Srividhya Prakash

SWE @Google trying to fight abusive content across Google's products. Ex @yahoomail. Masters in Comp Engg @SJSU. Tamil. Like to play badminton and ski ⛷️ :)

[Srividhya Prakash @silvershadow94]

@pradeepcep Fair, and ethical.

[Vir Das @thevirdas]

Colbert isn’t the first comedian to lose a battle to a political establishment. From Lenny Bruce to current day comics, comedians who don’t bend usually lose. Ironically it sharpens their voice to the very best version of itself. He will move on being the best he’s ever been, the most relevant he’s ever been and with the sharpest pen he’s ever had. If he’s willing to take risks, what he does next will be his legacy, not a long run of a very mainstream CBS show. The freedom that comes with a deep defeat at the hands of insecure power is priceless. Man I hope he uses it!

[Srividhya Prakash @silvershadow94]

Guna, antha raaja sir paatu..

[Peter Girnus 🦅 @gothburz]

I'm the VP of AI at Apple.

I've been here since 2011.

I watched Siri launch.

It was revolutionary.

For about six months.

Then Google Assistant came out.

Then Alexa.

Then ChatGPT.

We kept saying we were "focused on privacy."

Privacy is what you say when you're losing.

Three years ago the board asked about our AI strategy.

I showed them a slide that said "On-Device Intelligence."

They nodded.

They didn't know what it meant.

Neither did I.

But it had a picture of a neural network.

Neural networks look impressive.

Even when they don't work.

Last year someone asked Siri to set a timer.

It opened a Wikipedia article about timers.

Tim saw the meme.

He didn't laugh.

He scheduled a "strategic offsite."

Offsites are where we go to admit failure privately.

I presented three options.

Option 1: Build our own LLM.

That would take four years.

We don't have four years.

Option 2: Buy a startup.

We looked at twelve.

They all wanted $40 billion.

For teams of nine people.

Who would leave after the acquisition.

Option 3: Call Google.

The room went quiet.

Google is the enemy.

We've spent fifteen years pretending we're better than Google.

Our entire brand is "not Google."

But Google has TPUs.

We don't.

Google has Gemini.

We have Siri.

Siri still can't reliably add items to a grocery list.

I called Sundar.

He picked up on the first ring.

He'd been waiting.

They all wait.

Eventually everyone calls Google.

I asked for TPU access.

He said yes.

I asked for Gemini integration.

He said yes.

I asked how much.

He said one billion dollars.

I said that's a lot.

He said "per year."

I paused.

He said "you don't really have a choice."

He was smiling.

I could hear it.

We announced it as a "strategic partnership."

Partnership means we're paying them.

The press release said we're "enhancing Siri's capabilities."

Enhancing means replacing.

We said the new Siri arrives "late 2026."

Late 2026 means 2027.

Maybe 2028.

Definitely not 2026.

A reporter asked if this means Apple lost the AI race.

Our comms team said we're "thoughtfully deliberate."

That's not an answer.

But it has enough syllables to sound like one.

Internally, we're calling it "Project Humble Pie."

Someone suggested "Project Brain Transplant."

HR flagged that as "not brand-aligned."

The engineers are relieved.

They've been trying to make Siri work for years.

Now they can blame Google.

Blame is a renewable resource.

Tim did a podcast.

He said AI is "a profound technology."

He's never used ChatGPT.

I showed him once.

He asked why it was typing so slowly.

I said that's how it works.

He said "Siri should be faster."

I said "Siri will be Google."

He said "don't say that publicly."

I won't.

Publicly, we're "leveraging industry partnerships."

Leveraging means surrendering.

But with dignity.

We still have the best hardware.

We still have the ecosystem.

We still have the brand.

We just don't have AI.

So we're renting it.

From the company we've mocked for two decades.

The one billion dollars is a licensing fee.

The real cost is the narrative.

We were the innovators.

Now we're the integrators.

But the stock is up 3%.

Wall Street doesn't care about innovation.

Wall Street cares about not falling behind.

We're not falling behind anymore.

We're being carried.

By Google.

For one billion dollars a year.

