Unclecode (Hossein)

@ChatGPTapp Could you add multi-select bulk delete/archive for recent sessions? Plzzz

Ummm It more looks like a temporary band-aid solution because Claude can't operate on its own plan. I remember The Opus 4.7 hype was so attached to the plan, and now this shows we're not there yet. I don’t like it, by the way, using a hook to classify with Haiku and logging responses in fire-and-forget mode is much better than forcing the model to keep track and consume tokens.

really fucking good

Ummm It more looks like a temporary band-aid solution because Claude can't operate on its own plan. The Opus 4.7 hype was so attached to the plan, and now this shows we're not there yet. By the way, using a hook to classify with Haiku and logging responses in fire-and-forget mode is much better than forcing the model to keep track and note and consume more tokens.

a prompt I've been using a lot recently: implement <SPEC> and while you do, keep a running implementation-notes.html file (or markdown) with decisions you had to make weren't in the spec, things you had to change, tradeoffs you had to make or anything else I should know

"You cannot govern a technology you've only been briefed on." - @VivianBala The second-order signal in this story is the one worth keeping: when a Foreign Minister ships on a Raspberry Pi, the space of possible solutions just got bigger. Problems still scale exponentially, so the answer isn't fewer developers, it's more of us, working on harder problems. And what makes this work at Pi-scale isn't a bigger model. It's better data. A small LLM fed clean, well-sourced, well-curated context will out-reason a frontier model fed sludge. Data quality is the lever, not parameter count. That's the thesis behind Context4AI - what we're building in Singapore on top of @crawl4ai: authoritative extraction first, model second. Curate the input, the rest gets cheap

Tell me who is the robot? The human has to answer this or the one made this Captcha!!!!

@jasonzhou1993 😂 this is not anymore bot detection, even challenging for human, this police can use in road when wanna know if a driver is drunk or not!!

@unclecode things are going cra cra!

WTF is this Captcha!!

How do you keep Claude working until the job is done? Claude Code helps with this in a few ways, including one we shipped recently: /goal.

@alex_whedon @edreisMD Applied 🤞

Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is: - 52x faster than FlashAttention at 1MM tokens - Less than 5% the cost of Opus Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter. @subquadratic finds and focuses only on the ones that do. That's nearly 1,000x less compute and a new way for LLMs to scale.

> Open-source models (Kimi K2.6, DeepSeek v4) are now on par with Opus/GPT-5.5 at a fraction of the cost no hitting limits constantly > The real insight isn't who "won" - it's that the moat is gone. Anyone can run frontier-quality models cheaply > The harness matters as much as the model > Benchmarks are basically broken/meaningless - the model-vs-model war is a distraction > There's no best model, just the best one for you - Windows/Mac/Linux analogy news.ycombinator.com/item?id=479932…

Kimi K2.6 just beat Claude, GPT-5.5, and Gemini in a coding challenge thinkpol.ca/2026/04/30/an-…

@Dan_Jeffries1 @ylecun Well, one of these folk he is criticizing is himself when a year ago he was saying “kids no need to learn coding anymore, leave it AI”!! This was exactly a nonsensical thing a CEO would say intentionally to hype up his gou market. tomshardware.com/tech-industry/…

Jensen is one the smartest and most far seeing folks the world. "If an AI scientist warns people that AI is going to permeate across radiology and radiologists are going to get wiped out, it might seem helpful but it's hurtful. If we convince everybody not to be radiologists and we now need radiologists, that actually is hurtful to society. "It is hurtful to convince all the young college graduates not to study software engineering because we are going to need more software engineers than ever. That's hurtful." "Scaring people with nonsensical things, which are not going to happen, that this is an existential threat, there's a 20% chance that is is existential, that's ridiculous. "That it's going to wipe out 50% of college level jobs. "That is it going to completely destroy democracy. "These kinds of comments are not helpful. They are made by...CEOS. And you become a CEO, maybe you adopt a God complex and somehow you know everything." Brutal. And right.

OpenAI Symphony is actually great You can 5x Coding agent outcome with it + Good codebase harness Played with it past few days and here is my setup: - Playwright CLI - Boot skill - WORKFLOW.md ... Broken it down in 12 mins below so you can do the same 👇

Did you see the claude.ai/design onboarding animation? So cool! Agree?

Do u get this? I get it time to time, then I reply “continue” it works! Thinking to add in my hook to retry automatically by itself 😅 @ClaudeDevs

@claudeai How to hand off Claude code ? This is amazing, i hope claude code also get “edit on designer” then we can continue there like ultraplan

Introducing Claude Design by Anthropic Labs: make prototypes, slides, and one-pagers by talking to Claude. Powered by Claude Opus 4.7, our most capable vision model. Available in research preview on the Pro, Max, Team, and Enterprise plans, rolling out throughout the day.

Some useful tips for Claude Code with Opus 4.7: 1/ Auto mode (Shift+Tab): skip permission prompts, let Claude run complex long tasks like deep research and refactoring without interruption 2/ /fewer-permission-prompts: scans your session history, finds safe commands that keep triggering approvals, auto-whitelists them 3/ /recap: short summary of what the agent did and what is next, useful when returning to a long session after a break 4/ /focus: hides intermediate work, shows only the final result, for when you trust the model enough to let it run. Spend mire time on planning, so detailed that then u let it go without babysitting. 4/ /effort or Effort level: Opus 4.7 uses adaptive thinking instead of budgets, tune effort up for complex tasks, down for quick ones 5/ Give Claude a way to check itself, 2-3x improvement in output quality, more important than ever with longer autonomous runs. For example, make your own custom skill that runs 1/ end-to-end tests 2/ /simplify 3/ /ultrareview 4/ send PR. One command, full cycle.

Quick read of the chart: 4.7 leads on tool use, computer use, MCP. Quietly regresses vs 4.6 on agentic search and cybersec vulnerability reproduction. This is an orchestration tuning, not raw capability. Mythos Preview is the headline you missed if you only looked at the orange column.

Introducing Claude Opus 4.7, our most capable Opus model yet. It handles long-running tasks with more rigor, follows instructions more precisely, and verifies its own outputs before reporting back. You can hand off your hardest work with less supervision.