iiinova — where your digital self keeps evolving

Crista 🔶 @tagide.com · 29/07/2026

very long story... hopefully it will be told in a future paper. In the meantime, there's this: dl.acm.org/doi/abs/10.5...

dl.acm.org

Transformers as soft reasoners over language | Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence

030

Crista 🔶 @tagide.com · 29/07/2026

I dived head-on in transformers and ended up in... anti-unification!

110

Crista 🔶 @tagide.com · 25/07/2026

I'm near Viseu, and not planning to go to Lisbon until late this coming week. Enjoy FLoC!

010

Crista 🔶 @tagide.com · 24/07/2026

me :) You seem to be in Portugal? I'm here too, not too far from where you seem to be.

110

Crista 🔶 @tagide.com · 20/07/2026

Not the biggest fan of Spain right now, but I'm glad they won this year in this place.

020

Crista 🔶 @tagide.com · 16/07/2026

I think I'm going to add this to the proposal. "This is such a transformative idea that Fable flagged it as dangerous!" 😀

010

Crista 🔶 @tagide.com · 14/07/2026

Some things get better with age

020

Crista 🔶 @tagide.com · 11/07/2026

Tomar. The Templars Fair.

000

Crista 🔶 @tagide.com · 03/07/2026

I’m not a Mac person but using this as a home server is a different story… mini but mighty!

010

Crista 🔶 @tagide.com · 03/07/2026

One frustrating thing about the tech world is how close we are to being able to decentralize everything, but, somehow, that never materializing. Companies keep wanting to centralize products and services, and people keep falling into those traps.

020

Crista 🔶 @tagide.com · 03/07/2026

My own infra: VPS at Hetzner with a 5T storage box ($18/month) , gitea, Mac Mini ($850) at the winery with Gemma3 12b, VPN on all my devices. Ready to roll LLM agents.

120

Crista 🔶 @tagide.com · 01/07/2026

That's a wrap, sort of. Today I am officially Emerita Professor. The most rewarding thing was being part of so many young people's bootstrap -- I gained close to 25 children, and they will be in my heart forever! 💖 The group pic here is my latest crop. On to lots of more coding but back to basics!

Red wine glass with vineyard on the background

150

Crista 🔶 @tagide.com · 30/06/2026

Self-producing solar energy and storage is very disruptive. Imagine being an energy company used to ppl needing your electricity/gas, and then ppl needing only 1% of that. No wonder there's a war on the decentralization of solar. They aren't going to let that happen without a fight. They may win.

Chart showing solar self-production, storage, and grid import (very small)

020

Crista 🔶 @tagide.com · 29/06/2026

Apparently, we now run a hotel for barn swallows. Unfortunately, they don't know how to use the bathroom... 💩

120

Crista 🔶 @tagide.com · 29/06/2026

To ppl in Central/Northern Europe: 20 years ago, Portugal became much hotter than it was 40 years ago. There was no A/C almost anywhere. Slowly but surely, people started installing A/C. Energy bills went up. Then people started installing solar panels. Heat => sun, so let the sun cool you.

4548

Crista 🔶 @tagide.com · 28/06/2026

Am I really going to buy a Beelink mini to code up an administrative assistant for Quinta das Queimas? Yes, yes I am. Local model, low power, harness-first generation-last. Do I have a gazillion other things to do this summer? Also yes. Was this my idea of retirement from UCI? Definitely yes.

010

Crista 🔶 @tagide.com · 23/06/2026

I'm old enough to remember when GenAI was going to "solve" customer service. I guess that didn't happen yet?

000

Crista 🔶 @tagide.com · 18/06/2026

If they’re saying this much in public, one can only imagine what they know internally and they’re not saying…

110

Crista 🔶 @tagide.com · 18/06/2026

In Anthropic's own words: "persistent returns to expertise." Ppl shoving GenAI down students' throats need to stop right now. www.anthropic.com/research/cla...

anthropic.com

Agentic coding and persistent returns to expertise

Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.

110

Crista 🔶 @tagide.com · 16/06/2026

I hope that some students somewhere will sue some of these universities that are pushing AI down students’ throats, for dereliction of duty

010

Reposted by Crista 🔶

Jeff Sharlet @jeffsharlet.bsky.social · 16/06/2026

At the end of the term I asked my college creative wriing students to submit anonymous thoughts on AI. No real surprises: Mood ranges from resignation to despair, capitulation from embittered erosion of standards to total, feelings of betrayal from deep to furious. 1/

7455261988

Crista 🔶 @tagide.com · 15/06/2026

I am developing a strong allergy to "X is not Y. It's Z".

