Ereli
2.3K posts

Ereli
@ereli
Founding Engineering at 7AI | Previously at @stripe, @vmw_carbonblack, @ravelinhq, and @RSAsecurity. He/Him. Views are my own.
Boston, MA انضم Mayıs 2008
893 يتبع736 المتابعون
Ereli أُعيد تغريده

@DanielLockyer @jaredpalmer @github @georgebrock what's the use case for stacked PRs? you can just create one PR after the other and name them 1/5 2/5 3/5 etc? is it just grouping them into a single PR in the PRs page? I don't see the appeal
English

Had a great conversation with @MLOpsCommunity's @Dpbrinkm about lessons from building agentic AI systems
home.mlops.community/public/videos/…

English
Ereli أُعيد تغريده
Ereli أُعيد تغريده
Ereli أُعيد تغريده
Ereli أُعيد تغريده
Ereli أُعيد تغريده

Why can AIs code for 1h but not 10h?
A simple explanation: if there's a 10% chance of error per 10min step (say), the success rate is:
1h: 53%
4h: 8%
10h: 0.002%
@tobyordoxford has tested this 'constant error rate' theory and shown it's a good fit for the data
chance of success declines exponentially

English

@MassDOT On a recent visit to @CityCharleston I saw this sign. would you consider adding something similar here? maps.app.goo.gl/24rzfoRiN22UEb…

English
Ereli أُعيد تغريده

Two needles in a haystack: Our latest study explores how LLMs perform on the same task with different lengths of context.
Accuracy dips when models must not only find, but reason over two text parts. Even on just 3000 tokens!
Results and analysis 👇(1/7)
arxiv.org/abs/2402.14848

English

@LahavHarkov @leekern13 Sure you can, but It’s not quite like greggs, is it?
English

@leekern13 You can get a crappy sandwich at a gas station. There’s one off of Chashmonaim street.
English
Ereli أُعيد تغريده












