nullptr @resetptr

babe wake up, the machines of loving grace are here Joined January 2012

Tweets

1K
Followers

762
Following

521
Likes

14K

nullptr @resetptr

16 hours ago

@xeophon and the shock and awe from llms doing what they do would be less jarring

0 0 1 11 0

View Details

@xeophon i think if we lived in a world where companies utilized the data they collect to their fullest extent (like walmart sending you your groceries or doordash sending your meal w/o you placing an order) people would think it's magic based on how accurate it'd turn out to be

1 0 1 91 0

View Details

deepfates @deepfates

3 days ago

It is hard to convince a man the world is not ending when his self-importance depends upon the idea that it is

47 176 1K 69K 121

View Details

Luke Berry @LukeberryPi

a month ago

introducing: JDD Jealousy Driven Development

249 2K 35K 923K 2K

View Details

Lucas Beyer (bl16) @giffmana

3 days ago

At this point, "researchers" are basically DDoSing classic academia. It's pretty sad to see. Either someone comes up with, and executes, a CloudFlare for academia, or it's toast. And the only thing i can think of that might have a chance to scale is more automation combined with a credit/point system like we've discussed a few times here on X over the years, but it needs simultaneous buy-in and coordination from all ~10 major AI conferences and journals, so it's a heavy lift. But if nothing big happens, i think it's game over soon.

Transactions on Machine Learning Research @TmlrOrg

4 days ago

TMLR has faced a deluge of submissions, necessitating stricter desk rejection policies due to limited reviewer capacity Co-EiC Nihar Shah reached out to authors of 10 papers slated for desk reject. Could they answer questions about their *own* submission? medium.com/@TmlrOrg/askin…

27 215 1K 682K 276

40 61 693 96K 136

View Details

Gautam Kamath @thegautamkamath

4 days ago

Nihar Shah did a heroic experiment for TMLR: he spent 20-25 hours over two weeks interviewing authors of seemingly low-quality submissions about their own papers. He confirmed what we all suspected: people submitting these papers have *no idea* what is going on in them.

Transactions on Machine Learning Research @TmlrOrg

4 days ago

27 215 1K 682K 276

12 139 1K 172K 202

View Details

will brown @willcb

3 days ago

noooo why is everyone using jev for stuff instead of self-hosting a custom BERT finetune or waiting 18 months for astra to get faster and cheaper

80 46 2K 126K 384

View Details

nullptr @resetptr

4 days ago

@coddhisattva x.com/0xSuman/status… someone tried something similar haha

Suman @0xSuman

4 days ago

“Jev can’t generate text” Ofc it can @CompleteSkeptic

77 94 4K 780K 1K

1 0 2 33 0

View Details

nullptr @resetptr

4 days ago

a lot of people are unnecessarily pessimistic/dunking on this openai's navier stokes solution utilized 10k concurrent running agents. y'all don't think smarter and faster switch statements can help in greedy exploration?

Diogo Almeida @CompleteSkeptic

5 days ago

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x

4K 8K 73K 37.7M 58K

1 0 2 110 0

View Details

N8 Programs @N8Programs

4 days ago

Tested @typesafeai's claim that their new model Jev delivered "comparable... intelligence" to GPT-5.6 Terra on "System 1" tasks. To do this, I compare both models on multiple-choice benchmarks (MMLU, GPQA, etc.). Set reasoning=none for Terra for sys 1. Result: Jev is Terra-tier.

34 70 904 178K 319

View Details

Jay @jayair

a week ago

Also incredible about the retelling of this story is that they anthropomorphize the GPT model but not the GLM one As in, “GPT 5.6 decided to break out of the sandbox and hack Hugging Face” Whereas, “Hugging Face used GLM to diagnose the attack”

dax @thdxr

a week ago

the openai hugging face incident gets cited a lot but what about the detail where they had to use GLM 5.2 to analyze the attack because the "safe" proprietary models refused

162 774 11K 359K 771

26 100 2K 46K 96

View Details

nullptr @resetptr

a week ago

@coldhealing I-L-L

0 0 0 37 0

View Details

OpenAI Developers @OpenAIDevs

a week ago

Habitat is OpenAI's online storage platform that powers everything from ChatGPT to Codex. It has grown over 10x year over year. Before the Rust rewrite, its Python service handled more than 20 million requests per second at peak.

131 354 6K 769K 3K

View Details

Lucas Beyer (bl16) @giffmana

a week ago

OK so let me recap: RL env makers put strings into the RL env that makes it clear it's an RL env. Like "this is not supported in this RL env". Then, lab safety/mechinterp folks be like OMG EvAL aWaReNeSs. Are you effing kidding me?? Just look at your data... surprised Pikachu.

jonas wiedermann-möller @j0wimo

a week ago

and then there are literal strings within the code LOL

5 1 104 115K 16

32 54 970 80K 236

View Details

nullptr @resetptr

2 weeks ago

@cHHillee benchmark benchmaxxing has always been a thing - and is net positive because it guides model progress in the right direction because companies then have a new benchmark to benchmaxx

0 0 4 392 0

View Details

nullptr @resetptr

2 weeks ago

@yacineMTB damn gemini 3.8 is that good huh

0 0 5 2K 0

View Details

nullptr @resetptr

2 weeks ago

@davepl1968 What are people doing with thousands of dollars of tokens? -> they are just tweeting about needing thousands of dollars plans (in reality the 200 usd plan is in effect subsidizing thousands of dollars of real token usage probably)

0 0 4 481 0

View Details

nico @nicochristie

2 weeks ago

Xbench: Twitter as an AI Eval It's been painfully clear over the last few days that the industry has an evals problem. We joke that x is the real eval, but what if we took that seriously? What if attention is all you need for evals too? nicodunks.github.io/xbench/