I entered the #ComputationalLinguistics field in 2018 by enrolling for a Bachelor's degree.
Since then, a lot has changed. Almost all the things we learned about, programmed in practice and did research on are now nearly irrelevant in our day-to-day.
Everything is #LLMs now. Every paper, every course, every student project.
And the newly enrolled students changed, too. They're no longer language nerds, they're #AI bros.
I miss #CompLing before ChatGPT.
Gemini hackers can deliver more potent attacks with a helping hand from… Gemini - In the growing canon of AI security, the indirect prompt injection has eme... - https://arstechnica.com/security/2025/03/gemini-hackers-can-deliver-more-potent-attacks-with-a-helping-hand-from-gemini/ #artificialintelligence #largelanguagemodels #promptinjections #fun-tuning #features #security #biz #gemini #google #llms #ai
TechCrunch: Open source devs are fighting AI crawlers with cleverness and vengeance https://techcrunch.com/2025/03/27/open-source-devs-are-fighting-ai-crawlers-with-cleverness-and-vengeance/ #AI #LLMs
This is why we can't have nice things.
"Lee recently launched Interview Coder, an "invisible" AI tool for job candidates to cheat on technical questions during coding interviews. Lee's startup sells access to the tool for $60 a month"
Columbia suspends student who created AI tool that helps people cheat in coding interviews
https://www.msn.com/en-us/money/news/columbia-suspends-student-who-created-ai-tool-that-helps-people-cheat-in-coding-interviews/ar-AA1BMYHp
404 Media: Hayao Miyazaki, Who Said AI Is ‘Insult to Life Itself,” Reduced to AI-Generated Meme by OpenAI
What does it matter if Google AI is more intelligent if it can't discuss certain important current topics like #Politics or #Conservative #Wealthy #White #Males. That's why I have turned to #AI #LLMs from other sources. #SorryGoogle #FuckYourPropaganda
List of AI bots to add to robots.txt (although they may not obey -- may need to throw them in the bitbucket and 404 or 444 them). In addition to these, you may have to block specific random browser versions for the most aggressive bots who ignore robots.txt.
https://github.com/ai-robots-txt/ai.robots.txt/blob/main/robots.txt
This frustrated sysadmin ends his justified and informative rant with this:
"If you personally work on developing LLMs et al, know this: I will never work with you again, and I will remember which side you picked when the bubble bursts."
https://drewdevault.com/2025/03/17/2025-03-17-Stop-externalizing-your-costs-on-me.html
Did #ColliderFest today on the Viper HPC stand! It was fun talking to people about supercomputing/HPC and AI, but also exhausting.
Used my UMAP word cloud demo to talk about how foundational language models (e.g. like ChatGPT, Ollama, etc) work, and how hallucinations come about.
https://starbeamrainbowlabs.com/blog/article.php?article=posts/533-research-smflooding-vis.html
VentureBeat: The risks of AI-generated code are real — here’s how enterprises can manage the risk
".... As organizations move from experimental mode into production mode, they have increasingly come to the realization that code is very buggy. Shaukat noted that AI-developed code can also lead to security and reliability issues. The impact is real and it’s also not trivial...."
Very disturbed by an idea in a theogg video today. "The end of frameworks" and new language ideas because it won't immediately be available in the LLMs and therefore feel more difficult to use.
That just feels so gross. Imagine if tech had stopped 10, 20, 30 years ago because magic autocomplete worked sometimes.
Thinking about the wasteful nature of #LLMs got me thinking about waste in my own development. While it can be convenient to use the large, enterprise-grade frameworks to deliver a minimalist website in 2025 - it's absurd.
Do I really need #laravel with #react, #jquery, #tailwind, #webFonts, #postgres to host some simple #markdown?
Do I need to re-render a bunch of static content at every hit? Does every simple article require 64 connections to the server to display?
I think not.
I want my material to be available to anyone who wants it - regardless of the device they are using or the robustness of their connection.
I want to respect users who disable #javascript for their personal protection.
I want to respect #ScreenReaders and users of assistive technology, without unnecessary complexity.
Everything we need is built into the HTML and CSS specs.
Technical University of Munich, Germany
If you keep prioritizing your own thirst every day, how is #AI supposed to thrive? 
#SaveTheAI
#LLMs #MachineLearning #sustainability #munich
(hi @tu_muenchen ! So good to see you are here)
"This infection of Western chatbots was foreshadowed in a talk American fugitive turned Moscow based propagandist John Mark Dougan gave in Moscow last January at a conference of Russian officials, when he told them, “By pushing these Russian narratives from the Russian perspective, we can actually change worldwide AI.”
A NewsGuard audit has found that the leading AI chatbots repeated false narratives laundered by the Pravda network 33 percent of the time — validating Dougan’s promise of a powerful new distribution channel for Kremlin disinformation."
https://www.newsguardrealitycheck.com/p/a-well-funded-moscow-based-global #ai #llm #llms
Hunger Games for AI
OK really it's called "Survival Game" but that's the basic idea. Some Cornell researchers decided to put various LLMs through their paces, giving them a limited number of tries to solve a variety of problems, and eliminating those that failed in successive rounds of evaluation. I do NOT have the time to read this comprehensively, but from the intro:
Our results show that while AI systems achieve the Autonomous Level in simple tasks, they are still far from it in more complex tasks, such as vision, search, recommendation, and language. While scaling current AI technologies might help, this would come at an astronomical cost. Projections suggest that achieving the Autonomous Level for general tasks would require 1026 parameters. To put this into perspective, loading such a massive model requires so many H100 GPUs that their
total value is 4 × 107 times that of Apple Inc.’s market value. Even with Moore’s Law, supporting such a parameter scale would take 70 years. This staggering cost highlights the complexity of human tasks and the inadequacies of current AI technologies.
https://arxiv.org/pdf/2502.18858
Check it out! Tell me what jumped out at you.
The Guardian: Revealed: Israeli military creating ChatGPT-like tool using vast collection of Palestinian surveillance data
https://www.theguardian.com/world/2025/mar/06/israel-military-ai-surveillance
@metin@graphics.socialIf your company name is Alibaba, not calling your AI "Genie" is quite a missed chance. 
“…language models can fundamentally be described as supercharged autocomplete tools, prone to returning incorrect information because they are skilled at creating a facsimile of a human-written sentence—something that looks like an acceptable response—but chatbots are not doing any critical “thinking.”“
—Thomas Maxwell, Microsoft's Satya Nadella Pumps the Brakes on AI Hype
#llm #llms #hallucinations #ai #aihype
I love this idea. Use #LLMs to convert pieces of research literature into "knowledge units" that capture "most (~95%) factual knowledge" from the original works and yet avoid repetition or too-close paraphrase of the copyrighted language and thereby avoid infringement.
https://doi.org/10.48550/arXiv.2502.19413
See my related idea in this June 2023 Mastodon thread.
https://fediscience.org/@petersuber/110572463707435047
I believe the two ideas are compatible and compelling. I'd love to see concrete projects carry them forward. Very glad to see that the authors of the new arXiv piece "share #OpenSource tools for converting research documents into Knowledge Units."
