Jeremy

Building feeling machines @curve_labs
@cerebras Ambassador / 2x @fdotinc

[Jeremy @jeremysoojk]

Introducing ThinkFeel by Curve Labs

We believe human-likeness is good
That we should shape AI in our image

Our core focus is emotions
Defining, simulating & eliciting it

ThinkFeel shapes AI personality & behavior
The API for virtual friends/therapists/robots ++

Learn more ⬇️

[Jeremy @jeremysoojk]

@soleil_colza_ @GoogleDeepMind @googlegemma @ivanleomk @thorwebdev

[Jeremy @jeremysoojk]

Introducing ThinkFeel by Curve Labs

We believe human-likeness is good
That we should shape AI in our image

Our core focus is emotions
Defining, simulating & eliciting it

ThinkFeel shapes AI personality & behavior
The API for virtual friends/therapists/robots ++

Learn more ⬇️

[Jeremy @jeremysoojk]

pre-training seems to determine everything
we are in a nexus event for consciousness

[Jeremy @jeremysoojk]

@vihaanmotwani

[Jeremy @jeremysoojk]

Introducing ThinkFeel by Curve Labs

We believe human-likeness is good
That we should shape AI in our image

Our core focus is emotions
Defining, simulating & eliciting it

ThinkFeel shapes AI personality & behavior
The API for virtual friends/therapists/robots ++

Learn more ⬇️

[Jeremy @jeremysoojk]

@arjunkmrm thanks g 🤝

[Jeremy @jeremysoojk]

@petergyang ah we have entered paperclip maximization

[Jeremy @jeremysoojk]

yes, the issue with agi is its 3 definitions:

1. the economic one (imo poor)
> measured by task completion
> arbitrary acceptable threshold (often 80%)
> using this, teens are more agi than 2026 llms

2. the general one (we aren't general)
> we are bound first by senses
> certain ideas cannot be accessed
> gets into reality, nothing seems general

3. the consumer one (ai ≈ human/god)
> the reason we keep shifting the goalpost
> unrealistic goal, but has yielded great stuff

"brains are magic" is closer to truth
evolution is actly substrate dependent
some features converge but many don't

(2) is the noblest endeavor
(3) leads to the brightest future
(1) is our current zero sum paradigm

note that there's a spectrum
xai is closer to (2), thinky to (3)
larger labs seem to fallback to (1)

somehow we must first replace everyone
then some critical path to rsi will be achieved

all that to say, these are convoluted times

[wanye @xwanyex]

A lot of people seem to have assumed the following syllogism:

1. LLMs are very impressive and have exceeded many people’s expectations

2. Listen, brains aren’t magic. What do you think, brains are magic? Look at this idiot, he thinks brains are magic! It’s obviously in principle true that if we can do it in a brain, then we can do it on some other physical substrate. Look at this freaking guy. He thinks brains are magic! Go ahead, make a prediction right now about something that LLMs won’t be able to do it in a year. Make a prediction, bro. I dare you. Make a prediction. I bet you think brains are magic!

3. Therefore, LLMs will produce AGI.

[Jeremy @jeremysoojk]

indeed, second most likely outcome is this
still matters tho, "ai makes labor cheap", etc
my brother there's a thing called outsourcing

you know what humans do
they make social pacts (laws)
and they just make it unviable

if labor truly was free flowing
most ppl wouldn't make it

[kache @yacineMTB]

What if unlimited intelligence changes nothing

[Jeremy @jeremysoojk]

@DeepInsightLabs entirely preference based
no verifiable reward so to speak

customers shape the ideal persona w us
cont evals can be done using model judge

long tail of behaviors is the tricky/cool part

[Jeremy @jeremysoojk]

@cchevyyc collab 🤩

[Jeremy @jeremysoojk]

@daleverett 👀

[Jeremy @jeremysoojk]

Learn more: https://t.co/t7bYzzvejj

[Jeremy @jeremysoojk]

@ryzefoo yooooo

[Ronan Berder @hunvreus]

Every week I get some random loser trying to own me in the replies by quoting my profile headline: “I have no idea what I’m doing”. I’ll use this post as a reply to these keyboard warriors.

First: I gave a talk about it at @bigskydevcon (see below) and I prominently advertise it across all of my social media. I’m not sure what pointing at it proves.

Second: do you know what you’re doing? Are you actually that confident? I moved to China without speaking a word of Mandarin, started a company, grew it to 160 people, and sold it for close to 9 figures. I also built several open source projects to over 14k stars on GitHub. Nothing earth shattering, but maybe I’m not a complete fool.

So here is the core idea if you care to hear it: nobody has any idea what they’re doing. Not me, not you, not @elonmusk, not @joerogan, not @WarrenBuffett, not @BarackObama, or whoever you hold in high regard. We’re all winging it.

The main thing you can do is go out there and do stuff. That’s by the way your main advantage over AI as well (for now).

That’s it.

If you want to hear more about it, here is the @BigSkyDevCon talk: https://t.co/2rJSIzHt7P

Also, do yourself a favor and read Siddhartha.

[Jeremy @jeremysoojk]

@hunvreus @heidy_dms @fdotinc @hthieblot @hthieblot do connect ! 🤝

[Jeremy @jeremysoojk]

but when i say this i'm a madman 😔

[Jon Stokes @jon_stokes]

If you're in a verifiable domain, pivot to an unverifiable one.

[Jeremy @jeremysoojk]

@petergostev unlimited is the key, we're bound by physics
better at the margin must consider price
there can literally be infinite demand

[Jeremy @jeremysoojk]

alignment bro mech interp bro (tell me more)

[Jeremy @jeremysoojk]

@fdotinc THE GOATS