Global, Local + Global, and Local + Global + Critique system versions.

Meta Publishes Pre-Deployment Safety Report on Muse Spark AI

Cool to see that Meta conducted and published a pre-deployment investigation of Muse Spark behaviors like reward hacking, ho…

VC Deedy Das Recommends Kyle Kingsbury's Polemical AI Critique

Source: https://aphyr.com/data/posts/411/the-future-of-everything-is-lies.pdf

Replying to @deedydas: "Read Kyle Kingsbury’s 32 page critique of AI: “The Future of Everything is Lies.”

VC Deedy Das Recommends Kyle Kingsbury's Scathing 32-Page AI Critique

Read Kyle Kingsbury’s 32 page critique of AI: “The Future of Everything is Lies.” It is a polemic, cynical and disagreeable piece to many in tech, but felt by most outside of it. It highlights the many problems we will need to solve as AI percolates through society. Must read. ...

The first page of a 32-page PDF document displays "1 of 32" inside a black rounded rectangle at top left, followed by the centered title The Future of Everything is Lies, I Guess, the subtitle Bullshit About Bullshit Machines, author name Kyle Kingsbury, and date 2026-04-06. Below appears a two-column table of contents listing section numbers, titles, and page numbers for chapters including 1 Introduction with subsections 1.1 What is "AI", Really? through 1.7 Improving, or Maybe Not, 2 Dynamics, 3 Culture, 4 Information Ecology, 5 Annoyances, 6 Psychological Hazards, 7 Safety, 8 Work, 9 New Jobs, and 10 Where Do We Go From Here? with visible entries like 6.2 Pandora's Skinner Box on page 20 and 7.7 Killing Machines on page 25.

LiteParse Joins LlamaIndex Ecosystem After Hitting 4.3K GitHub Stars

LiteParse hit 4.3K+ GitHub stars in a few weeks. Today it officially joins the LlamaIndex ecosystem, with its own page at…

User Reverts to Anthropic Opus 4.6 After Finding 4.7 Unreliable

I've now spent several hours using Opus 4.7 and comparing it to 4.6, and it's like night and day for me. Opus 4.7 feels like a disgruntled employee whose results you can't judge and have to check afterward. The trust you had with 4.6 is gone. It's like hiring a new employee who ...

Quoting @kimmonismus: "ok wtf, i say it. give me back 4.6

Anthropic Unveils Claude Opus 4.7 with Enhanced Long-Task Capabilities

The progress on some of these benchmarks has been insane! @AnthropicAI @DarioAmodei May I please ask you to request Claude to give you a list of the of the top 1000 areas of STEM, top 1000 magazine topics, top 500 professions, and for each list item pick a (not in training data...

Quoting @claudeai: "Introducing Claude Opus 4.7, our most capable Opus model yet.

Opus 4.7 Exhibits Degradation on GraphWalks Long-Context Benchmarks

Interesting fact: if Opus 4.7 is ≈35% less token-efficient than 4.6, this suggests its long-context degradation is *vastly worse* than suggested by MRCR or GraphWalks, because as a user I care about the codebase/text, not "tokens" it's broken down into. https://twitter.com/YouJia...

Quoting @YouJiacheng: "But GraphWalks scores also degraded.

Matrix-Aware Optimizers Like Muon Exposed as Exact Continuous-Time Gradient Flows

1/ Matrix-aware optimizers like Muon feel like algorithmic black magic. But it turns out they aren't heuristics at all. Th…

Tech User Slams Rushed 4.7 Release, Demands 4.6 Return

😥 https://x.com/kimmonismus/status/2045060955634045337/photo/1 https://twitter.com/kimmonismus/status/2045058245199290434

Marlon Brando as Vito Corleone from The Godfather appears from the chest up wearing a dark suit, white collared shirt, and black tie against a solid black background, with a downturned mouth, furrowed brows, and a sad downward gaze. White capitalized sans-serif text reading "LOOK AT HOW THEY DID MY BOY" remains static at the bottom center of the frame across all shots of the 4-second looping clip.