I'll present this as a win at the next all-hands.

Wins are whatever you frame them as.

The graph will go up and to the right.

It always does.

As long as you pick the right metric.

[Srividhya Prakash @silvershadow94]

Was searching for kuthu hop beats yesterday when YouTube suggested to me this grooooovy song - https://t.co/PhHsHpI4XM

Slight-ah restored my faith in YouTube recommendations.

[Srividhya Prakash @silvershadow94]

Ask Perplexity - "What is agent cardinality in the context of LLMs?" to know more.

[Srividhya Prakash @silvershadow94]

Doomsday begins with Agent Cardinality, 🥲

[Srividhya Prakash @silvershadow94]

A simple and solid use case for LLM's as a product/app -
1. Upload the little tag that comes in dresses on how to take care of the dress.
2. Take a photo of clothes to be washed in the washing machine.
3. What the settings should be (hot/cold water, rinse) is predicted.

[Sridhar Vembu @svembu]

All the LLMs and other deep learning models are based on neural networks. We can think of them as mathematical functions with hundreds of billions of parameters.

Those parameters (weights) are determined during training and we train these networks with trillions of tokens (text, images, videos that are split up into tokens to be ingested by the models).

We can say that every one of the trillions of tokens played a part in determining the value of each of the hundreds of billions of parameters.

The image I have in mind is a giant lake where we dissolve trillions of cubes of salt, sugar etc. After the dissolution we cannot know which of the cubes of sugar went where in the lake - every cube of sugar is everywhere!

Therein lies a problem: if we use a business database, such as customer relationship data, to train a neural network model (i.e to determine its parameters), when the customer changes that data or deletes the data, we do not know how to alter the weights of the model to account for this change in the data. Even if the model were dedicated to that customer, we still cannot guarantee the customer that their changes to the data will be reflected in the model.

In that sense, neural networks (and therefore LLMs) are NOT a suitable database.

This is a fundamental limitation of the current scientific mathematical approach and cannot be fixed only by technological fine tuning.

The RAG (retrieval augmented generation) architecture keeps the business database separate and augments the user prompt with data fetched from the database.

In that case, the model itself is not trained on the (potentially changing) customer data because that data is only used in the prompt.

But RAGs can only go so far.

I personally have come to believe more foundational work is needed. What does that look like? All I have right now are hunches. That is the existing part of scientific work!

[Srividhya Prakash @silvershadow94]

❤️

[Lokesh Kanagaraj @Dir_Lokesh]

Loved Un Aasai Mugam, a crowdfunded Independent film ❤️

https://t.co/QVEIA0xca7

My hearty congratulations to @VimalSantiagu, @VishwajithPai, @Maathevan and the entire team 🤗❤️

#Pavithraa #JaneThompson #DrKarthikeyan @kabervasuki @SenthamizhanAM @siennor #Jai #ManojKumar #DineshRaja #VincentST

[Srividhya Prakash @silvershadow94]

If you are doing it right, the new year is not the day you start but the day you give yourself a break.

#random

[Lex Fridman @lexfridman]

I'm in Kyiv to interview President Zelenskyy, trying to do my small part in pushing for peace.

This photo is of me visiting Babi Yar yesterday, a place where many in my family were slaughtered by Nazi forces in 1941. They were ordered to gather with valuables with the promise they'd be "resettled", and then forced to lay down in this ravine on top of other people's bodies and were shot. Over 30,000 people were slaughtered in this way in just 2 days.

Let me add another note, because sadly I'm attacked a lot online by all sides but in this case Ukrainian people. I'm told by many Ukrainian friends (living in Ukraine) that the attacks are voices propped up by Ukrainian bot farms. I disagree, and I think it's not a good way to operate intellectually, thinking that anyone attacking me is a bot, and anyone supporting me is a smart thoughtful human being 🤣 Maybe it's true sometimes, but it's better to assume it's not. I prefer to assume it's just a lot of passionate people who care about Ukraine and yes sometimes get caught up in the witch-burning hysteria of the crowd. The far left and far right in United States did this a lot over the past few years.

