Alex @AlexanderMoini

founder @ Theus AI bringing radical liquidity to CRE theus.ai DTX Joined December 2022

Tweets

760
Followers

497
Following

1K
Likes

3K

Alex @AlexanderMoini

a day ago

@yacineMTB holy gpu

0 0 0 21 0

View Details

Alex @AlexanderMoini

a day ago

@bryan_johnson @_katetolo best breed man, grew up with malinois' and you build such a deep connection with them. wonderful time you have ahead

0 0 0 397 0

View Details

@tanayj @AndrewCurran_ well if their finetuning is what it sounds like, which is giving reward for writing in distribution of other human writing, this may not work without labeling human writing as AI as well.

0 0 5 90 1

View Details

Alex @AlexanderMoini

5 days ago

the simplest version of this is mercor has to resell it with “meat on the bone” - google is able to pay more for it because they can capture more of the end value by selling tokens instead of the data. i’m also sure they probably underwrote it against what they think their experts could do for the same price.

0 0 2 437 0

View Details

Alex @AlexanderMoini

5 days ago

@nickcammarata facts, fuck parallelism

0 1 0 578 1

View Details

Alex @AlexanderMoini

5 days ago

stripe buys openrouter for: - customers - data on where tokens are going - bundling into stripe has interesting usecases like: - users of application layer products being able to see itemized llm spend by model in their bills from an app layer company. - byok setups - plenty plenty more.

0 0 0 15 0

View Details

Alex @AlexanderMoini

5 days ago

its not even model routing in the sense that you may think. there's a few things good about openrouter for the end user: - you have access to all the neoclouds & labs, for each model. because uptime mostly sucks, you need to have fallbacks to other providers. (i.e. openai -> azure hosted openai / fireworks -> togetherai -> cerebras) - you can easily add new models to your product without having to change code they now more recently have model routing based on the query itself in their fusion models, but this is circa mid '26.

1 0 0 21 0

View Details

Alex @AlexanderMoini

a week ago

@_AashishReddy kinda sounds like two kids on the playground discussing a fantasy. i read this with a lisp.

0 0 0 24 0

View Details

Alex @AlexanderMoini

a week ago

@colemurray i think they had me vesting for like 5, 10 years or something

0 0 7 2K 0

View Details

Alex @AlexanderMoini

2 weeks ago

@frantzfries "coincidentally"

0 0 1 74 0

View Details

Alex @AlexanderMoini

2 weeks ago

@suchenzang situationally aware

0 0 0 296 0

View Details

Alex @AlexanderMoini

2 weeks ago

@ArielKwiat as in there was a difference between that and the kaggle xgboost ensemblers

0 0 0 41 0

View Details

Alex @AlexanderMoini

2 weeks ago

@ArielKwiat feel like there's a big gap btwn rl and other ml. 2020 was the days of few-shot encoder-decoder finetunes, and SFT on t5-series models n stuff.

1 0 0 133 0

View Details

Alex @AlexanderMoini

2 weeks ago

@zekramu god forbid i have to do work

1 0 1 20 0

View Details

Alex @AlexanderMoini

2 weeks ago

@PhilipJohnston ngl dawg, kinda hard to imagine a guy building space data centers is sick from a bit of acceleration.

1 0 12 2K 0

View Details

Alex @AlexanderMoini

2 weeks ago

@shaunmmaguire With a small group of protesters in front, maybe.

0 0 1 658 0

View Details

Alex @AlexanderMoini

2 weeks ago

@var_epsilon lists the entirety of everything ever done at google

0 0 9 773 0

View Details

Alex @AlexanderMoini

2 weeks ago

@willcb @boardyai bill wrown btw if you're curious.

0 0 0 11 0

View Details

Alex @AlexanderMoini

2 weeks ago

always think about how much "ASSISTANT:" has defined how llms work. because they understand word semantically, and isn't purely a sep token between user messages, it almost certainly has heavily conditioned all llms sense of what they're supposed to be.