iiinova — where your digital self keeps evolving

And now you want to ban social media, right? First, you can't let the kids play in the streets and now the internet can't be like the streets.

000

If you tie your kid to a very narrow identity online, you can expect them to take some big emotional hits when that gets attacked as an extension of who they are at school. They can't escape.

100

Cyberbullying doesn't begin in schools. It begins in a society that can only go online if everyone has only one identity that has to match with your ID.

100

Michael Crichton - Wikipedia

No mention of the recent nominee for dumbest #AI take ever from the PotUS. No mention of the recent initiative taken by Bernie Sanders. Let's never forget how dumb the news can be: en.wikipedia.org/wiki/Michael... #Portugal

en.wikipedia.org

000

It's already friday and Portuguese media are still reporting on a cold slice of last week #AI news. This just in: ex-Anthropic employee says AI can kill us all. No mention of the renewed consensus from OpenAI, Anthropic and xAI. No mention of the official response from China. #Portugal

100

That approach will only fundamentally improve when models go beyond language in how they interact with the world.

000

Meanwhile, if your #HTML is just mumbo-jumbo, you're forced to address everything visually by scribbling on screenshots, hoping that the agent gets what you mean and ignoring bad structural choices that contribute to lingering issues.

100

I'm not even saying that you should look at the code, but often it helps a lot if you're able to inspect element into something that's wrong and be able to tell an agent what to do with actionable words. Like, this class, label or id, lose the padding, don't use absolute units, etc.

100

Ricardo Tavares @viewfromtheweb.com · 12/09/2026

People keep saying that frameworks like #tailwind or even #stylex are good for agents, but are they working in our current reality where even frontier LLMs still suck at frontend and #CSS?

200

guise, omarchy is just a bunch of fash scripts I mean, bash scripts just install arch, btw

011

Ricardo Tavares @viewfromtheweb.com · 10/09/2026

Apple to EU: "Want some?" #EU: "Brought enough for everyone?" #Apple: "Actually, nevermind."

000

Happy birthday #StarTrek May the force be with you 😉always.

010

Eventually, LLMs will still have limitations, but these won't be so relevant and it's also likely that AI will have moved on to something else by then. PMs can only take us so far anyway. Hopefully, scientists will still know what we're doing.

000

Whenever we can identify some way in which LLMs can't count Rs on the word strawberry, that's fixed now. Turns out that product managers can take over the world if we let them. Given infinite resources, language can be composed in infinite ways.

100

But the beauty of open source, is that we can all learn something every day. Maybe don't chrome extension into yt-dlp into mvp config RCE, but you do you. Never take no for an answer. Unless that answer is certainly.

100

This just in: to the surprise of pretty much zero people, the new vibe-coded bash-scripted arch-based OS gone viral is filled with weird vulnerabilities. The trick is to just keep adding money, I guess.

100

#javascript arrays amirite moar like disarrays

000

Ricardo Tavares @viewfromtheweb.com · 07/09/2026

#git hosting is on a different level once you have users that need to make seven commits each time they change the colour of a button

000

GPT 6 can solve all of them with a dedicated harness. It's not an easy challenge to benchmaxx, but here we are. Close enough to call it AGI, I guess. But benchmarks are meant to be broken.

000

Ricardo Tavares @viewfromtheweb.com · 07/09/2026

Wondering what is this AGI talk about? arcprize.org/tasks/ls20 It's just a benchmark that tries to offer puzzles that are created from scratch, much like procedural videogames. You can play them yourself in this link.

arcprize.org

ARC-AGI-3 Task #ls20

Play ARC-AGI-3 task ls20

100

Ricardo Tavares @viewfromtheweb.com · 06/09/2026

Blew up again. #astra coding today: See any issue? OK, fix 💥 Finish what you were doing 💥 Finish what you were doing 💥 Good night, I guess 🫠 #aiagents #codex #openai

100

Ricardo Tavares @viewfromtheweb.com · 06/09/2026

So I've dropped down from Pro to Plus to get the average #openai Astra coding experience. Good news: you can always use your phone to tell it to go ahead after it gets interrupted by the 5h limit. Bad news: you can blow through two 5h limits with a single prompt, no fast mode, GPT 6 High.

Screenshot of a Codex (ChatGPT) coding agent session showing a scrollable log of executed commands, including several 'tail', 'python3', 'jq', 'mkdir', and 'command -v ruff' entries with their run times. Below the log, an orange usage-limit notice reads: 'You've hit your usage limit. Upgrade to Pro (https://chatgpt.com/explore/pro), visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at 6:05 PM.' A downward arrow icon and a footer showing '2 files +337 −15' (added/removed lines) appear at the bottom.

100

DeepSWE shows you the number of agent steps, it's just not the default tab.

100

keyboards: *literally a key called enter* #apple: "listen, it's cmd-o to open a folder"

120

If you don't want the next ludites, don't call it the next industrial revolution.

000

Zooming in on the latest DeepSWE lines: deepswe.datacurve.ai Gemini is now winning on cost but dramatically losing on number of agent steps, which doesn't sound so good for coding. Also this benchmark is getting saturated. SlopCodeBench when? #gemini #google #apple #aiagents

130

What are we doing to help family and friends looking to upgrade from their windows 10 machines? Given hardware prices, do you deploy the #Linux option on their old computer and embrace your fate as tech support?

020

Some napkin math for #BSD CVEs discovered with AI assistance: #FreeBSD 27 #OpenBSD 1 #NetBSD 0 #DragonFlyBSD 0

010

They already know from Linux that maintainer bandwidth is scarce and that adding provenance helps reviewers. But it's not clear for anyone where a developer stands if they're fully responsible for their code while also having a label on it saying what model generated it.

020

Debian developers looked at the policy used by the Linux kernel they distribute and decided to also accept LLM-generated code, but have less mandatory disclosure. lists.debian.org/debian-vote/...

130

"The rise of artificial intelligence is the next industrial revolution." So... decades of growth powered by innovation, widespread suffering and environmental damage, followed by worker organization and political reform that eventually secure a broader share of that wealth across society?

110