Dinesh Ayyappan

"this thing is very good at a narrow subset of tasks it has optimized to be good at, often fails catastrophically at things outside its remit, and is always confident that it's right" feels pretty PhD-level intellect to me!

If it’s a slow motion train wreck, then does Gen AI delay or slow down the wreck at all? If teacher workloads are actually being reduced by outsourcing some of their job to GenAI, then maybe we will leave the profession a little more slowly. Or maybe we will just get more, other work to do. 🤷🏽‍♂️

Now is a fine time to think about what makes two artificially intelligent systems the same or different. Really liked Andrea Ferrario’s paper on this from #FAccT25: arxiv.org/abs/2506.03233

Thanks for sharing! After this explanation, what do your audiences want to clarify or learn next? Any memorable questions or trends?

The public meltdown happening between the world’s richest man 🙄 and the President of the United States 🙄🙄 once again begs the age-old question: are men simply too emotional to be in positions of power?

… like you mentioned in your limitations. Lots to ponder! Thanks for sharing ☺️

Really fresh perspective, Peter! Helpful vocabulary and theory for having this (important) conversation about embracing Gen AI ‘flaws’. Thinking about applications to teaching and also how this could be used to add friction that disrupts Type-1 thinking in contexts when Type-2 is preferred…

Need more communication like this! Many people don’t know where their tax dollars go.

“I can’t let you do that, Dave” but as a feature

An aerial panoramic photo of Barcelona in the evening.

If you're considering a CS PhD but concerned about pursuing it in US, I'm happy to chat about EU options. I just did this and am starting a #PhD in Spain this year on fairness and interpretability of LMs.
DM me and ask about applications, funding, programs, etc. #academicsky #computerscience

Sorry to miss you at CHI! Just emailed you :)

Nice experiment! Would love to chat, especially about education-related contexts!

Going to my first CHI! Happy to chat tomorrow or Sunday with anyone building for learners or with interpretability folks! #chi2025

As people's lives are increasingly being influenced by language models (with or without their consent), AI safety is more important now than ever. I'm sad that some of this research is losing support in the US, but I am also relieved to work in an ecosystem where trustworthy AI is a top priority.

Me wearing a sweater walking alongside a beach with a small child on my shoulders.

Super excited to share that this Fall, I will be moving to Barcelona to start a PhD at Pompeu Fabra researching language model safety and interpretability around questions of bias and discrimination. I've been working on my balancing act: deep learning by day, beachside shoulder rides by sunset 😅

• Al Overview The idiom "you can't lick a badger twice" means you can't trick or deceive someone a second time after they've been tricked once. It's a warning that if someone has already been deceived, they are unlikely to fall for the same trick again. Here's a more detailed explanation: • Licking: "Licking" in this context means to trick or deceive someone. • Badger: The badger is a wild animal, and the phrase likely originates from the historical sport of badger baiting where dogs were used to harass

Someone on Threads noticed you can type any random sentence into Google, then add “meaning” afterwards, and you’ll get an AI explanation of a famous idiom or phrase you just made up. Here is mine

okay now that you all have gotten a taste of America's rational and predictable new trade policy, everyone who wants to build your new factories here just form an orderly line

This is a really interesting experiment -- thanks for sharing! If I may ask: How did you choose the model to work with (and why one)?

Thanks for researching and writing 🙏🏽. You've made a fan!

Stunning. I would have loved having this on my wall when I was a physics teacher!!

Why the world is looking to ditch US AI models Content moderation systems are being abandoned and defunded, leaving many countries looking for alternatives.

Lots of coverage of the sentiment in this headline, but this article (by @eileenguo.bsky.social) also highlights the value of language-specialized models. So much nuance here ("slang usages and slurs, ... phrases written in a mix of languages) that's hard for people to grasp, let alone LMs.

Spain to impose massive fines for not labelling AI-generated content Spain's government approved a bill on Tuesday imposing massive fines on companies that use content generated by artificial intelligence (AI) without properly labelling it as such, in a bid to curb the use of so-called "deepfakes".

It's great to see Spain moving on the EU AI Act's guidelines on transparency. Looks like these massive fines are only for high-risk systems, but it's a good step towards public safety.

"Upper management does not understand what AI is and thinks it will solve untold problems. Junior devs rely on it too much without understanding their code first."

I imagine something like this is pervasive across roles and industries other than tech / software engineering too.

This is so amazing for science!

Thanks for sharing! This is a really helpful framework for understanding hallucinations. I am so curious about why there are so many Type A errors and so few Type B errors. It seems like both incorrect facts (Type B) AND correct facts (Type A) are having a hard time coming through! (is that right?)

Hello from Singapore! DM’d you

Hello from Singapore!! DM’d you.

I am too tired to write any more blogs this week but I listened to an hour of this and the quick summary of this for anyone interested so you do not have to listen to it is like:

A text excerpt: “… The tools of scientific thinking are powerful and of great value. The point is not to abandon them but to integrate the knowledge they provide into a broader, richer conversation about what we human beings are doing and should be doing, and why. Science is not a guide but a support a crutch, a tool, an accelerator-for human beings ready to think what they are doing.” Page xiii

Here in the forward, Allen uses Arendt’s chief proposal, “to think what we are doing” to frame the role of science: that it should support the conversation about what we should be doing, not lead it.