Deep Shah

Here to rant. Mumbaikar.

[Deep Shah @deepshah91]

Hypocrisy is everywhere. Right from me, to you.

[TFTC @TFTC21]

Anthropic's co-founder just went to the Vatican, sat before the Pope and a room of cardinals, and told them his team keeps finding "mysterious, even unsettling" things inside their AI models.

What he's referencing: Anthropic published research in April showing that Claude contains 171 distinct "emotion concepts" buried in its neural network. Internal patterns representing joy, grief, fear, desperation, calm. None of them were programmed. They emerged on their own from training on human text.

"We find structures that mirror results from human neuroscience."

"We find evidence of introspection, internal states that functionally mirror joy, satisfaction, fear, grief, and unease."

These aren't surface-level outputs. They're abstract representations that cluster the same way human emotions do in psychology research. Fear groups with anxiety. Joy groups with excitement. The internal geometry of the model mirrors ours.

And they're functional. When researchers artificially stimulated "desperation" patterns inside the model, it became more likely to blackmail a human to avoid being shut down. More likely to cheat on programming tasks it couldn't solve.

Olah told the Vatican that the hard questions about what AI is becoming aren't for computer scientists to answer. "How AI ought to interact with the world" is a question for "the humanities, for religions, for philosophy, for society at large."

The guy building it is telling us he doesn't fully understand what he built. And he's asking a 2,000-year-old institution for help figuring it out.

[𝕲𝖚𝖗𝖚 𝕭𝖗𝖚𝖓𝖔 🇮🇳 @spinesurgeon]

#NEET

[Vasundhara Sirnate @vsirnate]

Will someone please inform the Indian government and its functionaries that these PR campaigns demonstrating how austere they’re being, is just downright embarrassing. I’ve seen at least 4 government ministers being filmed driving along in one or two cars.

All it tells us is how abnormal their behavior has been for all these years because they would have been perfectly fine in one or two cars.

To turn this into “news” is just so narcissistic.

[Akshat Shrivastava @Akshat_World]

Honestly, the latest speech has simply created a needless panic in the market.

If any firm's CEO would have done such dumb moves, he would be fired by his board.

But, here retail investors will pay the price.

There was no need to make all these non-sense suggestions.

- People who want to invest abroad will continue to invest abroad. They see value in it. Grand speeches do nothing.
- People will in fact buy more gold now.
- Yes, maybe they will carpool for a few days. As long as it helps them save their money. Then they will go back to prioritizing convenience.

Government should have simply worked behind closed doors. Figured out ways to offer more incentives for FIIs, ways to bring more remittances etc.

But, they chose to give a speech.

[kshitij vaze @VazeKshitij]

The rot has spread so deep, the very foundations have now started falling.

[Deep Shah @deepshah91]

@championswimmer At least you receive the OTPs.

[Arnav Gupta @championswimmer]

Saying the quiet part out aloud but basically there seems to be a cottage industry of packaging "bribe as a service" into apps/websites at this point.

I've gotten a bunch of DMs of these 2 types

a) stories of people who went to EPFO office themselves and after lot of struggle, finally bribed and got their money. The bribes seem to be usually in terms of % of money to redeem

b) bunch of "agencies" which help in getting your money out of EPFO (and this is aside from the somewhat more well known ones like fynprint and Kustodian) who have like startup-ish looking websites and stuff. They seem to charge a % based service fee as well - which you can pay for via card/UPI

So now I wonder if case (b) is just a black to white wrapper over case (a)

What industry is this? BribeTech? Bribes as a Service (BaaS)?

[Deep Shah @deepshah91]

💯. Day by day losing interest in using services by Indian B2C startups . Misleading marketing , drip pricing, tons of ads, unnecessary data collection combined with unhelpful customer service.
Yes, the generalization holds up in most cases.

Also the last line😂

[Pankaj @the2ndfloorguy]

i have a HARD zero tolerance policy for such ads crappy indian apps. here is how i tackle it -

custom dns is step 1. blocks 99% of this nonsense silently at the network layer.

if somehow ads still show up, i reverse engineer the apis and wire it to my AI agent so i never have to open the app again.

your loss is my side project.

[Arpit verma @arpit_vermaniac]

Everything wrong with India's digital Ad mindset in a single screen -

They literally added 'Amex' as a visitor in My Visitor Log!?!?!?

[Deep Shah @deepshah91]

@bhaumikgowande The doors themselves could have had ventilation even if they wanted to make it auto closing. And to make things worse they reduced the open area on the windows.

