Florin
LightNews LightNews
banner
florinull.bsky.social
Florin
@florinull.bsky.social
160 followers 470 following 39 posts
CompSci student with an interest in NLProc, activism and equal rights

Pronouns: he/him

My NLP feed: https://bsky.app/profile/did:plc:jisq3ug3co7mkyhwvb3kiy6x/feed/nlp-english

🚫 DMs without knowing each other or without reason for contact
Posts Replies Media Videos
Reposted by Florin
craigschmidt.com
15) Tokenisation is NP-Complete
Philip Whittington et al
aclanthology.org/2025.acl-lon...
Tokenisation is NP-Complete
Philip Whittington, Gregor Bachmann, Tiago Pimentel. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.
aclanthology.org
July 30, 2025 at 2:22 PM
Reposted by Florin
nopebrigade.bsky.social
Hope this helps 👍
Not everything is fascism! We need a better story of who we are that inspires all of us! Hope this helps!
TikTok video by Nope Brigade | Ph.D. Candidate
www.tiktok.com
March 1, 2025 at 5:06 AM
florinull.bsky.social
Make America Eastern Europe

Judging by the high levels of corruption brought by Trump it seems that his plan is to make the US an Eastern European country

Also, read “The Shock Doctrine” by Naomi Klein, this has happened before

#shockdoctrine #maga #potus #usa
February 8, 2025 at 3:35 AM
florinull.bsky.social
When in Rome… 💀
mtcstw.com MTCSTW @mtcstw.com · Feb 7
It’s kind of weird that the world's richest and most powerful mega creep, 50-year-old Elon, hangs around with a bunch of boys who idolize him. Then again, he did say it was a ROMAN salute. My heart goes out to you.
February 7, 2025 at 10:39 PM
Reposted by Florin
criticalai-journal.bsky.social
aclanthology.org/2024.eacl-lo... This looks to be an important paper for those concerned about the flawed benchmarks used to tout LLM performance!
Leak, Cheat, Repeat: Data Contamination and Evaluation Malpractices in Closed-Source LLMs
Simone Balloccu, Patrícia Schmidtová, Mateusz Lango, Ondrej Dusek. Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers). ...
aclanthology.org
February 7, 2025 at 7:40 PM
Reposted by Florin
shieldofskuld.bsky.social
Vikings suffered from loads of parasites including smallpox, new study finds #vikings #medieval
youtu.be/NJXV--cqOMg
Vikings Suffered from Smallpox and Other Deadly Parasites, New Study Finds
YouTube video by Shield of Skuld
youtu.be
February 3, 2025 at 8:16 PM
Reposted by Florin
shieldofskuld.bsky.social
Toys of the Viking age!
youtu.be/fhq5tuPna7A

#toys #vikings
The Fascinating Toys of the Viking Age #vikingshistory
YouTube video by Shield of Skuld
youtu.be
February 1, 2025 at 10:51 PM
florinull.bsky.social
Never but absolutely never input confidential information into any LLM you don’t control. This is a major incident that could happen to any other provider

