shal

why not?

[shal @shalseban]

@nikitabier I do think world models would fit it in here a little bit -

https://t.co/6JnGnqZcUA

[shal @shalseban]

over the last week i built a web app that let my nephew play as his teenage dad in 1980s india. quite similar to the animus from assassin’s creed, and all it took was off-the-shelf ai models.

i worked with my cousin to fill in as much of the world as possible: where he was, what he wore, what he read, what music he listened to, what was around him.

for the stuff he didn’t remember, the world model basically sampled from a probability distribution conditioned on what was likely for that time and place: common clothes in 1980s india, what teenagers listened to, what a station or street might have looked like, etc.

for someone from the 70s or 80s, there’s a lot of this probabilistic filling-in that world models need to do. but you can imagine how different this gets for gen z and gen alpha. they’re going to leave behind thousands of photos and videos, messages, playlists, locations, voice notes, social graphs, game history, and probably a lot more.

so the amount the world model has to invent shrinks pretty dramatically.

while we might never find out whether our parents actually swam across crocodile-infested rivers and fought a couple of lions to get to school every morning, it’s not a stretch to imagine the next generation experiencing their ancestral past more like this.

i’m especially curious what that does downstream to parenting and society in general.

here’s a screen recording of me playing with it for a while after i fudged the names, places, dates, and faces from my cousin’s photos. everything in the video is realtime generation and me controlling the characters in a webapp.

[shal @shalseban]

s/o to folks @reactorworld. they are doing some amazing out there.

[shal @shalseban]

over the last week i built a web app that let my nephew play as his teenage dad in 1980s india. quite similar to the animus from assassin’s creed, and all it took was off-the-shelf ai models.

i worked with my cousin to fill in as much of the world as possible: where he was, what he wore, what he read, what music he listened to, what was around him.

for the stuff he didn’t remember, the world model basically sampled from a probability distribution conditioned on what was likely for that time and place: common clothes in 1980s india, what teenagers listened to, what a station or street might have looked like, etc.

for someone from the 70s or 80s, there’s a lot of this probabilistic filling-in that world models need to do. but you can imagine how different this gets for gen z and gen alpha. they’re going to leave behind thousands of photos and videos, messages, playlists, locations, voice notes, social graphs, game history, and probably a lot more.

so the amount the world model has to invent shrinks pretty dramatically.

while we might never find out whether our parents actually swam across crocodile-infested rivers and fought a couple of lions to get to school every morning, it’s not a stretch to imagine the next generation experiencing their ancestral past more like this.

i’m especially curious what that does downstream to parenting and society in general.

here’s a screen recording of me playing with it for a while after i fudged the names, places, dates, and faces from my cousin’s photos. everything in the video is realtime generation and me controlling the characters in a webapp.

[shal @shalseban]

over the last week i built a web app that let my nephew play as his teenage dad in 1980s india. quite similar to the animus from assassin’s creed, and all it took was off-the-shelf ai models.

i worked with my cousin to fill in as much of the world as possible: where he was, what he wore, what he read, what music he listened to, what was around him.

for the stuff he didn’t remember, the world model basically sampled from a probability distribution conditioned on what was likely for that time and place: common clothes in 1980s india, what teenagers listened to, what a station or street might have looked like, etc.

for someone from the 70s or 80s, there’s a lot of this probabilistic filling-in that world models need to do. but you can imagine how different this gets for gen z and gen alpha. they’re going to leave behind thousands of photos and videos, messages, playlists, locations, voice notes, social graphs, game history, and probably a lot more.

so the amount the world model has to invent shrinks pretty dramatically.

while we might never find out whether our parents actually swam across crocodile-infested rivers and fought a couple of lions to get to school every morning, it’s not a stretch to imagine the next generation experiencing their ancestral past more like this.

i’m especially curious what that does downstream to parenting and society in general.

here’s a screen recording of me playing with it for a while after i fudged the names, places, dates, and faces from my cousin’s photos. everything in the video is realtime generation and me controlling the characters in a webapp.

[shal @shalseban]

Super 📈 on this initiative!

[Liz Wessel @lizwessel]

Contrarian career advice: This is an EXCELLENT time to go into sales. So many of the best / hottest AI companies will tell you how hard they’re working to find good sellers... because no one actually believes AI is replacing AE roles any time soon (sorry to all the sf billboards out there that suggest otherwise).

Obviously, as a candidate, it’s hard to break in if you don’t have the right logos or years of experience on your resume. As AI makes the mechanics easier to pick up, the team @firstround is betting we can teach high-potential talent the uniquely human parts of the craft.

So we’re launching the @firstround Sales Combine: We’ll train you up for FREE and put our name behind you when we make the intro to companies like @notion, @fal, @Rillet, @Gumloop, @Serval & @Verkada.

More below, apply by 10/3: https://t.co/3MJTXLtgyr

[shal @shalseban]

@_onlyrams commoditized AI company* then :)

[shal @shalseban]

If you’re building a category-defining AI company selling to enterprises, take advantage of the shadow CAC subsidy while it exists. https://t.co/I90hjQD9gP

[shal @shalseban]

@AdamHoltererer same goes for claude/cowork/code.

[shal @shalseban]

it would be efficient if founders identify themselves with what they bring to the table instead of whats missing.

i am the
- sme founder
- growth founder
- design founder

avoid introducing yourself as the non-technical founder.

[shal @shalseban]

@oldestasian kudos to you on spotting that but am i missing something - is this really that hard a problem to solve for them?

[shal @shalseban]

@pursarth

[shal @shalseban]

i heard you if say cambrian explosion three times in SF - you get a guaranteed seed round

[Gajesh @gajesh]

OAI cooked here. 1B model to redact all PII at client side.

[shal @shalseban]

hah, downloading it right now.

pretty novel - I have seen some other interesting patterns of what the new container of apps would look like. Did not imagine the ideation journey would start with an image.

jolted down some other patterns here about this space - https://t.co/ix0ZtBMqsA

[shal @shalseban]

@gabemonroy Exciting!

System of record companies will have the biggest impact in enabling agent autonomy by making current environments more error tolerant.

https://t.co/gxl7ZNJdzo

[shal @shalseban]

@gabemonroy Exciting!

System of record companies will have the biggest impact in enabling agent autonomy by making current environments more error tolerant.

https://t.co/gxl7ZNJdzo

[shal @shalseban]

https://t.co/aThCXANz7N

[shal @shalseban]

@bensig huh, would not have guessed resident evil entering memory for agents race. good stuff!

[shal @shalseban]

+1 to not default to a sandbox for every interaction

however, a little bullish on optimizing the F out of the sandbox layer efficiently at scale. my guess - the SOR companies will lead the charge here, decentralizing the sandboxing to derisk actions on their system and it will not be led by a horizontal sandboxing platform service.

[shal @shalseban]

https://t.co/aThCXANz7N

[shal @shalseban]

@kat_kampf @GoogleAIStudio ✋

[shal @shalseban]

@DennisAnthony__ Did you use Sam3D Object or Sam3D Body?

I thought Body version is pure mesh and no texture.