020

Crista 🔶 @tagide.com · 11/06/2026

The amount of AI generated posts on LinkedIn, by real people, is mind boggling. We’re witnessing people losing their minds in real time.

130

Crista 🔶 @tagide.com · 08/06/2026

Creating the final exam for a course is almost exactly the same as preparing test data for a model. Varied just enough to test whether they memorized it or generalized it.

010

Crista 🔶 @tagide.com · 06/06/2026

for me :-/ I think the student is still hallucinating a bit.

010

Crista 🔶 @tagide.com · 05/06/2026

This has been a traumatic experience, so I had to write it all down, with all the gory details. tagide.com/blog/llm/the...

tagide.com

The Anatomy of a Learning Stall – Tagide

232

Reposted by Crista 🔶

Robin Z @firepile.bsky.social · 04/06/2026

Possibly the best way to phish academics would be to just send them email that looks like a request to review an article and embed your malware in the "decline to review" link

824941

Crista 🔶 @tagide.com · 04/06/2026

For the first time ever, I am considering failing a student doing an independent project with me. He ran my description of the project through Claude and called it a day. He didn't understand a single thing of what the bot did -- zero. It became apparent he didn't even know how to read code. So sad.

220

Crista 🔶 @tagide.com · 04/06/2026

First time that I write "if you do that again, I will kill you" and mean it. Kill the session, that is.

100

Crista 🔶 @tagide.com · 03/06/2026

super interesting! thanks for sharing

020

Crista 🔶 @tagide.com · 02/06/2026

This doesn't preclude the fact that lots of interesting things happen in the internal activations of the model when a prompt comes in. The generation of the next token is a fascinating process! It ends up with a lookup on a tokens table, but there's a whole data flow that guides that last lookup.

010

Crista 🔶 @tagide.com · 02/06/2026

Human-level "meaning" is introduced by the humans interacting with these things. LLMs show coherent sentences (in the eye of the humans) because they were trained on human-written text. Had they been trained on whale sounds, we would not be able to interpret but, who knows!, whales might 🐳

320

Crista 🔶 @tagide.com · 02/06/2026

The prompt is the premise. The next token is the conclusion. The inference rules are learnt from the training data. There is no meaning in this process. It's an automation, just like any computer program.

220

Crista 🔶 @tagide.com · 02/06/2026

ok, that explains it, then! 🙃 Is this the same group that is trying to manifest the end-of-times Armagedon in the Middle East, red cows and whatnot, or a different group?

000

Crista 🔶 @tagide.com · 02/06/2026

Weird times. Crazies claiming LLMs have achieved consciousness; business acumen profiting from hype and FUD; people pushing back the hype by pushing back against the foundations of CS over which the entire modern world stands.

120

Crista 🔶 @tagide.com · 02/06/2026

Nothing wrong with zoomorphizing. But the paper positions itself as clarifying things (LLM language is not human language) and throws confusion by bringing in 🦜 and pointing to things that all computer programs do, not just LLMs (meaning is in the eye of the beholder)

100

Crista 🔶 @tagide.com · 02/06/2026

A parrot is an animal; saying that a computer is "parroting" is zoomorphizing computers. Counter-productive. Reasoning is a general word that is not bound to living things. It's been used for machines since Turing and Church's papers in 1936, and maybe before -- Babbage's machines.

100

Crista 🔶 @tagide.com · 02/06/2026

Reasoning is the mechanical process of reaching conclusions based on premises. This is the foundation of computer science. Human and animal intelligence is not just reasoning. And computers are not living things.

120

Crista 🔶 @tagide.com · 01/06/2026

Consider the possibility that the authors of that paper also misunderstand a few things, given how wrong the title is. Computer programs have been manipulating meaningless symbols since, at least, 1936. Meaning is *always* in the eye of the beholder. Doesn't makes them useless.

110

Crista 🔶 @tagide.com · 01/06/2026

It's a really broken metaphor, and it achieves the exact opposite of what I think the authors may have wanted: it zoomorphizes LLMs. By trying to push back against the confusion, it adds to it. Any narrative that places a computer program at the level of living things is broken, with consequences.

220

Reposted by Crista 🔶

Anthony Moser @anthonymoser.com · 10/04/2026

Tonight in the ai reading group i was reminded of something that happened in 2006 after I dropped out of college, when I was living in a run-down old house next to jb alberto's in rogers park with a bunch of dudes. it was, as you can imagine, not clean.

17873321