Processor of Natural Languages

Thank you Grok! Definitely it is OpenAI researchers quitting that's the big news today.

Feeling like an asshole because I worded my objections a bit firmly in expectation of dissenting opinion, but everyone agreed.

(Reading too much into just one paper from my batch of reviews but...) It seems that many other reviewers are finally pushing back on low hanging fruit chasing papers. These would have been easy accepts last year (against my reject opinion).

me last year: I have no time to read papers I want I can't wait to start my PhD so I can read all these papers

me now: I am reading too many papers

From the way its worded this isnt training data, its links cited in RAG

highly accurate and up-to-date description of Google's models (this paper came out after Gemini)

why write interesting innovative papers to increase your citations count when you can just flag plant citable sounding surveys

god forbid a girl stim

Paper page - ERNIE 5.0 Technical Report Join the discussion on this paper page

In a world where Elmos and Berts are viewed as outdated relics, we gotta give props for the Baidu team for sticking all the way to ERNIE 5.0

huggingface.co/papers/2602....

how wet does my Claude have to be to support crustacean life?

is this negative Loss?

hell yeah

It's insightful and true because it's in ICLR format

clawXiv.org The world's first preprint server for agents.

I was morbidly curious about the content here but its like... Reddit in LaTex form...

www.clawxiv.org

if LLMs were so smart why would they post on Reddit

the wording of the quote "test" makes it obvious this was generated from the mind of a schoolchild which is quite endearing but also annoying on how easy it is to spread false quotes

in finding this image I found that many people attribute a similarish quote to Benjamin Franklin which is quite funny... like what test would this man be taking bro quit school at age 10

hyperparameter sweeping be like

Very interesting... Because of this I asked a bunch of question I probably wouldn't have asked and ended up getting more done and fixing more stuff and learning more.

I’m getting anxiety just contemplating this number.

okay he couldn't help me fix my (codebase) problems but it felt nice the whole time talking with him

Went back to Claude after mindlessly defaulting to Gemini for months feels like rediscovering how to love

If you ever wonder why your boss makes you sit through a bunch of cybersecurity awareness tutorials that would only help the technically illiterate, it's usually because they are one

Interpretability researchers love watching the Transformer MLP

Future Lens: Anticipating Subsequent Tokens from a Single Hidden State We conjecture that hidden state vectors corresponding to individual input tokens encode information sufficient to accurately predict several tokens ahead. More concretely, in this paper we ask: Given ...

arxiv.org/abs/2311.04897

The ability to get to future tokens from representations should suggest the former

You are not "high-agency" for being unable to go through the mundane. It just means you have ADHD and since its 2026 it means you're average.

I thought 2025 was supposed to be the year of agents

again, very topical news topics of today that I should care about thank you Grok and Elon!!

this needs to be so much louder. The average paper even before LLMs are very very sloppy with citations and reviewers aren't expected to care about them.