#llms

Digital Tar Pits - How to Fight Back Against A.I.

A new movement aimed at poisoning A.I. models like ChatGPT has gained traction after hackers have been attempting to trap said models in a never ending ‘Tar Pit’ of nonsense. After reading an Ars Technica interview, I tracked down a hacker developing tools to poison AI training data. Tools such as ‘Nepenthes’ are designed to confuse and corrupt the models that scrape the internet for their learning. But can we really stop A.I. from turning the web into a mess of low-quality, regurgitated slop?

#AI #LLM #LLMs

#AI models struggle with paragraph and document-level reasoning, often overgeneralizing and misinterpreting individual sentences, according to a computer scientist who analyzed different #LLMs: https://buff.ly/jILhPZ6
Manas Gaur, University of Maryland, Baltimore County

Most #LLMs over-generalized scientific results beyond the original articles

...even when explicitly prompted for accuracy!

The #AI was 5x worse than humans, on average!

Newer models were the worst.

Accepted in #RoyalSociety Open #Science: https://doi.org/10.48550/arXiv.2504.00025

@evan I'm not a fan of AI models, even local LLM's.

"I want AI to do my laundry and dishes so that I can do art and writing...
...Not for AI to do my art and writing so that I can do my laundry and dishes."

In my opinion, it diminishes creativity.

#AI #LLM #LLMs

Wikipedia and Kaggle Release Structured Dataset to Aid AI Development, Counter Scraping

#AI #AITraining #Wikipedia #Kaggle #AIData #MachineLearning #OpenData #Wikimedia #Datasets #BigData #DataScience #LLMs #NLP #Google #Alphabet

https://winbuzzer.com/2025/04/17/wikipedia-and-kaggle-release-structured-dataset-to-aid-ai-development-counter-scraping-xcxwbn/

Interesting (short) paper of game-based training and evaluation of agentic behaviour in LLMs: Leon Guertler, Bobby Cheng, Simon Yu, Bo Liu, Leshem Choshen, Cheston Tan.: "Textarena"

https://arxiv.org/html/2504.11442v1

#llms #AI #generativeai #agents #agenticAI #evaluation

Next in line of our ISWS 2025 tutor superheroes will be John Domingue from Knowledge Media Institute (KMi), the Open University’s technology research and innovation centre in UK.

https://2025.semanticwebschool.org/

#semweb #superhero #tutor #AI #semanticweb #knowledgegraph #isws2025 #academiclife #summerschool @fizise #llms #reliableAI #responsibleAI @AxelPolleres @albertmeronyo @enorouzi

El lado del mal - Exploiting Package Hallucination at Scale contra Vibe Coders con LLMs https://www.elladodelmal.com/2025/04/exploiting-package-hallucination-at.html #VibeCoding #LLMs #Hacking #LLM #PackageHallucination #Malware #AI #IA #Exploiting #ChatGPT #Llama #DeepSeek #InteligenciaArtificial #ArtificialIntelligence

We are very happy that Heiko Paulheim from University of Mannheim, Germany, is joining the ISWS 2025 tutor team again. ...and as you can see by yourself, he seems to be proud, too, to be part of our summer school ;-)

https://2025.semanticwebschool.org/

#semweb #summerschool #isws2025 #knowledgegraph #AI #responsibleAI #reliableAI #generativeAI #llms #semanticweb #academiclife #PhDs @fizise @albertmeronyo @AxelPolleres @pgroth

Readings shared April 15, 2025. https://jaalonso.github.io/vestigium/posts/2025/04/15-readings_shared_04-15-25 #CommonLisp #Emacs #FunctionalProgramming #Haskell #ITP #IsabelleHOL #LLMs #LeanProver #Lisp #Math #Programming #RacketLang

Smithery https://buff.ly/NSpTkyz

"Smithery is a platform to help developers find and ship agentic services."

#MCP #genAI #LLMs #api360

https://www.the-independent.com/tech/bots-internet-traffic-ai-chatgpt-b2733450.html
#cybersecurity #bots #AI #LLMs #Bytedance #ByteSpider

You can experiment with LLMs like GPT-4.1 and Llama Maverick free with GitHub Models:
https://github.com/marketplace/models
And even use Python frameworks like LlamaIndex, Pamela Fox explains: http://blog.pamelafox.org/2025/04/how-to-use-any-python-ai-agent.html

Rate limit info: https://docs.github.com/en/github-models/prototyping-with-ai-models#rate-limits

#GenAI #GitHub #LLMs

Kimina-prover preview: Towards large formal reasoning models with reinforcement learning. ~ Numina & Kimi Team. https://github.com/MoonshotAI/Kimina-Prover-Preview #LLMs #LeanProver

https://www.europesays.com/uk/21435/ Google unveils new prompt engineering playbook: 10 key points on mastering Gemini, other AI tools | Technology News #AI #AIChatbots #AIGeneratedOutputs #ArtificialIntelligence #GenerativeAI #GoogleGemini #GoogleVertexAI #JSONOutputFormatAI #LLMs #ModelConfigurationsAI #PromptEngineering #PromptingTechniques #Technology #TextPrompts #UK #UnitedKingdom

I’m looking forward to testing the Nvidia Jetson Orin to run #LLMs locally. Max 25W power seems much more realistic for an #AI utility and would be a massive privacy boost.

https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-orin/nano-super-developer-kit/

When AI generates new content, is it creating, recombining, or simply recalling pieces of what it’s seen before?

