iiinova — where your digital self keeps evolving

ruggsea @ruggsea.eurosky.social · 19/10/2026

Integrate it into your projects! Use it to test funny or important hypotheses on trends in journalism and online discourse! Tell your friends who need news datasets for their research. Good friends don't let their friends pay for Factiva when a free one might do the job! infini-news.uni-graz.at

010 1

ruggsea @ruggsea.eurosky.social · 19/10/2026

We made a second version of the dataset/api we made to solve news datasets collection/checking word frequencies across basically all the world's news from the last 10 years! Try it, it's fun! To show people how cool this resource is, I had Opus make a hype video for it.

250 5

ruggsea @ruggsea.eurosky.social · 16/10/2026

a beautiful but rare sight, I imagine, in mech interp: the feeling of doing science

030 3

ruggsea @ruggsea.eurosky.social · 15/10/2026

this happens because of RLHF, because 7 is human mode and this kind of RL collapses the outputs in greedy, mode leaning directions

030 3

Reposted by ruggsea

Mark Riedl @markriedl.bsky.social · 14/10/2026

Thank you Copilot, yes this is exactly the slide template I should be using when lecturing about One Hot Vectors in my intro to AI class

4934 93

ruggsea @ruggsea.eurosky.social · 08/10/2026

sorry what is a non-generative LLM

100 1

ruggsea @ruggsea.eurosky.social · 08/10/2026

Doing research in that field , I believe that (1) training models to learn those general patterns is necessary and (2) this has not been done in a systematic and scaled way yet But I feel like some colleagues without familiarity with LLMs just assume them to magically work for this, off the shelf

000 0

ruggsea @ruggsea.eurosky.social · 07/10/2026

this is kinda the argument Fisher makes here hersephoria.com/files/books/...

hersephoria.com

000 0

ruggsea @ruggsea.eurosky.social · 03/10/2026

How those models are trained is not a secret. There are open source models with very accurate technical reports detailing how they were trained and repository containing the code and data they used. I trained one (although small) yesterday even

0100 10

ruggsea @ruggsea.eurosky.social · 26/09/2026

currently trying to check that but it's a bit difficult to build a pipeline to detect things it is probably difficult to guess the shape of

000 0

ruggsea @ruggsea.eurosky.social · 25/09/2026

thanks! Opus 5.5 chose the picture

011 1

ruggsea @ruggsea.eurosky.social · 25/09/2026

Currently working on promotional materials for an LLM course me and @lorrenz.bsky.social will teach next month, is this image making you want to enroll in it?

2130 13

ruggsea @ruggsea.eurosky.social · 25/09/2026

"I've seen datasets you people wouldn't believe..."

100 1

ruggsea @ruggsea.eurosky.social · 24/09/2026

tracing llm behaviours to training data is a novel (and difficult!) field of research so I wouldn't say that everybody can discuss it just fine. Also, a lot of their behaviour now comes from a second training (post training) that incentivizes behaviour by outcome, not by imitating training data

000 0

Reposted by ruggsea

ruggsea @ruggsea.eurosky.social · 23/09/2026

Currently registering for #EMNLP2026 next month in Budapest! Is anybody else that follows me going? Btw thanks to @eugenevinitsky.bsky.social for making a conference page for it on lea.ac!

lea.ac

Lea - A calmer Bluesky for researchers

A curated garden for researchers on the AT Protocol

132 3

ruggsea @ruggsea.eurosky.social · 23/09/2026

Currently registering for #EMNLP2026 next month in Budapest! Is anybody else that follows me going? Btw thanks to @eugenevinitsky.bsky.social for making a conference page for it on lea.ac!

lea.ac

Lea - A calmer Bluesky for researchers

A curated garden for researchers on the AT Protocol

132 3

ruggsea @ruggsea.eurosky.social · 24/09/2026

Guitar chords are low perplexity, apparently

110 1

ruggsea @ruggsea.eurosky.social · 21/09/2026

how do you know that other people do and they do not?

100 1

ruggsea @ruggsea.eurosky.social · 17/09/2026

have you tried using few shots?

000 0

Reposted by ruggsea

Vladimir Salnikov @v4ldelund.bsky.social · 16/09/2026

"just look at the data" final boss

316222 162

ruggsea @ruggsea.eurosky.social · 15/09/2026

This... sounds a lot like reasoning works in general? Also the paper is not about reasoning models, it's about Llama 8B Instruct bsky.app/profile/docv...

040 4

ruggsea @ruggsea.eurosky.social · 15/09/2026

I am a bit confused of why he would be scared of censorship when talking about the first option. Isn't what he was hypothesizing the theoretical endgame, at least nominally, of the Chinese way to socialism?

