iiinova — where your digital self keeps evolving

Devoxx Belgium @devoxx.com · 12h

Reposted by Richard Groß

❤️🎙️ Devoxx Belgium Speakers on Bsky 👀

0136

Reposted by Richard Groß

Josh W. Comeau @joshwcomeau.com · 03/05/2026

I’ve spoken to a few course creators now, and we’re all seeing the same trend. Revenue down 50%+. Fewer people engaging with our content. People switching to LLMs, which slurp up all of our work and regurgitate it, without consent or compensation. It feels pretty bleak. 😅

711012

Reposted by Richard Groß

Josh W. Comeau @joshwcomeau.com · 03/05/2026

And for #2, it’s absolutely true that AI can be fantastic learning tool, but it can only answer the questions that you know to ask. There’s a vast ocean of knowledge out there, and until you become an expert, it's hard to tell which parts are important (or true, for that matter 😅).

79710

Richard Groß @richargh.de · 13h

You aren’t. You went from sculpting to chiseling. Or from author to editor. Very different jobs. And if you stop working on code, just work on the firehose for the code, then even that metaphor breaks down. Not sure what we are then but it’s definitely not what we signed up for.

030

Also how do they figure their api provides no way to prefil? Just add a couple of „thinking“ blocks into the json messages array.

000

In the past Claude‘s safeguards have been very successfully bypassed with at first simple DAN attacks, then echogram(which I don’t think is actually solved today) up to most recently role confusion. Maybe there is no publicly known attack right now but the first statement is misleading.

100

Also this is deceptive: > Claude’s safeguards blocked the requests that used deceptive prompts. > The Anthropic API provides would-be attackers with no way to prefill Claude’s thinking.

100

Interesting. It looks like Anthropic vibe-coded the report. Did no one review this for clarity before they released it?

100

It’s strange that they are first doing this now. And only a $500 subscription, not a $1000 one.

011

Yes. And stoicism is arguably also the opposite of smartphone usage :)

130

The second one is that dopamine rush after the struggle, when you’ve solved the problem. LLMs instead provide these small dopamine rushes of constant progress. But they also move the baseline. With LLMs you expect the progress all the time and that creates massive dissatisfaction when you struggle.

I had a conversation about this the other day. So I think it’s two things that create happiness. One is the flow state, which is working right at the edge of your abilities and a bit beyond. Aah what a rush. LLMs do not give that.

Simon Willison @simonwillison.net · 28/09/2026

Reposted by Richard Groß

I've published detailed notes and an annotated transcript to accompany the video of the keynote I gave at @wearedevelopers.bsky.social World Congress North America in San Jose on Friday - here's my rundown of everything that's happened with LLMs in 2026 so far simonwillison.net/2026/Sep/27/...

simonwillison.net

2026 in LLMs (so far)

On Friday I gave the closing keynote at the WeAreDevelopers World Congress North America in San Jose. I tied together the key trends from the past year into a chronological …

310522

Richard Groß @richargh.de · 28/09/2026

💯 Unfortunately multi-tasking is a lie.

000

Richard Groß @richargh.de · 28/09/2026

> early evidence suggests prompting can feel addictive, heavy reliance on AI can undermine learning and running agents in parallel can lead to a state cognitive overload dubbed 'AI brainfry'. In line with this, nearly 1/2 of respondents felt AI did not have a positive impact on their mental state.

180

Yes. The outrage about this is bonkers bsky.app/profile/rich...

Probably. Honestly my Ai video detector, my brain, is crap.

000

Still mighty impressive to have that many on standby.

000

Where're All The AI Chips?

About those data centers > Microsoft Only Has 2GW of AI Specific Data Center Capacity, Despite Claiming It Added A Gigawatt A Quarter Three Quarters straight > Microsoft Is Warehousing An Estimated $50 Billion to $100 Billion In GPUs More here: www.wheresyoured.at/wherere-all-...

wheresyoured.at

If you liked this piece, you should subscribe to my premium newsletter, and you can subscribe on the following links: $70 a year, $18 a quarter, or $7 a month. In return you get a weekly premium news...

