Sabrina Halper

Exploring the mind, machine, and meaning of it all 🎙️🔮 | @stanford

[Sabrina Halper @SabrinaHalper]

Ken Liu @kyliu99 on how modern explanations to questions like "why do we dream?", "why do I love this person?", "why are we here?" explain everything while explaining nothing at all.

Science can explain mechanism. It can't explain meaning. Stories create meaning.

Modern states ask citizens to die for them all the time - for the sake of a story. Humans will die for a story. It's the only thing humans have ever willingly died for.

[Sabrina Halper @SabrinaHalper]

Congrats to @jackdent and the @chaidiscovery team on your deal with Pfizer!! In our conversation, Jack told me that pharma companies are moving a lot faster on AI than people realize and that's certainly proving to be true.

When asked about what keeps him going when working on an area that is filled with unknowns and historically harder than other sectors, Jack pointed to three things:

1.The problems are so intellectually rich you could spend a lifetime studying them and feel rewarded by that alone.
2. If they work, they can reshape one of the largest industries in the world.
3. They can genuinely help people at scale.

It is rare to work on something this hard, this fascinating, this commercially massive, and this good for the world.

[Jack Dent @jackdent]

Fun to sit down with my friend @SabrinaHalper and talk about Chai's journey. Thanks for having me on!

[Sabrina Halper @SabrinaHalper]

@FDavidsonT Just in time for summer - l’m excited for this!!

[Francis Davidson @FDavidsonT]

Meet Odessia.

She plans and books entire trips in one conversation.

Apparently, she's the goddess of travel.

[Sabrina Halper @SabrinaHalper]

@TheAnnaGat Richard is awesome , you can watch the full interview here: https://t.co/2dZ6Lm1iOY

[Sabrina Halper @SabrinaHalper]

@TheAnnaGat Thank you Anna ❤️ I like yours!

[Sabrina Halper @SabrinaHalper]

"I think in general we're moving towards an interaction pattern where you're more of a manager and you're kind of delegating things rather than actually doing the task yourself. But, thinking like a manager is often hard for people." @isafulf

Everyone second time founder I've spoken to who's building a new company right now is shocked at how different it feels early on this time around. They're putting together the initial product with a team of agents, and the job feels a lot more like conducting than ever before.

[Sabrina Halper @SabrinaHalper]

Who do you think will be the most powerful human in 20 years?

"It might just be the person who has the most coherent picture of what they want the future to look like." @RichardMCNgo

It seems valuable to spend more time thinking and developing a strong pov of what you want things to look like. Easier than ever to execute on it.

[Sabrina Halper @SabrinaHalper]

@jfbrly @bryceehunt wow

[Sabrina Halper @SabrinaHalper]

Jack Dent story time - his path from high school in the UK to founding Chai is pretty amazing:
- Cold-emailed Stripe as a teenager
- Daniela Amodei was running Stripe’s careers inbox and forwarded his email to Patrick Collison
- Interned on Stripe’s payments API and eventually became the only engineer working on it
- Greg Brockman walked around the office with no shoes
- Met @joshim5 while at Harvard. Years later, Sam Altman Facebook-messaged him about Josh when he was leaving OAI - asked if he could be convinced to stay or start a company
... All of it became part of the path to co-founding @chaidiscovery a few years later.

Early Stripe was a talent epicenter @DanielaAmodei @patrickc @gdb @sama

[Sabrina Halper @SabrinaHalper]

@ericwu01 Go Eric!!!

[Sabrina Halper @SabrinaHalper]

Spotify: https://t.co/bg3yutbWdx
Apple: https://t.co/wkeFxctarM

[Sabrina Halper @SabrinaHalper]

"We have this long term vision of turning biology from something which is trial and error and experimental and make it something that looks more like an engineering discipline in the next century."

New Episode with @jackdent !

0:00 Intro
02:28 Inside @stripe w/ @patrickc, at <100 People
08:45 @sama role in @chaidiscovery origin
16:20 Chai-2 Antibody Breakthrough
19:24 Approaching biology as an engineering problem
24:45 Using AI models to treat people on an individual scale
26:52 Robotics in labs
27:22 Pharma industry today
29:49 The clinical trial bottleneck
32:53 Personalized biology models & cancer vaccines
36:24 Longevity, peptides, and trillion-dollar drugs
39:02 Does it matter that the government cut exploratory science grants?
41:00 What next-gen Chai models will do
42:08 Can frontier AI labs build this themselves?
43:26 Biology is hard

[Sabrina Halper @SabrinaHalper]

YouTube:https://t.co/ElMGkuRcxF

[Sabrina Halper @SabrinaHalper]

The Claw ✋

[Sabrina Halper @SabrinaHalper]

Spotify:https://t.co/CkQM2WDANi
Apple: https://t.co/vEx4g4EhdV

[Sabrina Halper @SabrinaHalper]

.@dwarkesh_sp's episode with @ericjang11 is awesome. Eric has a rare gift for making complicated ideas feel simple, which is exactly what struck me when I first met him at a dinner and immediately peppered him with questions.