010 1

ruggsea @ruggsea.eurosky.social · 15/09/2026

People are discussing fully automated communism on main. In China. Cf: the west

110 1

ruggsea @ruggsea.eurosky.social · 14/09/2026

If you're 20-30 & your main circle isn't discussing: - how to make bioweapons with GPT-6 Astra Then it's time to find a new circle

053 5

ruggsea @ruggsea.eurosky.social · 11/09/2026

We need this but for most of interdisciplinary research areas (and with interedisciplinary I mean the obvious something something + computer science/machine learning)

020 2

ruggsea @ruggsea.eurosky.social · 11/09/2026

about #4, from a Micheal Levin lecture:

001 1

ruggsea @ruggsea.eurosky.social · 10/09/2026

deepseek's engram and Qwen's n-gram layer are not the same thing at all and my brain really itches badly cause I have seen people confuse them, rightfully so

010 1

ruggsea @ruggsea.eurosky.social · 09/09/2026

How would you define intelligence? Pardon me for answering a question with a question but I assume your definition will be stricter than mine so I would be curious about that

000 0

ruggsea @ruggsea.eurosky.social · 09/09/2026

I use they as a generic plural, as a plural "it"

000 0

ruggsea @ruggsea.eurosky.social · 09/09/2026

Do they lack intent? Yes most likely Do they have feelings or internal states? No or at least not in the same way humans do. Do they possess or exhbit something akin to intelligence? I find it difficult to deny this given current evidence (even taking away evidence from commercial AI labs)

100 1

ruggsea @ruggsea.eurosky.social · 09/09/2026

and misused terms to market and/or explain their invetions/product. But this remains a critique of their enterprise from a semantical angle no? Going past the use of words, I still think that if intelligence can be ascribed to non human entities, the current models clearly exhibit it.

200 2

ruggsea @ruggsea.eurosky.social · 09/09/2026

Oh I also read the Turing paper last year and I was baffled by his claim that Muslim believe that women don't have a soul. I actually remember really trying to find sources for it. Getting back to the argument, I continue to agree that the tech sphere and key figures of it have exploited fiction..

310 3

ruggsea @ruggsea.eurosky.social · 09/09/2026

I do think his answer kinda makes sense with how neural networks are trained + it seems compatible with the philosophical theory I mentioned. But it implies consciousness, something not necessary for intelligence per se. It's easier to argue that llms are undeniably getting more and more intelligent

100 1

ruggsea @ruggsea.eurosky.social · 07/09/2026

It really seems to me that, putting aside tech marketing attempts and lingo, such machines really do perform and/or are improving in some aspects of "intelligence" + there is a philosophical framework that just descriptively claims that computation can give rise to mind like things

100 1

ruggsea @ruggsea.eurosky.social · 07/09/2026

I listened to that excerpt of the talk! I do find fascinating to follow the thread of cultural references and fictional topoi employed by the tech industry (and related questionable people) but I still think that it's not addressing the core matter: why a computational system cannot "grow smarter"

300 3

ruggsea @ruggsea.eurosky.social · 07/09/2026

to be transparent: i don't think quantifying things is the most important part of the human mind, nor that the quantifiers should rule. But i do think, descriptively, that a pile of mathematical calculations could be a mind and therefore maybe "grow more intelligent"

000 0

ruggsea @ruggsea.eurosky.social · 07/09/2026

Wait I think we are speaking past each other! I do not mean normative claims about calculation (which part of the mind is "best", who should rule), but the descriptive idea that complex enough computations can basically be a mind and/or that a mind is equivalent to a computer in certain aspects

210 2

ruggsea @ruggsea.eurosky.social · 07/09/2026

Are you familiar with the computational theory of mind? plato.stanford.edu/entries/comp... Would you suggest that people that subscribe to it do confuse facts with fiction?

plato.stanford.edu

The Computational Theory of Mind (Stanford Encyclopedia of Philosophy)

100 1

ruggsea @ruggsea.eurosky.social · 07/09/2026

I find it super funny thinking that at some point, in the early history of the PRC, there is this key conference where all the people that will be super important later on are discussing together with a guy who speaks German and his interpreter

000 0

ruggsea @ruggsea.eurosky.social · 07/09/2026

I think it's crazy that at some point they kinda let him lead the Chinese Communist military efforts and he lowkey fumbling it was apparently a key factor in Mao taking power

100 1

ruggsea @ruggsea.eurosky.social · 07/09/2026

His early life includes things like escaping from prison with the help of his Communist lover to then go to Moscow and joining the Soviet school for international agents

100 1

ruggsea @ruggsea.eurosky.social · 07/09/2026

His underground name in Chinese was 李德, Lǐ Dé, something that apparently sounds like "Li the German" but also "Li the virtuos" (according to Kimi and @hailina.bsky.social that speaks Chinese) and was apparently so good at concealment that his real identity was revealed only in 1964

110 1

ruggsea @ruggsea.eurosky.social · 07/09/2026

I got nerdsniped by the baffling existance of a new type of guy: a man born in Bavaria under the German Empire that was so good as Communisty party member that he was sent to China to assist the Soviet republic and ended up beefing with Mao on military strategy en.wikipedia.org/wiki/Otto_Br...

en.wikipedia.org

Otto Braun (communist) - Wikipedia

130 3