@realmcore_ Profile picture

akira

@realmcore_

oss code gen

Similar User
Luca photo

@br3akthematr1x

Kristie McNealy photo

@KristieMcNealy

carmocamargo photo

@ACarmocamargo1

Pinned

Hey everyone. Even though we’ve shipped fast, the tool is nowhere near good enough. There’s a reason I don’t use it daily. We figured it out. Now to build it. See you in a few weeks. LFG 🫡


Alright chat, we got this shit. End to end testing tomorrow, and then private beta starts rolling out! If you are on the beta list, watch out for a message from me ;)


For what it’s worth it works

i started learning rust 3 hours ago shipping rust code to prod on a friday LFG get owned scrumlords



i started learning rust 3 hours ago shipping rust code to prod on a friday LFG get owned scrumlords


akira Reposted

i have interviewed >100 people for this research engineering role at morph labs and have made an offer to and hired only 1. we have headcount for 1 more - our goal is to bring the next billion autonomous SWEs online with frontier-level (and beyond) evals and post-training for AI…

i am hiring a full-time research engineer to work on creating superhuman quality synthetic data for evaluating and post-training AI SWEs. our goal is to bring the next billion autonomous AI SWEs online as quickly as possible. scalable and trustworthy evaluation and post-training…



akira Reposted

The average CS student just doesn’t know how to program it’s so sad


For context I already tried the minecraft game. It’s incredible. The point I was making was that it’s so good in fact a lay person puts it in the same bucket as the real deal

I don’t think you guys understand how insane it is that people consider this to be CATEGORICALLY EQUIVALENT to minecraft literally unbelievable that in peoples heads it’s the *same* thing



I don’t think you guys understand how insane it is that people consider this to be CATEGORICALLY EQUIVALENT to minecraft literally unbelievable that in peoples heads it’s the *same* thing

For the low low price of Manhattan's entire power grid, you can now play Minecraft but without any consistent bitrate and no object permanence!



akira Reposted

ANNOUNCING: Windows Support for Aphrodite (Experimental) [Installation Guide] github.com/PygmalionAI/ap… Enjoy your extremely optimized, high-throughput local AI, courtesy of @AlpinDale Please pay your respects by retweeting, telling your friends, or creating and open-sourcing…

Experimental Windows support has landed in Aphrodite. You can now use a performant implementation of PagedAttention on Windows w/ NVIDIA GPUs! It's still in the testing phase and there might be issues, but preliminary benchmarks show no performance degradation compared to Linux.

Tweet Image 1


akira Reposted

they do this thing called the putnam competition, it's like leetcode but everyone fails


[x dot com the everything app] has one shorted the entire middle of the bell curve that makes up 98% of the population and they are completely oblivious to it.

Instagram has one shotted the entire middle of the bell curve that makes up 98% of the population and they are completely oblivious to it.



akira Reposted

Introducing Oasis: the first playable AI-generated game. We partnered with @DecartAI to build a real-time, interactive world model that runs >10x faster on Sohu. We're open-sourcing the model architecture, weights, and research. Here's how it works (and a demo you can play!):


Cool yeah so we now have it hard coded, going to parameterize an eval harness. Should be able to help people set up evals for various tasks after I figure this out.

CLAUDE IS WRITING EVALS AND THEY’RE EXTRAPOLATING PAST THE PROVIDED DIFFICULTY WITHOUT HALLUCINATION we stay winning now I can bootstrap an eval set basically for FREE



vaguepoasting is back on the menu


ewma is so so good (see d2v2 by FAIR)


If an autocomplete tool can one shot your code, the code you’re writing is too in-distribution This is a subtweet, take it personally


The only reason I use types in python is for autocomplete


1.5 years tops

i cant comprehend people thinking AGI in less than 15 years 4 years since GPT3, and we have *mostly* gotten better engineered and optimized models RL to boost decision making (beyond naively better preference optimization) is starting to make agents work, but just barely i…



United States Trends
Loading...

Something went wrong.


Something went wrong.