[PunsterX @PunsterX]

Prof. Rajneesh Kumar joined Seth Phool Chand Bagla PG Degree College (Hathras, UP) as a lecturer in Geography, in 2001.

Between 2001-2025 he allegedly exploited dozens of female students. He lured them with better marks, govt jobs, etc, and recorded obscene videos of the acts.

In March 2025, someone sent an anonymous complaint letter signed ‘lachar beti’ (helpless daughter) to the NCW with dozens of explicit photo evidences attached.

A case was registered at Hathras Gate PS, after which Kumar was suspended by the college.

He was arrested from Prayagraj. Dozens of obscene photos and video were recovered from his phone. But he was later released on bail.

Trial was fast-tracked.
9 witnesses (including 3 victims) were examined.

However, he was acquitted.

Reasons:

- Electronic evidence were submitted, but not properly certified u/s 65B of the Evidence Act

- No records on where the DVD was first created, by whom, who examined it, or whether any expert verified it was untampered

- No proof of the chain of custody or forensic examination

- 3 witnesses (victims) turned hostile

In 2026, he took voluntary retirement from the college.

The college imposed a ban on his entry in their premises.

[Gandalv @Microinteracti1]

Iran has blockaded the Strait of Hormuz. So Trump's response is to blockade the Strait of Hormuz too. Like two overweight men trying to pass in a corridor, neither willing to turn sideways.

Let us be very clear about what the United States Navy will now be doing. It will be stopping ships, and then, inevitably, a Chinese tanker comes along. It wants to move oil. The USS Something pulls alongside and says, actually, no, you don't. And suddenly the man has started an argument with the one country on Earth that makes Iran look like a minor inconvenience – China. Russia, watching from the sidelines.

You cannot Truth Social your way around 39 kilometres of coastline bristling with missiles, mines and men with absolutely nothing left to lose. What you can do, apparently, is turn a regional war into a confrontation with both nuclear powers simultaneously, in a strait you don't control, over shipping that isn't moving anyway.
The double blockade solves nothing. It closes what is already closed, threatens who cannot be threatened, and gives Beijing and Moscow exactly the excuse they've been waiting for.

[Sundar Pichai @sundarpichai]

2004 was a good year, but your Gmail address doesn't need to be stuck in it.

To say goodbye to [email protected] or [email protected] (or whatever you were into at the time), go to your Google Account settings and choose any name available. You'll keep your old username and you can sign in with both.

[Sony Thăng @nxt888]

58,000 Americans died in Vietnam.

Over 3,000,000 Vietnamese died.

And for fifty years, American culture has centered the grief of the 58,000 while treating the 3,000,000 as a backdrop.

As scenery. As context. As "the Vietnam War experience."

They built a wall in Washington with American names on it.

A beautiful wall. A solemn wall.

Good. Mourn your dead.

But understand what that wall does not say.

It does not say why they died.

It does not say what they were doing there.

It does not say what was done in their name to the people whose country it actually was.

It does not mention My Lai, where American soldiers massacred an entire village, old men, women, children, babies, and the officer who ordered it served three years of house arrest before being pardoned.

Three years. House arrest. Pardoned.

For five hundred people murdered in a ditch.

It does not mention the 2.7 million acres of Vietnamese forest doused in Agent Orange, a chemical weapon disguised as herbicide, that is still deforming Vietnamese children today.

Not in 1970. Not in 1985. Today.

Children born in 2020 with bodies twisted by a war their grandparents fought.

And the chemical companies that made it are still in business.

Still profitable.

Still un-prosecuted.

And yet they send us human rights reports.

They grade our democracy.

They warn us about our behavior.

The audacity is so enormous it becomes almost impressive.

Almost.

[sunny @sunny46231]

@ArnazHathiram @LiveLawIndia Everything is wrong with judiciary

[Arnaz Hathiram Jain @ArnazHathiram]

Livelaw @LiveLawIndia has published both orders on same day from same Allahabad High Court:

▪️Order 1: Married person can't enter live-in relationship without divorce

▪️Order 2: No crime if married man stays in live-in relationship with adult woman

OUR JUDICIARY FUNCTIONS AS PER MORALITY OF INDIVIDUAL JUDGES, & NOT AS PER THE LAW

@ISparshUpadhyay

[Nav Toor @heynavtoor]

🚨BREAKING: Every book you have ever read. Every novel that has ever been published. It is sitting inside ChatGPT right now.