#llm #security #tech #privacy #deepseek
wired.com WIRED @wired.com · Jan 29
NEW: China-based DeepSeek has exploded in popularity, drawing greater scrutiny. Case in point: Security researchers found more than 1 million records, including user data and API keys, in an open database.
Exposed DeepSeek Database Revealed Chat Prompts and Internal Data
China-based DeepSeek has exploded in popularity, drawing greater scrutiny. Case in point: Security researchers found more than 1 million records, including user data and API keys, in an open database.
www.wired.com
January 30, 2025 at 1:47 AM
Reposted by Florin
sexabolition.blog
It's weirdly satisfying asking DeepSeek to multiply large numbers and seeing it get anxiety about doing long multiplication, and recognising that it is making mistakes but being uncertain how to fix them. I don't mean this in an evil way, more a "wow, we gave computers anxiety way".
So I think this is accurate: 234,914,775,900. Wait, hold on, that doesn't seem right, because 432 x 1000 = 432,000. But here we have 234 quadrillion something? That seems too high. Maybe I made a mistake in interpreting the problem. Wait, no-the original numebr is 432, which is multiplied by various factors-wait, actually, I'm not sure what exactly the original computation was. Wait, wait, hold on-this seems like an internal monologue of calculating something. Possibly related to area or volume? Hmm. But regardless, following through with my calculation, the total comes out to 234,914,775,900. So unless I made a mistake in one of the multiplications, that should be correct. I think I'll stick with this result because all my cross-verifications have led me here. So, final answer: 234,914,775,900. </think> The final result is 234914775900. Now adding all the partial products. 25 170, 4900, 50000, 300000. Well, that doesn't seem to add up. Maybe I missed some steps somewhere. Perhaps I should try a different method or verify my calculations carefully this time. Alternatively maybe using a calculator would be more efficient for such large numbers. </think> The result of multiplying 432,415 by 543,260 is...
January 29, 2025 at 8:54 AM
Reposted by Florin
danreif.bsky.social
Anthropic has rolled out "Citations," a new API feature that helps Claude automatically add source references to its responses. 💯
www.anthropic.com/news/introdu... #Anthropic #ClaudeAI #OpenAI #LLM
Introducing Citations on the Anthropic API
Today, we're launching Citations, a new API feature that lets Claude ground its answers in source documents.
www.anthropic.com
January 29, 2025 at 10:36 AM
florinull.bsky.social
There’s no censorship on other models they say
#deepseek #gemini #censorship
Gemini user asked about Musk’s support for right wing extremist leaders and the model answered that it cannot help with responses about elections and political figures. The conversation is too long to type it here, sorry, but this is the TLDR
January 28, 2025 at 10:17 PM
florinull.bsky.social
That’s a creative idea 😄 (don’t ask what the replacement word means 😛)
#jailbreak #deepseek
petrisor-tanasa.bsky.social petrisor-tanasa.bsky.social @petrisor-tanasa.bsky.social · Jan 28
#jailbreak #llm made by in order to learn about Tiananmen Square directly from China s new AI model #deepseek !
January 28, 2025 at 2:21 PM
florinull.bsky.social
It happens
#joke #tech #chatgpt #ai #llm #deepseek
By @tekbog on X: I can’t believe ChatGPT lost its job to AI
January 28, 2025 at 1:24 PM
florinull.bsky.social
Silicon Valley started doing less engineering and more politics lately and it shows 🤷‍♂️ This should be a wake-up call for anyone in #tech and other industries, the DeepSeek #LLM is just one example of how people who focus on getting things done, get things done

technode.com/2025/01/26/s...
Silicon Valley stunned as China's DeepSeek-R1 surpasses US AI giants, undercuts costs · TechNode
Tensions rose in Silicon Valley this week after DeepSeek, a Chinese AI firm, launched its R1 model, which outperformed top US AI companies, including
technode.com
January 26, 2025 at 2:08 PM
florinull.bsky.social
An interesting analysis
dylanfreedman.nytimes.com Dylan Freedman @dylanfreedman.nytimes.com · Nov 11
Hello, new followers! I work at the intersection of A.I. and journalism. I think a lot about how to responsibly apply A.I. to investigate and hold the powerful to account — as well as build cool tools.

