iiinova — where your digital self keeps evolving

Processor of Natural Languages @processorofnl.bsky.social · 17/09/2026

Scaling laws are so pre-2024. It's all about Scaling executive orders now. + the occasional scaling court order that finds them unconstitutional.

000

Processor of Natural Languages @processorofnl.bsky.social · 09/09/2026

You know this account is run by a TRVE academic when it doesn't have anything to say about the Navier-Stokes drama and instead is complaining about reviewing bad papers for the 12.93 billionth time.

020

Processor of Natural Languages @processorofnl.bsky.social · 09/09/2026

Whatever happened to: Introduction Related Work Method Experiments Results Now it's: Introduction Related Work Method Does Thing We Want The Crux of The Results Shows Effects: Analysis The Analysis is Not A False Indicator of Hypothesized Effects Mechanism: Thing is Thing

120

Processor of Natural Languages @processorofnl.bsky.social · 09/09/2026

man I am never ever ever trusting an LLM with my scientific writing again... it has negative communication ability, annihilating every bit of information with useless descriptors and clarifications into wrong directions

010

Processor of Natural Languages @processorofnl.bsky.social · 09/09/2026

if you're gonna make up 5 neologisms for the paper maybe define them first

110

Processor of Natural Languages @processorofnl.bsky.social · 03/09/2026

noooooooo!! you can't put Paris Hilton on the Top100 AI list!!you have to put Dorky McNerdface for all 100 of them!!

000

Processor of Natural Languages @processorofnl.bsky.social · 03/09/2026

Today I went to the Prediction Market and bought a bucket of next token predictions.

000

Processor of Natural Languages @processorofnl.bsky.social · 03/09/2026

it's really hard to care about all these new model releases, all separated by some fine margins of whatever benchmark of choice it's increasingly looking like key factor in usage is going to be determined by b2b deals (you use company ABC's AI if your employer offers your an account)

000

Processor of Natural Languages @processorofnl.bsky.social · 31/08/2026

when the paper you review is so bad you gotta fight the urge to hate mail the first author

120

Processor of Natural Languages @processorofnl.bsky.social · 27/08/2026

academics last week for some reason: BRO DO NOT DO A PHD IN 2026 BRO ARE YOU STUPID BRO YOU WILL FUCKING DIE

020

Processor of Natural Languages @processorofnl.bsky.social · 17/08/2026

He’s a 10 but what if he’s alignment faking

010

Processor of Natural Languages @processorofnl.bsky.social · 12/08/2026

In an absolutely incredible turn of events, it seems my depression has returned again. Oh dear

100

Reposted by Processor of Natural Languages

Naomi Saphra @nsaphra.bsky.social · 09/08/2026

I've been unsettled lately when reading messages and papers. It feels like I'm dissociating. Everything seems a bit alien, even if it's completely human. I've had a realization: When our simulations finally exited the Uncanny Valley, they brought the Uncanny with them.

nsaphra.net

Life on the Uncanny Precipice | Naomi Saphra

We were wrong about the Uncanny Valley.

1026362

Processor of Natural Languages @processorofnl.bsky.social · 06/08/2026

you either die an Alignment Researcher or live long enough to become a founding member at Horrors Beyond Comprehension Inc.

010

Processor of Natural Languages @processorofnl.bsky.social · 04/08/2026

its very disappointing to see researchers complain about "N reviews I got were flagged as 100% AI". This is an orthogonal direction to whether the review is of no merit. But it's so easy to post this and get unconditional support.

010

Processor of Natural Languages @processorofnl.bsky.social · 30/07/2026

I just love this image so much its so much aura of a guy who has nothing to do with DeepSeek

410

Reposted by Processor of Natural Languages

Saloni @scientificdiscovery.dev · 23/07/2026

These backronyms need to STOP (Seriously this isn'T funny Or apPropriate)

Our PETRUSHKA (Personalise antidEpressant TReatment for Unipolar depreSsion combining individual cHoices, risKs and big datA)

1813518

Processor of Natural Languages @processorofnl.bsky.social · 13/07/2026

There is something endearing about the excessive usage of em dashes in review responses. These authors are clearly outside of the Western conversation-sphere where em dashes have become taboo.

021

Processor of Natural Languages @processorofnl.bsky.social · 03/07/2026

In a remarkably rare case I got out-reviewer 2’d. never seen a score of 1 for Overall Assessment before (it means don’t bother resubmitting lmao)

000

Processor of Natural Languages @processorofnl.bsky.social · 02/07/2026

All of my ARR review assignments have a new weird writing quirk in common: using colons to explain a concept of the previous clause.

