CM @TVKVijayHQ : Velachery J7 Police Station has been under the control of Rowdy Thesigan for the past 20 yrs.He can occupy any open terrace of any building. The open terrace of 18, Sakthi Apts, Sarathy Nagar 2nd St,Velachery,Ch-42,has been occupied by him.Strict action is needed

This is hilarious. This is what AI was made for. I love it. 100% accurate.

"Modi has not done a single press conference in 12 years." So what? Not taking questions from the press is not some moral crime in itself. The media is not a council of high priests conducting a legitimacy exam that every Prime Minister must clear for democratic certification. In a democracy, legitimacy comes from public mandate and governance, not from standing before cameras to satisfy a section of the media that mistakes access for accountability. As such, today's media ecosystem is not some sacred, neutral institution operating above politics. Much of it is deeply polarized. Different political camps have their favorite journalists, narratives, and ecosystems. Questions are designed less to seek answers and more to manufacture headlines, extract viral moments, or build personal brands. And those who hold frequent press conferences are not automatically standing on higher moral ground either. We have all seen planted easy questions from pliable journalists and difficult questions brushed aside with “aap rival party ki bhasha mat boliye,” deflections, or outright evasions. The number of press conferences a leader does is not a certificate of transparency. Otherwise, by that logic, Trump, who interacts with the press almost daily, should automatically be considered the most transparent, most democratic, and the finest leader in the world.

For you feed has gotten worse since this implementation. Even after snoozing videos and crypto, every other post i see is a video. @nikitabier please fix this.

Claude Mythos is Delusional

2140 paying users complaining about Anthropic secretly reducing its usage limits for Claude last week. This is what you call a SCAM in 2026. @AnthropicAI @claudeai The usage exhaustion below is within 1 prompt (No Code). More on: reddit.com/r/ClaudeCode/comm… #claude #claudecode

Transformers v5.3.0 adds EuroBERT (multilingual encoder, 8192 ctx, bidirectional), VibeVoice ASR (60min audio, 50 langs, diarization), TimesFM2.5, PP-DocLayoutV2, OlmoHybrid, ModernVBert, and Higgs Audio V2. github.com/huggingface/trans…

Just wrapped a 9-day Codex session powered by @OpenAI gpt-5.3-codex with High reasoning. 421 prompts. 12,112 tool calls. 1.19B tokens. It ran like an autonomous staff engineer, and then some. Started with raw OpenAPI specs. Built a full Go CLI from them. When capabilities were missing mid-implementation, it extended the CLI itself, then kept going. No hand-holding. What it shipped: OpenAPI to Go CLI: generated from specs, then self-extended whenever gaps surfaced during real usage Repo discipline: Git init, commit history, governance rules, artifact index from nothing CLI hardened: TDD standards, regression baselines, flag consistency, auth validation Helm platform rescue at scale: 2,310 files, 1,124 templates, 124 chart definitions across a multi-version enterprise platform chart suite. The primary root chart alone declares 43 direct dependencies, with subcharts nested 4 levels deep, one subsystem alone has 124 templates and 13 nested charts. Central config tops 6,300 lines. Enterprise-platform Helm complexity, not app-chart complexity. And Codex diagnosed live failures inside it. Mac to Windows pivot: hit an architecture wall on Mac, switched to a Windows PC (with Docker access I had given it), stood up a kind cluster there, and codified the new path, autonomously 4-wave Argo CD GitOps rollout: overlay-only changes across multiple namespaces, zero manual cluster drift Full stack live: DB, identity layer, broker, REST APIs, admin config smart-view UIs WSO2 PKCE/OIDC SCIM: fixed redirect/callback issues, validated flows, wrote the docs 2 auth paths proved end to end: UI to identity to WSO2 to app, and token exchange to REST CRUD with correct boundary denials This wasn't autocomplete. It was a staff engineer that reads specs, builds the tools it's missing, and when one machine won't cooperate, finds another.

I have this in CLAUDE.md but in claude cowork, opus keeps overwriting the files. Is there a better way to enforce this? @bcherny @trq212 @adocomplete

Gave Codex-5.3 a complex Helm chart with many child charts and asked it to do an install, then went to bed. Woke up to a perfectly working environment. Tried this with Opus a couple of weeks ago, it gave up in 20 minutes. @OpenAI has kicked ass with Codex 5.3!

told 5.3-codex "no shortcuts, no fake tests" right before context compacted. post-compaction it continued like nothing happened. deleted the coverage-gaming tests, wrote real ones, suite passed. didn't repeat myself once. compaction is where most models reset on you. 5.3-codex doesn't. @OpenAI

haha claude!

Placed my first @Swiggy order using my AI assistant. Feels weird but good.

Post compaction "memento" moment lol

left: opus traced the VERSION_CONFLICT chain correctly but misidentified the layer at fault. blamed transaction system design when the bug was in read callers not handling rejected receipts. right: 5.3-codex wrote python to parse logs, tsx to reproduce the race, isolated the exact bug and the fix. close doesn't ship.

opus 4.6 spent 32 seconds and suggested a fix. 5.3-codex spent 5 minutes reproducing the bug outside the app with raw HTTPS calls to api.telegram.org, tested default vs forced IPv4, isolated it to IPv4-specific routing. this is why I keep saying we've turned a page with 5.3-codex. one of many examples where it just outperforms opus. still use opus constantly and keeping my Claude Max plan, but knowing when to use which model matters now more than ever!