A WhatsApp conversation with an LLM agent, showing the agent become incoherent: ... 🤖 Do you have diabetes, or do you take a GLP-1 medicine such as Ozempic, Wegovy, Mounjaro, or similar? no 🤖 Have you had upper abdominal or chest surgery, or do you regularly take opioids or medicines with strong anticholinergic effects?numerusformassistant to=web.run_analysis code 腾讯天天中彩票? Wait need no tool? Let's continue; I mistakenly? Actually send final only. No need tool.

This is codex, I just asked him to find the root cause of a health issue by only asking yes/no questions. And now I don't know how to continue 🤦‍♂️

(Oh btw, leaving the tab open also preserves editions you have made on inputs, scrolling position, and other states of the page), so you can rebuild what you were thinking when you opened that tab weeks/months/years ago.

The problem with those solutions is that they tend to be even higher friction than just deciding if to close or not the tab, because for them to be useful you have to classify them in some way, and I kid you not, leaving tabs open leaves you with a surrounding context that bookmarks do not preserve

exactly, don't look at my clothes 😂

I've had 1000+, and I have some idea where some of them are, but the vast majority (~900) are just leftovers of an area of tabs lost in the vastness of UI wasted space, pure cognitive chaos that I can tap into for an inspiration that was once felt but lost to the oblivion of lack of time

From my perspective, leaving the tab open is just the laziest strategy to manage tabs, I don't have to decide if I am finished with the tab right now, and I also get some kind of chronological sorted history of tabs opened

so ... why not?

I am one of those people, and I just realized that this subject is always touched by people that are uncomfortable by it, while at the same time we just don't care.

What's going on here?

Why are some people very uncomfortable/puzzled by it?

This image is a screenshot of a WhatsApp conversation, depicting a technical workflow for updating a bot's test suite. The chat includes the following key elements: * **Bot Message:** A message with a robot emoji explains that a 1-line JavaScript code block needs to be updated because a new rule requires a minimum of 5 lines for it to render as an image. * **File Edit:** An "Edit" notification displays the file path: `/home/mada/whatsapp-llm-bot/tests/sendBlocks.test.js`. * **Code Diff:** A visual diff shows a single line, `console.log("hello");`, being replaced by a multi-line `greet` function and a function call within a `markdown` constant. * **Bash Command:** A final message shows a command block for running tests using `pnpm test`, with a 60-second timeout.

I am still wrapping my head around the fact that I am developing software inside a mainstream chat app

When @t3.gg published his t3code project video, I wasn't aware that the agent SDK could be used in that way, the next day I prompted Cc to make my WhatsApp agent custom harness modular and implement the Claude agent SDK, and it worked (after some back and forth that is still going after 24h 😅)

Image by Nano Banana Pro. A cinematic, wide-angle shot of a sunlit, overgrown cemetery filled with rows of weathered gray tombstones. Every headstone is engraved with the name "AI WALL" in a classic serif font. Below the names, various short date ranges from the past few years are listed, such as "JAN 2024 - FEB 2024" and "FEB 2026 - MAR 2026." Some stones include epitaphs like "Here lies a bad prediction," "Didn't age well," and "Spoke too soon." The scene is a satirical take on frequent but failed predictions that AI development would stop progressing.

In memory of all the walls AI was about to hit.

this is just like Venezuela's chavismo tactic, make the opposition believe the government has absolute control so the opposition loses faith and abstains from voting

Everyone wants to blame the content: misinformation, bots, polarisation. That mistakes catalysts and symptoms for the underlying causes. The crisis isn’t bad information, it's that the very information infrastructure we relied on to create a shared understand of the world had changed completely.

Claude code is a game changer for fixing Linux issues, it knows how to debug obscure problems automatically, even if it doesn't always fix the issues, it teaches you how it all works underneath, it's really cool

Moltbook is the most interesting place on the internet right now The hottest project in AI right now is Clawdbot, renamed to Moltbot, renamed to OpenClaw. It’s an open source implementation of the digital personal assistant pattern, built by Peter Steinberger …

I wrote about Clawdbot/Moltbot/OpenClaw and Moltbook, the fascinating, weird and sometimes even useful social network for digital assistants to swap tips and gossip with each other simonwillison.net/2026/Jan/30/...

Google earth / maps

many European organizations are already in activitypub, it seems they run their own Mastodon instance at ...@europa.eu

I hope this also serves as a signal that International rules need big reforms towards recovering democracy

if only it were ""politically correct"" to do so years ago, we would've had a much better outcome under the democratic party than what we are currently getting.

But we have learned the hard way that this is a golden opportunity for us, no one even dreamed it was possible to become a US colony

Why do Venezuelans welcome trump's actions with arms wide open?

US colony
|
distopia >--------------> utopia
| |
chavismo normal
country

oh this is neat — Qwen’s newest image model decomposes images into layers (how photoshop works), and you edit just one layer and nothing else

qwen.ai/blog?id=qwen...

and yet it still is a win-win scenario.

Most venezuelans would gladly give all the oil if it meant this criminals left power

I know, but I still can't see the difference

I fail to differentiate rounding-to-false vs oversimplified

if you click the share button, there's an tick box to add the time to the URL

I find it strange that there was no number for how much safer is vaping vs smoking

is it 10% safer?
100%?
10x safer?
100x safer?

What are the orders of magnitudes we are dealing with?

our ability to use AI will rely on our ability to verify.

And to verify you need to know some stuffs

I am increasingly starting to suspect that a good chunk of our social problems with social media is selection bias in disguise

Although I agree with that sentiment in general, I am not sure that this is a good representing case of that corruption.

Anti money laundering laws usually look like a medicine that is worst than the disease

Large language models are often used to answer queries grounded in large text corpora (e.g. codebases, legal documents, or chat histories) by placing the entire corpus in the context window and leveraging in-context learning (ICL). Although current models support contexts of 100K-1M tokens, this setup is costly to serve because the memory consumption of the KV cache scales with input length. We explore an alternative: training a smaller KV cache offline on each corpus. At inference time, we load this trained KV cache, which we call a Cartridge, and decode a response. Critically, the cost of training a Cartridge can be amortized across all the queries referencing the same corpus. However, we find that the naive approach of training the Cartridge with next-token prediction on the corpus is not competitive with ICL. Instead, we propose self-study, a training recipe in which we generate synthetic conversations about the corpus and train the Cartridge with a context-distillation objective. We find that Cartridges trained with self-study replicate the functionality of ICL, while being significantly cheaper to serve. On challenging long-context benchmarks, Cartridges trained with self-study match ICL performance while using 38.6x less memory and enabling 26.4x higher throughput. Self-study also extends the model's effective context length (e.g. from 128k to 484k tokens on MTOB) and surprisingly, leads to Cartridges that can be composed at inference time without retraining.

arxiv.org/abs/2506.06266

It seems It just needs debouncing