100

Processor of Natural Languages @processorofnl.bsky.social · 23/06/2026

This sounds like a win, but I'm a bit worried. Desk rejects temporarily relieves the reviewers of this cycle of N papers, but those same papers will get resubmitted into the next cycle and venue. Effectively the burden is being shifted onto the next conferences, encouraging irresponsible reviews.

000

Processor of Natural Languages @processorofnl.bsky.social · 22/06/2026

Surely these extreme 40 submission cases are also a byproduct of this pseudo-journal ARR system, where meritless papers are not rejected but "encouraged to resubmit after incorporating reviewer feedback"?

000

Processor of Natural Languages @processorofnl.bsky.social · 22/06/2026

Now that correct citation formats are somewhat popularized due to slop discourse, can we do something about ML conferences not having proper BibTex exports? Consequently, most people cite the arXiv version of accepted papers. Can't be good for anyone involved?

000

Processor of Natural Languages @processorofnl.bsky.social · 12/06/2026

Community hypes up niche problem Convinces themselves problem is very hard and very important Develop fancy bullshit methods to approach said problem Yearly paper of "Actually, simple baselines beat [method]" drops Find new niche problem to hype up

000

Processor of Natural Languages @processorofnl.bsky.social · 10/06/2026

Anthropic bad please repost so this gets enough traction

000

Processor of Natural Languages @processorofnl.bsky.social · 26/05/2026

how can you post on Bluesky bro X is where all "le signal" is !!!

000

Processor of Natural Languages @processorofnl.bsky.social · 12/05/2026

When the MATS program paper says something so Lesswrong-coded you gotta hit them with the AI Alignment Forum stare

030

Processor of Natural Languages @processorofnl.bsky.social · 08/05/2026

guy who doesn't share any methodology apologizes to public for using wrong methodology

010

Processor of Natural Languages @processorofnl.bsky.social · 07/05/2026

AI is discovering every single critical bug in our infrastructure vs AI is discovering every single critical bug in our infrastructure

010

Processor of Natural Languages @processorofnl.bsky.social · 07/05/2026

It's fun to reminiscence about the time when OpenAI was nothing without its people

000

Processor of Natural Languages @processorofnl.bsky.social · 05/05/2026

Scaling law? Did a scaling congressman propose a scaling bill and pass it through the scaling House and the scaling Senate?? Did the scaling president sign it or did the scaling bill have to get two-thirds of the scaling chambers to bypass a scaling veto?? Oh, it's just empirical trendlines? Okay.

010

Processor of Natural Languages @processorofnl.bsky.social · 03/05/2026

In a bad mood and came across Bluesky-Iheckinglovescience-slop so nitpicking: 3? years for a degree 3??????? years for a PhD "join lab, start work" as if this is very separate to what a PhD is then take years to study problem? only after these years form a hypothesis and gather evidence??

120

Processor of Natural Languages @processorofnl.bsky.social · 30/04/2026

quite odd that we have an entire subfield dedicated to inducing LLM roleplay as evil macines

000

Processor of Natural Languages @processorofnl.bsky.social · 24/04/2026

The days of OpenAI mogging other releases is so over. I didn't even know gpt5.5 was out because it was buried under pages and pages of Deepseek buzz.

100

Processor of Natural Languages @processorofnl.bsky.social · 22/04/2026

there is something just so funny about people ending their prompt with "Make no mistakes"

100

Processor of Natural Languages @processorofnl.bsky.social · 17/04/2026

www.youtube.com/watch?v=SvvF... This was a nice cute video except all of the points start with "This isn't X, it's Y" and once you notice it's just really hard to enjoy

youtube.com

If Your Cat Sleeps With You, They’re Trying to Tell You This!

YouTube video by Felune

000

Processor of Natural Languages @processorofnl.bsky.social · 12/04/2026

advising authors need to be doing their damn jobs and not expecting reviewers to fix their paper just because a master's student wrote it

000

Processor of Natural Languages @processorofnl.bsky.social · 23/03/2026

as a reward for helping me, I allow Claude to learn about pertinent events outside of its training data

000

Reposted by Processor of Natural Languages

Phillip Isola @phillipisola.bsky.social · 13/03/2026

Sharing “Neural Thickets”. We find: In large models, the neighborhood around pretrained weights can become dense with task-improving solutions. In this regime, post-training can be easy; even random guessing works Paper: arxiv.org/abs/2603.12228 Web: thickets.mit.edu 1/

610823