Anyway, in the previous post, I already correcting a bunch of lies spread about me online about my my background. I explained my family roots in Ukraine, and now let me add some more context to the pile about my previous visit to Ukraine during the war.

I visited Ukraine in summer of 2022, traveling to Bucha, Borodyanka, Kyiv, Kryvyi Rih, and several places on the front in Kherson Oblast.

This trip was personal. Most of it was not recorded, and was not meant to be recorded. I had two goals for the trip:
1. To interview President Zelenskyy
2. For me to personally understand and feel the reality of this war.

For the first part, President Zelenskyy eventually agreed, and that's why I'm back in Kyiv.

For the second part, I spoke with hundreds of people off-mic (not recorded, just human to human), including soldiers, civilians, politicians, artists, religious leaders, journalists, economists, historians, and technologists. I recorded only a tiny number of these, with no intent to publish them as standalone episodes, but instead to maybe consider including them in a documentary-style video as part of a Zelensky interview (if it happens during the 2022 visit), kind of like David Letterman did. But the project quickly fell apart and started to not make sense, not in the way I was approaching it. As I was speaking with people (off-mic), the conversations I enjoyed having most and that I felt would powerfully show the beauty and pain of Ukrainian people in this war would be with hundreds of soldiers and civilians. The interviews I DID record were simply just not good conversations, and it's my fault, and I take full responsibility for that. They were short (by my standard: ~1 hour) where I asked disparate generic questions, which resulted in shallow generic conversation. I quickly realized that I would need to change my approach. I would need to either make a documentary by recording hundreds of conversations with soldiers & civilians or do full normal deep-dive 3-5+ hour podcasts with individual people. I agreed to do the latter with a few folks I met, including President Zelenskyy. I did an episode with Ukrainian historian Serhii Plokhy in this style.

Almost all the people I spoke with on and off-mic have reached out with support and total understanding. Many have become good friends. Still, I'm deeply sorry for the many ways I've failed in this effort, but I promise I'm working really hard to get better.

I really do try with all my heart to speak to people from all sides with empathy, depth, and compassion.

I'm sure the attacks will continue, but at least now you have some more context.

Sorry for the long post, and any mistakes (I didn't proofread). I'm writing it looking over Kyiv as the sun rises.

Happy Holidays. I love you all ❤️

[Srividhya Prakash @silvershadow94]

Might be useful for people doing research.

[Thoreo @thoreoai]

the kids at @cursor_ai built different fr

[Srividhya Prakash @silvershadow94]

This could be an ongoing practice that happens every week that takes national issues and current events into consideration to come up with questions. (5/5)

[Srividhya Prakash @silvershadow94]

This cultivates problem-solving and critical thinking which are very useful skills. (3/5)

[Srividhya Prakash @silvershadow94]

The answer essays of all students in a class/cohort can be later passed to an LLM to come up with data visualization on what were the most common themes on students minds, and they addressed by teachers and experts (industry professionals invited to give talks). (4/5)

[Srividhya Prakash @silvershadow94]

The emphasis should not be on the feasibility of such a scenario but about the pros and cons on kinds of policies that would need to come into existence/come into existence, implementation mechanics, diversity and any other broad topics students can think of. (2/5)

[Srividhya Prakash @silvershadow94]

Thought provoking and engaging questions like "What would it take to send a cohort of 500 people to the Olympics?", "what would happen if Indian tech CEOs of major companies were to be appointed in key positions of the government?" should be asked to students in school. A 🧵.

[Srividhya Prakash @silvershadow94]

@pradeepcep @msft4startups @_buildspace @_nightsweekends 👏

[Srividhya Prakash @silvershadow94]

Maybe it's the late night, but I couldn't stop laughing.

[sheikh usamaro (族) @0xUsamaro]

VCs asked me for a demo

so i sent them this

[Srividhya Prakash @silvershadow94]

For interested folks, Google Cloud experts will lead online classes for 4 weeks, focused on the latest gen AI products, best practices, & solutions to help you & grow your business. Register here - https://t.co/vSQl6fx7SJ

Starts in a week.