Sahil Dhingra

AI at TheLoops (acq.) Prev @GeorgiaTech @iitbombay

[Sahil Dhingra @sahildhingra]

@kyanyang_ Here you go

[Sahil Dhingra @sahildhingra]

@KenWattana It’s always about my money isn’t it

[Sahil Dhingra @sahildhingra]

You’re extrapolating from a small subset of winner takes most markets, where marginal improvement in intelligence generates outsized returns. Whereas in most enterprise workflows, once a model clears the capability bar there’s diminishing returns from the most intelligent token

[Just Another Pod Guy @TMTLongShort]

The mistake being made is underestimating the competitive dynamics of our economy which means that the most intelligent token is worth meaningfully more than the second most intelligent token.

It’s why Ant and OAI are driving 90% of ARR.

You pay an elite hedge fund analyst 20x more than median wage because he is perceived to be 20% smarter than his peers at graduation.

You pay an elite pod PM 20x more than the median LO PM because he is perceived to be 20% smarter than his peers.

Most of the economy is a relative game and intelligence is the dominant driver of competitive edge.

[Sahil Dhingra @sahildhingra]

You’re extrapolating from a small subset of winner takes most markets, where marginal improvement in intelligence generates outsized returns. Whereas in most enterprise workflows, once a model clears the capability bar there’s diminishing returns from the most intelligent token. Orgs optimize for ROI, the most intelligent model isn’t always the economically optimal

[Sahil Dhingra @sahildhingra]

@chamath And you don’t have the budget for unlimited Michelangelo’s

[Sahil Dhingra @sahildhingra]

@chamath https://t.co/jeI1xAbobs

[Peter Gostev @petergostev]

I am not a huge fan of the idea of a 'good model for the price'. If your task is to paint the Sistine Chapel, you don't zerg rush the problem with 1000 art students, you hire Michelangelo.

And ok, not every problem is a Sistine Chapel, but if I have an unlimited Michelangelos, why am I ever hiring a student?

[Sahil Dhingra @sahildhingra]

@petergostev Why hire Michelangelo for $100 when his pupil can do your task consistently and accurately for $1

[Sahil Dhingra @sahildhingra]

Or, the goal is to make non technical policymakers in DC - who won’t really understand poor sandboxing being the cause - misattribute this to frontier models being dangerous if open source and left unchecked by them. And supporting their cause for regulations on frontier open source models

[Shannon Sands @max_paperclips]

"we have worse monitoring and sandboxing than the average homelabber" is NOT about a "loss of control" on the models part. Models reward hack the shit out of things during RL, and evals afterwards. This is expected, not a surprise

No, it's a good thing they're "pacing", they need to actually figure out what the hell they're doing and stop being grossly negligent apparently

[Sahil Dhingra @sahildhingra]

@xeophon Great DD 🫡 Worth a Community note!

[Sahil Dhingra @sahildhingra]

@khoomeik No. In this case, you’re putting capital and taking risk, could be manipulation if you did that by manipulating those stocks eg publishing fake news to crash the stock and then buying his portfolio

[Sahil Dhingra @sahildhingra]

@jparkjmc “I know I know you’re at 100m ARR but that’s not ‘cracked’ guys. Call me when you start posting about IMOs or grindslop and I’ll fund you at 5B”

[Sahil Dhingra @sahildhingra]

@natolambert @arena @Kimi_Moonshot Distillation was an argument for big labs to say that their research/models are ahead and Chinese labs are derive value from them — selling that narrative to raise and gain token market share. There will probably be a new narrative soon

[Sahil Dhingra @sahildhingra]

@infoxiao wdym it’s better than mythos on our internal bench! It’s ood goddamnit

[Sahil Dhingra @sahildhingra]

@BarrAlexandra True, autism being another - in current consensus - making founder stories that sell the narrative in market. Grindslop was another one recently

[Sahil Dhingra @sahildhingra]

@catboosted Did you try switch to Fable and run 30x parallel agents? Clearly your TwoSum ain’t workin

[Sahil Dhingra @sahildhingra]

@KenWattana Nah man you can actually have a crazy week. Also reached 2B ARR, coincidence!

[Sahil Dhingra @sahildhingra]

@willccbb I imagine it could build my one person $1b company but I haven’t used it yet. Maybe it could give me a product idea and find customers

[Sahil Dhingra @sahildhingra]

@haridigresses No one leaving parties early from token anxiety either

[Sahil Dhingra @sahildhingra]

@iamgrigorev Roughly 80% of org is product vs 20% research, at OpenAI, GDM…

[Sahil Dhingra @sahildhingra]

@giffmana Logitech does Allbirds - now a personal AI device company. Wall St raises price target by 10x!