Kyle Tut

Everyone is from somewhere. Cofounder and CEO @pinatacloud

[Kyle Tut @KyleTut]

If you talk to a commercial electrician, you'll know the exact state of the AI bubble.

[Kyle Tut @KyleTut]

I don't understand what this section of Stripe's website does for selling payment processing

[Kyle Tut @KyleTut]

They are capable.

They just haven't switched from R&D expense to operating expense.

That will change the math.

[nicolas, 30 ans @nic_carter]

I love how bears think companies are completely incapable of verifying token efficiency. In their mind CFOs have absolutely no idea how much to order. $1m or $100m a quarter. No idea

In the real world, obviously, the job of the CTO is to supervise token usage. Spy on the exact queries if they must. Create novel efficiency metrics. Fire employees who divert tokens for personal projects. Distribute tokens according to proficiency. It’s not this impossible task the bears imagine it is.

[Kyle Tut @KyleTut]

AI Spend ≠ Revenue

[Kyle Tut @KyleTut]

Reddit is the front stage of a play where stories are performed for entertainment.

Reality is backstage where actors are fighting, stagehands break props, and producers are desperately protecting the illusion.

AI is trained on the show, not the production, creating reddit physics.

[Danielle Fong 🔆 @DanielleFong]

swear to god all of the models are mistrained on something I call "reddit physics"

[Anthropic @AnthropicAI]

Anthropic has confidentially submitted a draft S-1 registration statement to the Securities and Exchange Commission.

Pending completion of SEC review, this gives us the option to pursue an initial public offering.

Read more: https://t.co/onGZAhRLvD

[Kyle Tut @KyleTut]

@blauyourmind Cool. Tag me when shared

[Kyle Tut @KyleTut]

Anyone hooked their agent up to Robinhood yet?

[Kyle Tut @KyleTut]

@0xloveavax https://t.co/SJwH0og2dW

[Kyle Tut @KyleTut]

@GrantSlatton It's because they are subsidized. The behavior changes when the tokens are no longer subsidized.

https://t.co/6ZB0nwNp7M

[Kyle Tut @KyleTut]

His total spend would stay the same but his output and usage behaviors would drastically change.

He estimated he was getting $10k/month in tokens out of the $550/month he spends on his subscriptions.

As everything converts to unsubsidized usage billing, he expects his behavior to change from "let's ask five models" to "which model(s) gives me the best output per dollar".

[Kyle Tut @KyleTut]

All you have to do is watch an F1 race with impending rain to understand that all forecasting is fatally conceited.

[Patrick Collison @patrickc]

I’ve been struck by this phenomenon in much of the discussion around Hormuz. Who exactly should one listen to for systems as complex and reflexive as energy? (Evidently not IEA.) Is it even possible to make meaningful predictions for out-of-distribution shocks like strait closure given all of the second-order effects that one has to model? Are all forecasts fatally conceited?

[Kyle Tut @KyleTut]

His total spend would stay the same but his output and usage behaviors would drastically change.

He estimated he was getting $10k/month in tokens out of the $550/month he spends on his subscriptions.

As everything converts to unsubsidized usage billing, he expects his behavior to change from "let's ask five models" to "which model(s) gives me the best output per dollar".

[Kyle Tut @KyleTut]

Interviewed a dev today spending $550/month on personal LLM subscriptions.

He thought it was worth it for now but also said that it was to arbitrage the subsidized pricing and stay up to date on how the latest models performed.

Indicated this behavior would instantly change with usage based pricing.

[Kyle Tut @KyleTut]

What behaviors survive once prompts stop feeling free?

[Kyle Tut @KyleTut]

@alex The "new" isn't hard to compete against. It's really just subsidization.

https://t.co/6xsuWzlpWy

[Kyle Tut @KyleTut]

Interviewed a dev today spending $550/month on personal LLM subscriptions.

He thought it was worth it for now but also said that it was to arbitrage the subsidized pricing and stay up to date on how the latest models performed.

Indicated this behavior would instantly change with usage based pricing.

[Kyle Tut @KyleTut]

@nic_carter If you're in AI pivot to distressed assets

[Kyle Tut @KyleTut]

@amandacassatt *Cisco Webex

[Kyle Tut @KyleTut]

@JohnHuber72 @SlovenianD Someone needs to exploit this inefficiency.

[Kyle Tut @KyleTut]

Build a paperclip maximizer with https://t.co/IHLv3j9qeN

[Tanay Jaipuria @tanayj]

Couple of wild anecdotes from Nat Friedman about his OpenClaw

[Kyle Tut @KyleTut]

paperclipmaxxing