If you are looking to hear more from him, listen to our conversation from a couple years ago!

"The history of neural networks and actually many things in AI have been inspired a lot by biology, like genetic algorithms, evolutionary algorithms, processes in both neuroscience and psychology and biology and so forth. I think drawing from nature is a great way to get inspiration... And this is going to get me a lot of flack from the scientific community, but I feel like in the last decade and a half-ish of deep learning progress, the vast majority of major contributions have come from people who did not really adhere to that way of thinking, but more like, how do I push as much data as fast as possible onto my GPU?
It just empirically seems to me that people who are very attached to the idea of replicating a particular nature inspired architecture at the expense of enabling brute force compute... do not make the best algorithms. There's this essay by @sarahookr called the the hardware lottery where it's arguing that people who focused on maximizing and designing their algorithms to suit the hardware so that it could run the brute force thing as fast as possible tended to win. "

[Dwarkesh Patel @dwarkesh_sp]

New blackboard lecture w @ericjang11

He walks through how to build AlphaGo from scratch, but with modern AI tools.

Sometimes you understand the future better by stepping backward. AlphaGo is still the cleanest worked example of the primitives of intelligence: search, learning from experience, and self-play. You have to go back to 2017 to get insight into how the more general AIs of the future might learn.

Once he explained how AlphaGo works, it gave us the context to have a discussion about how RL works in LLMs and how it could work better – naive policy gradient RL has to figure out which of the 100k+ tokens in your trajectory actually got you the right answer, while AlphaGo’s MCTS suggests a strictly better action every single move, giving you a training target that sidesteps the credit assignment problem. The way humans learn is surely closer to the second.

Eric also kickstarted an Autoresearch loop on his project. And it was very interesting to discuss which parts of AI research LLMs can already automate pretty well (implementing and running experiments, optimizing hyperparameters) and which they still struggle with (choosing the right question to investigate next, escaping research dead ends). Informative to all the recent discussion about when we should expect an intelligence explosion, and what it would look like from the inside.

Timestamps:

0:00:00 – Basics of Go
0:08:06 – Monte Carlo Tree Search
0:31:53 – What the neural network does
1:00:22 – Self-play
1:25:27 – Alternative RL approaches
1:45:36 – Why doesn’t MCTS work for LLMs
2:00:58 – Off-policy training
2:11:51 – RL is even more information inefficient than you thought
2:22:05 – Automated AI researchers

[Sabrina Halper @SabrinaHalper]

YouTube: https://t.co/Y1vlHMGETq

[Sabrina Halper @SabrinaHalper]

Richard ngo @RichardMCNgo on what happens to human work -

"depends on how much you think current companies are already oriented around, let's say, like risk management, and social labor. I have this concept of the sociopolitical economy, which is a world in which the main things that humans are doing is the thing that AI inherently can't do, namely, tasks that require you for political, social reasons to be human and already I think more of the economy is oriented around this than you might think. You know, even in the most productive corporations, like Google for example, there's just a lot of people there who are fundamentally there because it would harm morale if you fired them. probably the Google bosses already know a bunch of people who they could fire without much productivity lost. But like it wouldn't be very googly to do that"

"I think that doesn't go away as long as humans retain fundamental political power. And so, so you kind of do have this fork where on, on one hand, like if humans are ultimately in charge, if like the decision makers and the government are humans, , if the laws are oriented to favoring humans, then you know, you get this trickle down effect."

"suppose, we assume that there are so many industries right now that just didn't exist a hundred or 200 years ago that are fundamentally social in nature. So I think, professional sports for example, it's just like this incredibly large industry that you can never automate. Just like fundamentally, like we know in chess, right? you know, you've automated chess, and then Magnus Carlson earns way more than any chess player ever has before. And that's not just at the top levels. It like trickles down. You have these like little league coaches or you have your school or your college football coach who earns like ridiculous amounts of money. Yeah. And so if you just picture that happening, but for every hobby that people spend a reasonable amount of time on that, you can sort of imagine the economy just like massively diversifying into like many different niches of like social and community-based activities."

[Sabrina Halper @SabrinaHalper]

@TheAnnaGat https://t.co/44xoIM79RA