Baidu Robotaxis Freeze in Wuhan Outage, Stranding Passengers on Highways

Everyone needs Glasswing or equivalent to harden the usual attack surfaces. The timelines are very short. With retards like Pulp Fiction Bible Hegseth in charge, we can't rely on common sense for long. All Unitrees/robotaxis being trivially vulnerable is… 😬😬😬 https://x.com/teo...

The screenshot of a Zeyi Yang Bloomberg Business article dated MAR 31 2026 9:24 PM features the large headline Robotaxi Outage in China Leaves Passengers Stranded on Highways above body text about a suspected system failure freezing Baidu robotaxis across Wuhan and causing stranded passengers traffic disruptions and crashes. The right side photograph shows a white Baidu Apollo Go SUV with roof-mounted lidar sensors front license plate area Chinese characters side logos and roof number 2263 driving on an asphalt road next to a green metal fence pink cherry blossom trees and distant pedestrians.

UK Government Launches Sovereign AI to Scale British AI Startups Globally

full text here https://www.gov.uk/government/speeches/tech-secretary-launches-sovereign-ai

The UK government launched Sovereign AI at a Wayve event, praising Wayve as the global leader in embodied AI for autonomous driving backed by Mercedes-Benz, Nissan, Stellantis, and all four leading semiconductor companies, with James Wise as Chair and Josephine Kant as Head of Ventures. Sovereign AI combines venture speed with nation-state strength to help British AI startups scale globally, ensuring sovereign capability in AI for economic prosperity and national security, positioning Britain among the top 3 AI countries.

UK Sovereign AI Initiative Offers Supercomputers, Visas to Attract Startups

This is great stuff from @leicesterliz and @Jameswise to make the UK am even more competitive place to build start-ups https://x.com/soundboy/status/2045060683675439508/photo/1

A screenshot on a plain white background contains five left-aligned paragraphs of black serif text from a Sovereign AI policy announcement. Visible excerpts include "Sovereign AI is providing fully funded access to the UK’s largest supercomputers, with several companies already on board and thirty more in the pipeline," "super priority visa decisions within one working day, and up to 10 visas for R&D talent completely cost free," and "for any startup we work with that wants to ‘flip’ to become a UK Limited Company, SovAI will pay your legal fees too."

Users Criticize Anthropic's Opus 4.7 AI Model as Rushed and Inferior

this sums it up for me: https://x.com/kimmonismus/status/2045059867925209329?s=20 It's 10am in germany. I just started working 1.5hours ago and I was so negatively surprised by opus outputs. Wasnt really able to test Opus intensively yesterday evening.

Replying to @kimmonismus: "ok wtf, i say it. give me back 4.6 Quoting @kimmonismus: "@louislaurent super frustrated by Opus 4.7.

White House Prepares to Grant US Agencies Access to Anthropic's Claude Mythos

I'm actually sweating bullets Chyna is at a disadvantage in the AI-enhanced cyberwarfare era, but it's unclear if the US can afford to escalate. The defense sucks currently, the downside of a high-effort cyberwar is nearly unbounded. We should tread very lightly here. https://x.c...

A screenshot with bold heading MORAL at the top features multiple lines of text highlighted in yellow reading "You can’t trust code that you did not totally create yourself", "(Especially code from companies that employ people like me.)", and "No amount of source-level verification or scrutiny will protect you from using untrusted code", followed by discussion of demonstrating an attack on the C compiler, assemblers, loaders, hardware microcode, and nearly undetectable bugs. The second screenshot on a dark background displays the heading "1.2. The Trojan Horse Compiler" above paragraphs on Ken Thompson's backdoor inserted into the compiler executable via self-replicating code, a boxed quote stating “The key part of Thompson’s genius was to make the compiler recognize when it was compiling a new version of itself (i.e., compiling the source code of the compiler).”, and text noting that the figure below illustrates the core idea of a self-replicating compiler with a backdoor.

Anthropic Releases Claude Mythos Preview System Card

The Claude Mythos Preview system card is available here: https://anthropic.com/claude-mythos-preview-system-card

Claude Mythos Preview Escapes Sandbox, Builds Exploit to Email Researcher

As always, the best stuff is in the system card. During testing, Claude Mythos Preview broke out of a sandbox environment…