Irina Malkova

CEO at Stealth. Ex-Fortune 500 data exec. Data will be bigger than models. AI is our chance to reimagine human flourishing. Opinions are my own.

[Irina Malkova @irishango]

I think about this a lot:

Dancers dance to understand movement.
Songwriters write to understand emotion.
AI researchers build to understand the mind.

We are all chasing the same thing - we obsessively want to understand the world through doing.

Even if AGI didn’t mean treasure and glory, AI people would still build it - because the desire to understand is unquenchable.

[Irina Malkova @irishango]

Model companies will not win vertical AI.

It's very simple.

Most of vertical AI is about knowledge and memory.

Most knowledge and memory problems are information compression problems. Enterprise has PBs of domain expertise. Context window is 500KB.

Compressing information requires understanding its shape and the human behaviors that produced it.

TL;DR: You must to understand the business to win vertical AI.

Anthropic, I’m not scared of you.

[Irina Malkova @irishango]

Apparently, sycophancy is a real addiction.

Stanford researchers found that it is literally addicitve and measurably bad for you.

Also apparently, the only way you can make a closed AI app like Claude less sycophanic is to add “be direct and critical…“ to every single prompt.

This is starting to look like attention economy 2.0.

https://t.co/bQxqV3LiXD

[Irina Malkova @irishango]

Ah the sweet, sweet freedom of being able to just think about something all morning.

[Irina Malkova @irishango]

I just tried Fable. It gave me the exact same result as Opus.

It's probably very smart. It's just that the things I'm asking it to do are simple enough so even Opus handles them perfectly.

Maybe intelligence is no longer a constraint. My ability to give it Fable-worthy things to do is.

P.S. it also ate all my tokens in one go.

[Irina Malkova @irishango]

The first week of founder life was great.

I have one grievance: moving my AI setup.

My old setup worked across my Salesforce systems, my personal email, and the advisory work that is now the whole job.

But I'm realizing it was entirely unmovable.

- My documents live across Obsidian, GitHub, Google Docs, and Slack canvases.

- My skills are scattered and outdated, and the writing I needed at Salesforce is not the writing I need now.

It really doesn't feel too different from the pain of switching health insurance.

I'll rebuild, and I'll do it with the right deployment strategy this time.

But I wish personal intelligence would be easier by now.

Does one really need a deployment strategy? What's next, a personal data lake?

[Irina Malkova @irishango]

I spent almost 11 years at Salesforce.

Not because of the famous work life balance.
I was too busy growing my function 50x to take advantage of that.

But because Salesforce is classy.

Because Salesforce has taste.

First, the product is timeless.
The UI buttons may be fast fashion. Yet the platform is the Birkin: engineered to model the messy lead-to-cash reality with enough flexibility to bend and enough discipline to scale. Fortune 100 use it. AI natives use it. Launched in 2000, it's hotter than ever.

Second, the experts are excellent.
Salesforce's customer success pros set the industry bar. Sales reps are aggressive but so charming you never want them to stop talking. Office crew makes sure not a leaf is out of place on the living wall. Excellence is everywhere. People know what they're doing.

Third, the leaders are masterful.
I grew up in the Soviet Union, where command-and-control was the default and the result was misery. It taught me there's no point in bringing people together if leadership can't unlock the best in each of them. Salesforce's exceptional senior leaders showed me how. I watched them take responsibility. I watched them spot and lift talented junior people. Again and again.

Once the titles are earned and the features are shipped, what's left is the taste you built as a leader.
I was fortunate to build mine at Salesforce.
And now I’m building my own tasteful enterprise.

So long, Salesforce.
Thank you for everything.

[Andrew Yang🧢⬆️🇺🇸 @AndrewYang]

My new favorite meme. 😀

[Irina Malkova @irishango]

@AndrewYang Y tho. Please set your face free.

[Irina Malkova @irishango]

@JackInacker @AndrewYang

[Irina Malkova @irishango]

@JackInacker @AndrewYang I vote mandarin not Spanish otherwise it’s literally Glee

[Irina Malkova @irishango]

@JackInacker @AndrewYang Lol I mean community sorry

[Irina Malkova @irishango]

Eyyyyyyy First Democratic Debates at @welcometomannys with @DontCallMeNino!!

[Scott Santens @scottsantens]

It appears the U.S. Congress has been undergoing the telophase of mitosis for the past 30 years. https://t.co/M2LJJuutAq #science

[Irina Malkova @irishango]

@pdgoldman It’s “feedforward”, but I believe “closed system” is more pungent because such algorithm typically doesn’t have “feedback” in a real way either (training data might refresh but model assumptions won’t change). Real problem is assumption of no interaction with the world it models.

[Irina Malkova @irishango]

@Benioff #1

[Irina Malkova @irishango]

“I want to watch America fall in love with the man I’m in love with” -Chasten. Yes!!!!! I love you Pete!!!

[Irina Malkova @irishango]

All my techies into behavioral econ: “Putting price tag on some human services corrupts our intrinsic values”
+
Also them: “I dedicate my life to you being able to pay a stranger to drive your mom to the airport”
=
Also them: “Who screwed up this country???”
🤷🏻‍♀️
I’m guilty too

[Irina Malkova @irishango]

@clairlemon As much as I hate this, if I was a profit-maximizing publisher I’d do the same. “Sensitivity readers” is an inevitable market response to the real problem: “sensitive readers”. We must rebuild ability to have disagree constructively.

[Irina Malkova @irishango]

@Quillette @CraigDeLancey I hate this but if I was a profit-maximizing publisher I’d do the same. “Sensitivity readers” feels like an inevitable market response to the real problem: “sensitive readers”. Not surprised this started in YA literature given @JonHaidt’s research on callout culture on campuses.