Praveen C

TIFR-CAM, Bangalore

[Logic & Applications @LogicPractice]

Navier-Stokes lost in translation: Why Lean verification of AI autoformalisation does not guarantee correct natural language proofs

https://t.co/hvPQaoyej1

[Bill Gurley @bgurley]

"Don't think for a second that because you are an 'alarmist' that you are doing a social good." Amen.

[Simon Maechling @simonmaechling]

I’m a scientist. I need to say this because the AI hype is getting ridiculous.

AI can design a molecule in seconds.
That doesn’t mean it discovered a drug.

It discovered something we scientists have never been short of:

Something to test.

Someone still has to make it.
Run the experiment.
Measure whether it works.
Check whether it’s toxic.

And ultimately prove it works in the real world.

AI hype tells us:
“Prediction is discovery.”
“Simulation is experimentation.”
“Generating a molecule is developing a drug.”

It isn’t. AI is making ideas incredibly cheap.

But every new idea creates something AI cannot generate: Evidence.

And the more hypotheses AI produces, the more experiments we’re going to need.

That’s the irony nobody seems to be talking about.

AI may not make laboratories obsolete.

It may make them more valuable than ever.

You can speedrun the thinking.
You can’t speedrun reality.

[Alvaro Lozano-Robledo @mathandcobb]

This article is lacking. The journalists should have actually investigated how "AI" found these curves. As it turns out and as we now know, they just used a lot of computing power to push the known construction and method of Elkies and Klagsbrun.

https://t.co/7oOlxOPYCJ

[Le Parisien @le_Parisien]

« Plus vous mettez de l’effort à surmonter le problème, plus vous êtes heureux » : Cédric Villani appelle les doctorants à ne pas utiliser l’IA générative
➡️ https://t.co/pje5J2WIer

[Arnab Ray @greatbong]

Started seeing this IISC professors viral video where he starts by saying “did you see MIT study of agents where they started to cheat” and then says that is AGI.

I cannot ,for the life of me, believe that a professor in class says this and it is viral and then I understand why Dhruv Rathee is Indias biggest AI expert.

[BURKOV @burkov]

The entire conversation about the dangers of building superintelligence (not a hypothetical one, but the one that will soon emerge from LLM scaling) is so ridiculous to me that I really wish I didn't know that this conversation is happening and people with some reputation in tech and science are seriously participating in it.

There's not a single indication that LLMs can be significantly improved beyond cases with verifiable rewards, which currently include only two: math derivations and coding.

Maybe math and coding are what *you* do most of the time at work, but it's not even one percent of what *people* do for a living!

How can so many educated people be so narrow-minded?! It blows my mind.

[Steven Strogatz @stevenstrogatz]

24 Fields Medalists speak out against "A Severe Misalignment of AI in Mathematics"

https://t.co/rpcCm3bh6G

[Praveen C @cfdlab]

@getjonwithit Is there a place where we can see what these new DG schemes look like ? Is it written up somewhere ?

[Pete Firman @petefirman]

I can make you see colours that don’t exist!

[Praveen C @cfdlab]

@chenna1985 If det(Jac) is everywhere negative it is not a problem. This happens if you dont use right-handed system. Problem is when det(Jac) changes sign, then it is zero somewhere and the inverse doesnt exist everywhere.

[Chapel Language @ChapelLanguage]

Today, we’re proud to announce the release of Chapel 2.8! Its highlights include improvements to Chapel’s tools ecosystem, as well as some nice advances in portability and performance.

For details, please see: https://t.co/6i5tVPnV8q

[Physics In History @PhysInHistory]

In the 1940s, Subrahmanyan Chandrasekhar was committed to his teaching role at the University of Chicago, despite being based at the Yerkes Observatory. Each week, he traveled 80 miles to teach a special course attended by only two students. The students were Tsung-Dao Lee and Chen-Ning Yang. They proved their mentor's faith was well-placed when they both won the Nobel Prize in Physics in 1957, years before Chandrasekhar received the same honor in 1983. Remarkably, this course went down in history as the only one where every attendee received a Nobel Prize, underscoring the extraordinary impact of Chandrasekhar's dedication and teaching.

📷 AIP Emilio Segrè Visual Archives, Physics Today Collection

[Praveen C @cfdlab]

@HiroNishikawa If there is a subroutine that returns the exact solution as a function of (x,t), then one can use AD to compute the source term, which gives source code for the source term. For fortran AD, try Tapenade from INRIA.

[Praveen C @cfdlab]

@HiroNishikawa Some such efforts exist, e.g, the HIOCFD workshops.

[Praveen C @cfdlab]

@HiroNishikawa If it is to test numerical fluxes alone, then this can be done. But if you want to test all aspects of methods like reconstruction method, it is not easy to do. What can also be useful is to develop a collection of testcases, quantities to compute, provide grids, etc.

[Hedgie @HedgieMarkets]

🦔 A neuroscientist who testified before the Senate says US schools weren't broken until tech companies convinced them they were. Jared Cooney Horvath found that test scores in Utah started declining right when schools implemented mandatory digital infrastructure in 2014. The US has spent $30 billion putting laptops and tablets in classrooms since 2002. According to international data, more time students spend on computers correlates with worse scores, not better. Gen Z is the first generation to score lower than their parents on standardized assessments.

Now the same cycle is repeating with AI. A Pew survey found more than half of US teens use AI for schoolwork. Teachers report students can't reason, think, or solve problems independently. Horvath argues that tools experts use to make their lives easier are not the tools students should use to learn how to become experts.

My Take
The "transfer problem" goes back to the 1950s. Students learn to master the tool but not the subject matter. Pressey and Skinner ran into this with teaching machines 70 years ago, and we're running into it again with AI. The tech changes but the outcome doesn't.

I think Horvath has it right. Learning requires friction. You have to struggle with a problem to actually understand it. AI removes that friction, which feels like help but functions as dependency. An expert can use AI effectively because they already know enough to evaluate the output. A student using AI to skip the hard part never builds that foundation. We're watching an entire generation learn to operate tools instead of developing the skills the tools are supposed to augment. The productivity gains go to the platforms, not to the kids.

Hedgie🤗

[Jonathan Haidt @JonHaidt]

Many college professors are discovering that students learn less when they have laptops open. Many of us are banning their use in class.

Putting computers and tablets on students desks in K-12 may turn out to be among the costliest mistakes in the history of education

[Cosmos Archive @cosmosarcive]

Sir Roger Penrose argues that current AI is a misnomer because it lacks the one ingredient essential for true intelligence: Understanding.

By applying the lens of mathematical logic, he explains that while machines are masters of following rules, the ability to understand why those rules work is a non-computational process. To Penrose, the leap from calculation to comprehension requires more than just better code; it requires Consciousness.

At the fundamental level, intelligence isn't just data processing; it's the conscious awareness that gives meaning to the math.

Credit: TheInstituteOfArtAndIdeas

[Praveen C @cfdlab]

@HiroNishikawa Paraphrasing what somebody else said. Anybody who has an area of expertise will quickly realize that these AI tools are pretty useless. What this shows is that most people dont have an area of expertise.