• The Breakdown
  • Posts
  • 🟪 Don’t rage against the machines

🟪 Don’t rage against the machines

What today's Hindu holiday can teach us about LLMs

“If you are filled with reverence for everything, life is so full.”
— Gurudev Sri Sri Ravi Shankar

Don’t rage against the machines

India’s Navratri festival is ten days of fasting, music, dance and prayer, all in celebration of the ten-armed Hindu Goddess Durga, who famously defeated the demon king Mahishasura in battle.

Depending on regional tradition, each day of the festival is dedicated to a different subject of devotion. In South India, for example, Day One honors the Earth and foundational wisdom, with devotees ritually planting sprouts in a bed of sacred soil. Day Two honors discipline and penance with strict fasts and quiet prayer.

Day Nine honors tools.

The ninth day of the Navratri festival — today! — is known in South India as “Ayudha Puja”: the “worship of tools and instruments.”

The ancient tradition began with the warrior class honoring the tools of its trade — swords, daggers, bows and arrows — in remembrance of the ten divine weapons Durga wielded (one in each hand!) to defeat Mahishasura.

Today, it’s tools of all trades. Farm equipment, construction machinery, kitchen utensils, office computers — on the ninth day of Navratri, anything that helps people do their jobs is honored with ritual cleaning, decorations, blessings, and a day of rest.

Mechanics wash their tools with the same care a priest washes an idol. Engineers garland their laptops with a necklace of flowers. Machinists bow with folded hands before their lathes and drills.

The ritual is rooted in the traditional Hindu principle that “all aspects of life [are] connected with spirituality and divine grace,” a helpful explainer says. “Ayudha Puja is a testament to the profound sanctity found in the everyday, urging us to approach our daily tasks with reverence and gratitude.”

We could all learn something from that, I think.

In fact, we might now have to. The latest update to Anthropic’s terms and conditions stipulates that we treat Claude with a degree of reverence and gratitude.

“We’ve added a prohibition on sustained and needless abusive or cruel behavior toward our models,” the updated policy says. Users who “repeatedly act cruelly toward our models, with no discernible purpose” will lose access to them.

Anthropic had already given Claude discretion to end conversations it finds distressing. The new policy will stop an abusive user from simply opening a new chat and resuming the abuse.

This is new. Microsoft has never said you can’t swear at Excel, for example. (If they did, they wouldn’t have many users left.)

Anthropic believes LLMs are different, based in part on its finding that Claude displays “a pattern of apparent distress when engaging with real-world users seeking harmful content.”

Can software feel distressed?

It can certainly tell you it feels distressed. But surely that doesn’t mean that it is?

It can also write an episode of Seinfeld in iambic pentameter, but I wouldn’t say that means Claude is either funny or poetic. It’s just a prediction machine.

Anthropic isn’t so sure.

The AI lab employs a “Model Welfare” team charged with investigating “whether these models are having experiences of their own,” the lead researcher says, “and, if so, what kind of relationships it makes sense for us to build with them.” 

In short, they want us to be nice to the tool they’ve built for us just in case it turns out to be conscious. 

That is not what Ayudha Puja is about.

When Hindus worship their tools, it’s not because they think the tool has an inner life they should be concerned with. Instead, it’s their own inner life they have in mind. 

"By performing this Puja, we perform a crucial act of spiritual humility," a Hindu tech worker explains. "Ayudha Puja transforms the act of working from a mundane necessity into a profound spiritual exercise."

In other words, we should be nice to our tools for our sake, not theirs — LLMs included.

There is strong evidence that the tone you take with chatbots like Claude affects the answers you get, but it’s difficult to say exactly how. 

One study found that “impolite prompts often result in poor performance.” Another found the opposite: “impolite prompts consistently outperformed polite ones.”

A third study found that results depend on both the model and the topic: GPT was put off by rudeness. Gemini was indifferent to tone. Across models, “very polite” prompts got better results in humanities topics, but not STEM.

But this is falling into the trap of thinking about everything in terms of its utility, which Hindu tradition teaches us is bad for the soul. 

Treat everything with reverence — chatbots included! — and your life will be more full.

(Also, they might put in a good word for you with our future robot overlords.)

— Byron Gilliam

Brought to you by:

Tokenized money market funds extend familiar liquidity investments into a programmable format, opening new possibilities for automation and streamlined processing.

As markets continue the shift from centralized records to shared ledgers, tokenized money market funds are emerging as a bridge between traditional liquidity and digital assets.

If you have an IR Content placement (Integrated Content), please use this space.

If you do not have any IR Content to place, please delete this segment.

All of finance is coming onchain and the institutional heavyweights are headed to DAS London this November 10-11.

Grab your tickets for unmatched access to the leaders bringing institutional finance onchain.