000

Dare Obasanjo @carnage4life.bsky.social · 26/09/2026

Where did they get all the trucks from?

200

Reposted by Richard Groß

AI companies: We’ve invented an unreliable guessing machine which sometimes doesn’t follow instructions. If you give it access to important systems it could kill us all. Also AI companies: Deploy this technology everywhere and replace your employees with it, otherwise China wins. IPO next month.

930886

Reposted by Richard Groß

Carl T. Bergstrom @carlbergstrom.com · 26/09/2026

Does my Speak-n-Spell feel pain when I type O-U-C-H? The jury is out. Philosophers and neuroscientists say that's fucking stupid, but I called a guy from the Center for Pretending that Household Electronics Are Sentient and he said there's no way to know for sure.

5843181182

There’s studies before Opus 4.5 where that might’ve been an argument but ever since then it’s basically all the same.

Obviously. The human language is free of ambiguity after all.

000

No, Hey should have just been written in Assembly like the other fascist likes suggesting.

100

I’ll ask Claude first bsky.app/profile/migu...

000

Ah the yes man strikes again. The perfect consultant.

100

Of all the things DHH could have done to trigger the Ruby/Rails community…Hey being rewritten in Rust is the most ridiculous thing. I mean, seriously. That is the problem? Are you sure that’s what you want to focus on?

151

LLMs by themselves, without a harness and cross-harness coordination, are still mindless mimics. Although I’d like to have a third word in there. Not sure which one though.

Whether those directions are followed is as dubious as predicting the weather.

I think I am settling on a new mental model/name for running coding bots to produce code: groundhog thunderstorm. A) no memory or learning whatsoever. If it’s not in the context it’s gone. B) away with anthropomorphism, weather metaphors it is. It’s a storm or several you try to give directions.

Miguel de Icaza @migueldeicaza.bsky.social · 24/09/2026

Reposted by Richard Groß

From twitter bad_pote

511730

Reposted by Richard Groß

James @43081j.com · 24/09/2026

Today we're launching ai-policy.dev! A collaboration of me, @danielroe.dev and @matteogabriele.npmx.social - this provides a few human curated AI contribution policies you can choose from. Our hope is that this will help you better define your stance and what kind of contributions you accept 🙏

ai-policy.dev

AI Contribution Policies

Ready-to-use AI contribution policies for open source projects.

1313540

Richard Groß @richargh.de · 24/09/2026

Pi.dev to the rescue. !! ls the bash output is for your eyes only.

pi.dev

Pi Coding Agent

A terminal-based coding agent

Richard Groß @richargh.de · 23/09/2026

Exciting. I wonder if TDD with agents still work as a design feedback and if you still do small steps. Exciting episode.

Richard Groß @richargh.de · 23/09/2026

Aah cool. Are the cards stored publicly? Honestly I think I know too little about ATProto. If I sign in with my account, does every card and every edit show up in my personal stream like a post?

Ah cool. My blog uses Asciidoc though (so much better tables 😁), so I’m not sure how easy that’ll be. I am considering migrating everything to Markdown so I can better c/p articles I wrote for magazines and maybe also get my blog visible in Obsidian.

But for my personal blog: really looking forward to bleeding edge features, buildless and converting to Mastro (from Jekyll).

Update-cycles are interesting. I always try to think of the PS4 (WebKit) or PSP browser and if I need to optimize for those. But most such uncommon browser are often Chromium or WebKit-based, so Firefox really is the browser most likely to auto-update. Unless it’s a company with Firefox LTS :)

That’s neat. Very cool idea, especially that you get semantic names. That uses container style queries right, which are only baseline since May 2026?

Richard Groß @richargh.de · 21/09/2026

Very cool. I am Sandwiching your talk. I‘ll talk about destructoring during the lunch break and securing coding agents right after. :)

Richard Groß @richargh.de · 20/09/2026

Your analogy reminds me of a flowery comic long past. I also love that Dungeon Crawler Carl has had such a strong effect on you. Now I also have to finish that series.

A single-page comic where one person is planting flowers and the other asks why he is so optimistic about the future brining flowers. Because she is planting flowers.