Hey, I've heard some scandalous stories as the creator of a podcast that interviews women in technology... Some had to be off the record.

"Harvard’s conflict with the federal government is on track to take yet another casualty. Facing financial uncertainty after the endowment tax hike and cuts to research funding, the University drastically reduced Ph.D. admissions to curb spending." www.thecrimson.com/article/2025...

Attributes to Support the Formulation of Practically Relevant Research Problems in Software Engineering arxiv.org/abs/2512.12699

AI Hallucination Cases Database – Damien Charlotin www.damiencharlotin.com/hallucinatio...

Uber, 99 Airbnb Bitcoin ChatGPT, Claude

Claude Codex Antigravity Copilot Replit Lovable Which one is the biggest failure? Which one loses the most money?

The Dangers of Relying on AI for Education Emily M. Bender youtube.com/shorts/IxNrn...

Jacu bird

"Profiting from the Randomness of LLMs in the Requirements Engineering Process" is the title of an invited talk at a software engineering conference

Backing Up Spotify Anna's blog annas-archive.li/blog/backing...

Do you think this kind of research helps companies that create LLMs?

Sad, but it seems everybody is using LLMs to code. Maybe @algebraic.dev can explain the rules on using AI on #LeanLang source code

From Zero to QED: An informal introduction to formality in Lean 4 https://lobste.rs/s/2rtmv9 ##compsci ##formalmethods

Vale pra LLMs também

👇

👀

I have just checked and Fernando Pereira appears to be Portuguese. I would be even sadder if he were Brazilian. Stochastic parrots is a great paper. Censorship does not make sense.

This year on Bluesky I wrote 2,908 posts and 2,893 replies. I received 23,095 likes, whereas 4,903 was from my most popular post, and apparently I love saying "que" and 👇! www.madebyolof.com/bluesky-wrap...

ALT: No, you're not a "prompt engineer", you're a sloperator. Colin M @vonexplaino.com

Team Deakins: KLEBER MENDONÇA FILHO - Writer / Director teamdeakins.libsyn.com/kleber-mendo...

"Trouw reported this week that even the best chatbot (Gemini 3) only get a third of their answers right, and nobody cares. AI grifters who want to sell books continue to babble on LinkedIn about "responsible use" and "better prompting", but you can't prompt something that gives random wrong answers"

"Disney [...] which has been notoriously protective of its intellectual property, opened up the video generator, saw the videos featuring Nazi Spongebob and criminal Pikachu, and decided: We want in." @felienne.bsky.social AI in week 51 www.felienne.nl/2025-51/?ref...

Puzzle: find the non hallucinated reference in this paper lex-localis.org/index.php/Le...

It's only useful when the result is irrelevant

What bothers me is that software engineering researchers are using these LLM systems built from stolen data without worrying about it, even publishing articles that help these companies become more popular. My hope is that I myself didn't care about this a few months ago.

Cowards

Read paper: 1) there's structure: we grant cognition is tractable --> distribution generated by tractable algorithm =/= random noise 2) if you know more structure, specify it, argue for cognitive plausibility, present a tractability proof. I'll wait. (for onlookers: screenshot not the proof)

"Manifesto for Software Development and Modeling in the Age of AI" From the title, I think it is garbage. But I will skim it next year www.linkedin.com/posts/kai-r-...

Should we care about reproducibility when the unethical aspects of LLMs are so strong? This is a real question, not a rhetorical one.

And please also, AGAIN, never forget that they are inherently unethical to use because no one gave permission for their research or writing to be used in it. It is all stolen, and if you're using the plagiarism machine, please keep that in mind.

"Although relational AI has potential therapeutic benefits" Do they? How can they be sure?

Mention to Geoff Hinton in The AI Con

13% of "Why We Fear AI: On the Interpretation…" de Hagen Blix, Ingeborg Glimmer amz.onl/56Etrcp

Unit and property-based testing in #elixirlang

Do we need something like that here in Brazil?

Cheap money makes everything more expensive.

Cars (so often as they're used now) are an evolutionary dead end. Let's just have and vision and start closing the gap. Individual car ownership in cities is especially wasteful.

Saturday mood

I shared this on Instagram

Second hand chatbot exposure!!!

Fernando Areias na Elixir Curitiba 2025 youtube.com/shorts/kn-ZY...

Raleigh, NC December 2014

I forgot to mention here that the ACM DL decision faced backlash here on BlueSky. Another thing I only thought of later: LinkedIn is too positive, and in academia we need criticism, a certain amount of negativity. That's why Mastodon and Bluesky are better.