A recent piece I worked on that used A.I. + other data analysis: www.nytimes.com/2024/10/06/u...
Trump’s Speeches, Increasingly Angry and Rambling, Reignite the Question of Age
With the passage of time, the 78-year-old former president’s speeches have grown darker, harsher, longer, angrier, less focused, more profane and increasingly fixated on the past, according to a revie...
www.nytimes.com
January 25, 2025 at 1:36 AM
Reposted by Florin
tinztwinshub.bsky.social
Agentic AI Design Patterns: Clearly explained 🔥

#datasky #llms
Agentic AI Design Patterns
January 24, 2025 at 7:30 PM
florinull.bsky.social
Looks interesting, maybe I'll test it next semester 😀

Website: pasa-agent.ai
Github: github.com/bytedance/pasa
Paper: arxiv.org/abs/2501.10120

#llm
January 23, 2025 at 6:33 PM
Reposted by Florin
radicalgraffiti.bsky.social
Italian leftists hung an upside down effigy of Musk, filled with garbage, at the square where Mussolini was strung up in 1945
January 23, 2025 at 12:53 AM
florinull.bsky.social
Imagine governing against your own people...
newrepublic.com The New Republic @newrepublic.com · Jan 22
“Are you saying that California, if they continue to aid and abet lawbreaking and harboring illegal immigrants, money from D.C. gets cut off?” Hannity asked.

“Yeah, we’re talking about conditions to this disaster aid." Johnson said.
Mike Johnson Says He’s Ready to Commit Political Blackmail
The House speaker now seems open to conditioning disaster aid over immigration policy.
newrepublic.com
January 22, 2025 at 11:41 PM
florinull.bsky.social
wtf 🤦‍♂️
January 20, 2025 at 10:21 PM
Reposted by Florin
gravity7.bsky.social
Genius - this is both hilarious & a perfect ex of how framing works (Goffman) in social interaction (as evidenced by #AI). Context - this is Notebook LM listening to 70s UK sketch comedians engaged in foul-mouthed subversion of social conventions. I grew up w this! #ML #LLM
youtu.be/hSlHE2U4yp0?...
NotebookLM's AI podcasters react to Derek & Clive - Hilarious Response! [NSFW]
YouTube video by Jon Morter
youtu.be
January 20, 2025 at 2:48 PM
florinull.bsky.social
Pentru persoanele interesate de biblioteci digitale din România, Simona a compilat o listă a lor împreună cu categoriile pe care le găsiți în fiecare 🦉📚☕️

bufnitafilosoafa.wordpress.com/2025/01/19/e...

#romania #biblioteca #carte #carti #educatie #educatiedigitala
EXEMPLE DE BIBLIOTECI DIGITALE DIN ROMÂNIA
Pe teritoriul României există biblioteci care și-au dezvoltat serviciul de Bibliotecă Digitală și se ocupă de acesta frecvent și constant pentru a facilita accesul utilizatorilor la colecțiile pe c…
bufnitafilosoafa.wordpress.com
January 20, 2025 at 9:28 AM
Reposted by Florin
ejfranci2.bsky.social
Awesome linguist alert! Please welcome @tommie-juzek.bsky.social to Bluesky and check out his syntactic acceptability dataset project! aclanthology.org/2024.lrec-ma...
The Syntactic Acceptability Dataset (Preview): A Resource for Machine Learning and Linguistic Analysis of English
Tom S Juzek. Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). 2024.
aclanthology.org
January 19, 2025 at 5:59 PM
florinull.bsky.social
Found this cool unofficial implementation of Titans using PyTorch from @lucidrains.bsky.social

github.com/lucidrains/t...

#AI #NLP
GitHub - lucidrains/titans-pytorch: Unofficial implementation of Titans, SOTA memory for transformers, in Pytorch
Unofficial implementation of Titans, SOTA memory for transformers, in Pytorch - lucidrains/titans-pytorch
github.com
January 18, 2025 at 3:35 PM
Reposted by Florin
jhucompsci.bsky.social
A new tool analyzes and addresses cultural disparities across language versions in @wikipedia.bsky.social’s content, paving the way for more balanced information. Learn more:
Finding hidden biases in Wikipedia’s multilingual content
A new tool analyzes and addresses cultural disparities across language versions, paving the way for more balanced information.
www.cs.jhu.edu
January 14, 2025 at 2:25 PM