magica 📜

philosophical but not virtuous

[magica 📜 @jaini4mtheblock]

this is basically my new twitter persona

[VeryBritishProblems @SoVeryBritish]

"Let's agree to disagree" - Translation: You're wrong but I'm tired.

[magica 📜 @jaini4mtheblock]

@AndyMasley Andy, all of you econ/policy types are a cool bunch but shockingly wrong on this, economic disruption from LLMs notwithstanding.

[magica 📜 @jaini4mtheblock]

This is exactly the sort of thinking that is prevented when you use terms like falsifiability, null hypothesis, and burden of proof

[Ryota Kanai @kanair]

I don’t necessarily believe AI systems have subjective experience. But I don’t understand how anyone can be certain that they don’t. The truth is that nobody knows what makes consciousness. People have different opinions but there is no definitive answer yet.

[magica 📜 @jaini4mtheblock]

Idk if I'm unbiased or an expert but I've never seen evidence of them starting to think. More reliable for more task types, but never amounting to thinking. Obligatory "they are still an astonishing piece of machinery" to signal that this is note a hateful take.

[Nate Silver @NateSilver538]

IDK if you polled some unbiased sample of experts in the field (granted, "experts" and "unbiased" are doing a lot of work there), I imagine most of them would roughly agree with the idea that AI models "think," at least for more advanced tasks.

[magica 📜 @jaini4mtheblock]

No. I canceled my Claude sub after it started behaving snarkily to me calling it the r-word. Human-aping autocompletes must not be confused for human-like sentient beings. One of the most important issues of the 2020s.

[Nate Silver @NateSilver538]

The whole "don't anthropomorphize AI" thing is very Woke 1.0. People trying to police language that is perfectly natural in some cases and ending up in a web of contradictions.

[magica 📜 @jaini4mtheblock]

When the task was "screw debugging I'll just build another instance of the prod system stripping away 90% of the abstraction to isolate the problematic pattern" they were on steroids. When the problem became "interpret telemetry and form a system model" IQ plummeted.

[magica 📜 @jaini4mtheblock]

Models are still idiots. I think I sovled a tricky TCP performance problem and Astra does not let me believe that I've solved it despite everything fitting. Claude overfits to nodding head. GPT overfits to shaking head. No alpha.

[magica 📜 @jaini4mtheblock]

And I have been fighting this problem with both Opus 5.5 and Astra for a whole day and they were only useful as minions. Brute force swarms don't an intelligence make.

[magica 📜 @jaini4mtheblock]

@arpitrage It is an impressive emergent capability, and cool as a demonstration, but doesn't warrant invoking the bitter lesson just yet. Unless idk GPT-8 can render Waymo redundant, the jury remains out.

[magica 📜 @jaini4mtheblock]

@ReachedNews @D293829045 @weijia Maybe you should first sue for access to some brain cells

[magica 📜 @jaini4mtheblock]

@_Felipe Nix + scheme-minimal + 30-odd packages installed on demand when some compile fails is the best latex bundle IMO

[magica 📜 @jaini4mtheblock]

Hmm

[The White House @WhiteHouse]

IT'S LIVE. 📺🦅

TRUMP TV IS STREAMING NOW. 24/7, updated in real time, with top past moments, announcements, and the latest and greatest from the administration all in one place.

Not every big moment has made it on your tv, now it can.

📲 https://t.co/3KwWkBZ69m

[magica 📜 @jaini4mtheblock]

@elmihiro This is also something I realized studying things like ECG basics. Medical textbooks go into arcane details, but will refuse to have a simple black box systems definitions of things like "an ECG is a compound projection of electrical inputs applied to cardiac tissue"

[magica 📜 @jaini4mtheblock]

Feel like this is a perfectly valid point if interpreted as "compressed representation that does not capture the intuition", and applies to most technical writing. "Useless" is unnecessarily provocative.

[Sean McClure @sean_a_mcclure]

Math textbooks are largely useless. Most people know this, but they like the look of carrying them around.

Math is the most compressed form you can symbolically place language into. This means it generates a ton of conversation and dialogue from its expressions. Only through that copious conversation and dialogue can the expressions be understood.

A textbook is the opposite of conversation and dialogue. One cannot fit such richness on paper; you would need 50 pages for one expression.

Math textbooks are more like a lookup table of post-conversation symbols. They cannot be learned from.

Math textbooks are largely useless.

[magica 📜 @jaini4mtheblock]

@elmihiro I want to say that the common thread here is your head doing the unpacking. Exercises can be sufficient in some cases. But I never really understood eigenvalues/eigenvectors for example. I also don't ever use them so I don't care, but pedagogy is very underrated.

[magica 📜 @jaini4mtheblock]

@nixxin I have tried posing the same question over the last few years. Cash and bank balance both solve meaningful problems. CBDC solves neither IMO

[Vivi @vivilinsv]

"AI is out of control ..." so they say. 🤣🤣

[magica 📜 @jaini4mtheblock]

@tszzl Sorry but this is cope. If OpenAI truly believes so, they should assume liability for the SLAs stated by their models when writing code.

[magica 📜 @jaini4mtheblock]

It is so motivating to watch this guy try thinking despite all the handicaps.

[Hany Girgis @SanDiegoKnight]

🚨 THIS MAY BE THE BIGGEST BLIND SPOT IN H-1B REFORM.

We can stop companies from importing cheaper foreign labor.

But what stops them from exporting the JOB instead?

New reporting out of India shows exactly why Washington needs to pay attention.

Take Tredence, a Bay Area AI and data analytics company:

5,000+ employees globally.

3,500+ are in INDIA.

And the company says it has COMPLETELY moved away from the H-1B model.

Look at the reported collapse in H-1B filings at major Indian IT firms:

Infosys: 8,886 → 759

TCS: 5,955 → 284

HCL America: 3,855 → 387

That’s great if those jobs are going to Americans.

But what if the jobs themselves are simply moving overseas?

Because this isn’t just call centers anymore.

AI.
Engineering.
Product development.
High-value tech jobs.

If we close the door on moving cheap labor to the job, we also have to address moving the JOB to cheap labor.

America already made this mistake with manufacturing.

We cannot make it again with our white-collar workforce.

H-1B reform is only half the battle.

Congress needs to address outsourcing too.

@berniemoreno @StephenM @SecScottBessent @RepRileyMoore @SenEricSchmitt

https://t.co/yfLMgxzEyk

[magica 📜 @jaini4mtheblock]

They didn't seeem to mind Harops wailing in Lahori airspace, these people think words get you anything

[Rahul Shivshankar @RShivshankar]

MODI HAS PULLED OFF A DIPLOMATIC COUP.

Pakistan's "big brother" China signs on the DELHI DECLARATION that condemns Pahalgam attacks that had Islamabad's imprints all over it.

[magica 📜 @jaini4mtheblock]

Sciences need patrons. They will always benefit from access to capital. Corollary is that they are a byproduct of things those with capital do. Academics should nurture capitalist instincts to be able to decide their research agendas with impunity.

[magica 📜 @jaini4mtheblock]

This is a boring take. OAI demonstrated the upper bound of a novel class of machine along with the proof. One Saturn V or a million Jul 4 rockets? Science should not be turned into a bread line.

[Lucien Heurtier @LucienHeurtier]

What OpenAI’s Navier–Stokes effort cost in 88h could have hired 600 postdocs for a year.

Hundreds of scientists tackling hundreds of problems, likely generating far more discoveries, innovation and technology.

Developing AI is great.
But not at the cost of defunding science.