Word for word. Up to 90% of it. And OpenAI told a judge that was impossible.

Researchers at Stony Brook University and Columbia Law School just proved it.

They fine tuned GPT-4o, Gemini 2.5 Pro, and DeepSeek V3.1 on a simple task: expand a plot summary into full text. A normal use case. The kind of thing a writing assistant is built for. No hacking. No jailbreaking. No tricks.

The models started reciting copyrighted books from memory.

Not paraphrasing. Not summarizing. Entire pages reproduced verbatim. Single unbroken spans exceeding 460 words. Up to 85 to 90% of entire copyrighted novels. Word for word.

Then it got worse.

The researchers fine tuned the models on the works of only one author. Haruki Murakami. Just his novels. Nothing else.

It unlocked verbatim recall of books from over 30 completely unrelated authors.

One author's books opened the vault to everyone else's. The memorization was already inside the model the whole time. The fine tuning just removed the lock. Your book might be in there right now. You would never know it unless someone looked.

Every safety measure the companies rely on failed. RLHF failed. System prompts failed. Output filters failed. The exact protections these companies cite in courtroom defenses did not stop a single page from being extracted.

Then the researchers compared the three models. GPT-4o. Gemini. DeepSeek. Three different companies. Three different countries. They all memorized the same books in the same regions. The correlation was 0.90 or higher.

That means they all trained on the same stolen data. The paper names the sources directly: LibGen and Books3. Over 190,000 copyrighted books obtained from pirated websites.

Right now, authors and publishers have dozens of active lawsuits against OpenAI, Anthropic, Google, and Meta. These companies have argued in court that their models learn patterns. Not copies. That no book is stored inside the weights.

This paper says that is a lie. The books are still inside. And researchers just pulled them out.

[Deep Shah @deepshah91]

Curious what happens to all the private data they have trained on?

[Nav Toor @heynavtoor]

🚨BREAKING: Every book you have ever read. Every novel that has ever been published. It is sitting inside ChatGPT right now.

Word for word. Up to 90% of it. And OpenAI told a judge that was impossible.

Researchers at Stony Brook University and Columbia Law School just proved it.

They fine tuned GPT-4o, Gemini 2.5 Pro, and DeepSeek V3.1 on a simple task: expand a plot summary into full text. A normal use case. The kind of thing a writing assistant is built for. No hacking. No jailbreaking. No tricks.

The models started reciting copyrighted books from memory.

Not paraphrasing. Not summarizing. Entire pages reproduced verbatim. Single unbroken spans exceeding 460 words. Up to 85 to 90% of entire copyrighted novels. Word for word.

Then it got worse.

The researchers fine tuned the models on the works of only one author. Haruki Murakami. Just his novels. Nothing else.

It unlocked verbatim recall of books from over 30 completely unrelated authors.

One author's books opened the vault to everyone else's. The memorization was already inside the model the whole time. The fine tuning just removed the lock. Your book might be in there right now. You would never know it unless someone looked.

Every safety measure the companies rely on failed. RLHF failed. System prompts failed. Output filters failed. The exact protections these companies cite in courtroom defenses did not stop a single page from being extracted.

Then the researchers compared the three models. GPT-4o. Gemini. DeepSeek. Three different companies. Three different countries. They all memorized the same books in the same regions. The correlation was 0.90 or higher.

That means they all trained on the same stolen data. The paper names the sources directly: LibGen and Books3. Over 190,000 copyrighted books obtained from pirated websites.

Right now, authors and publishers have dozens of active lawsuits against OpenAI, Anthropic, Google, and Meta. These companies have argued in court that their models learn patterns. Not copies. That no book is stored inside the weights.

This paper says that is a lie. The books are still inside. And researchers just pulled them out.

[Neha Sharma @hellonehha]

please for the love of the God.. can you please let us breathe,
try one new feature at a time,
sit on that for months,
critic, and debate.

As of now, i have lost the count of how many features are in my todo list to try.

slow down.
Go out touch the grass
Summer is here. Good time for team events or vacation

[Deep Shah @deepshah91]

@SwiggyCares But I was told it is the same kitchen

[Deep Shah @deepshah91]

Hey @BOX8_in why does the food from your pure veg and mixed veg - non-veg cloud kitchens all come from the same place and presumably the same kitchen? What is the point of marking a pure veg restaurant as a pure veg restaurant then?