Enter OLMoTrace, a new tool from the Allen Institute for AI (AI2):
Links AI outputs directly to the model’s training data
Built on open models and datasets
Top-tier performance

If you’ve ever wanted to better understand your AI — and build even more trust in its outputs — this is an incredible resource.

Try it here: https://playground.allenai.org/

#AI #GenAI #GenerativeAI #LLMs

The latest fresh hell in #AI use: a service that calls your elderly loved ones because you can't be arsed to. The framing is incredibly selfish.

I think the quote from the company's CEO is absolutely key:

“I created inTouch as a personal project to help me care for my aging mother (88) and aunt as I lived far away from them,” he wrote in an email. “I needed the reassurance my mother was safe even on days when I couldn't call her, I wanted her to know I was still in touch with her even if I didn't call her. I don't want to replace calls and visits from the family, never, these are the best (I fly to see her every 2 weeks), I want to complement them.”

Notice the CEO's selfish framing, which mirrors the way he expects his customers to think about this:

I created...
I needed reassurance...
I wanted her to know...
I don't want...

Not one time do the needs of the elderly relatives enter the thought process. Not once does he talk about her wants/needs and then discuss how this #LLM addresses it.

For me the output of #LLMs is frequently judged against an alternative of absolutely nothing. In my own words it's like: "I can call her once a week. If she wants to connect more often than that, she can talk to the bot. Isn't talking to the bot better than not talking at all?"

But in no way does this put the recipient of the call in the picture. And I especially like how the journalist had their own elderly mother offer opinions and she instantly points out the speed and business-like nature.

So much awfulness and selfishness.

@baldur Thank you for this! It's important to spread this message. Moreover, it is important to understand that #AI / #LLMs like #Google #Gemini already refuse to analyze and discuss #White #Conservative #Male politicians: they #Obnubilate #Obfuscate and #Omit. It is important to understand that we can not go back. #CaroleCadwallader is correct: we are living in a #DigitalAutocracy. Use #AI from other countries. Competition among them is one way to resist and rebel. But we can not go back.

How to enjoy writing in spite of the lure of generative AI

Over the last year I’ve been working on a book How to Enjoy Writing exploring the implications of generative AI for academic writing. I felt I had something important to say about the personal reflexivity involved in working with large language models, but in recent months I’ve realised that I lost interest in the project. Given the book was about cultivating care for our writing, as opposed to rushing through it with the assistance of LLMs, I’ve decided to break it up into blog posts which I’ll share here:

The lure of machine writing and the value of getting stuck
The Eeriness of Writing With Claude: When AI Mirrors Your Voice
Thriving in Creative Darkness: Free Association and LLM Collaboration
The Ethical Grey Areas of Machine Writing in Higher Education
Machine writing and the challenge of a joyful reflexivity
The Ebb and Flow of Writing: From Struggle to Unconscious Fluency
Will Claude tell you if your writing is crap? The danger of LLMs for wounded academic writers
Generative AI and the creative confusion of academic writers
Using Generative AI for functional rather than expressive writing
The Joy of Academic Writing in the Age of AI
The Objects With Which We Write: The Materiality of Academic Writing in a Digital Age
How LLMs change the relationship between thinking and writing
Machine writing and keeping your inner world awake
Finding Joy in the Creative Darkness: Reflections on Writing and Stuckness
The subtle pleasures of LLM’s psuedo-understanding
We urgently need to talk about the temptations of LLMs for academics
Generative AI and thriving in creative darkness
Academic writing has always been in flux

This is Claude’s summary of the core argument which unites these posts into a coherent project. One of the reasons I lost my enthusiasm for the project was the manner in which its capacity to imitate my style, sometimes doing it when I hadn’t asked, disrupted the psychology of my enthusiasm for what I was doing:

The core argument of the book is that generative AI forces academics to confront fundamental questions about why we write and what writing means to us beyond mere productivity. While machine writing offers tempting solutions to the difficulties inherent in academic writing, these difficulties are actually integral to the creative process and intellectual development. If we embrace AI tools primarily as efficiency mechanisms to produce more outputs more quickly, we risk losing the joy and meaning that make writing worthwhile in the first place. Instead, we should approach AI as a conversational partner that enhances our thinking rather than replacing it, staying with the productive "trouble" of writing rather than seeking to escape it. This reflexive approach to writing technology allows us to resist the instrumental acceleration of academic life while still benefiting from AI's creative potential.

However I’ve used Claude to support the editing of these blog posts based on the 80% complete draft of the book, simply because I wouldn’t get round to it otherwise. It has copy edited extracts, condensed them at points, chosen some titles and generally polished the text. There’s a few bridging sentences it provided but nothing more than this. I’m glad it’s given this project a public life because I feel like I was saying something valuable here. But I wasn’t willing to produce a second book on generative AI in two years, as it felt like I was stuck in a performative contradiction which was increasingly uncomfortable.

Instead my plan is to focus on doing my best intellectual work by focusing, for the first time in my career really, on one thing at a time. I’ll still be blogging in the meantime as the notepad for my ideas, but I’d like to take a more careful and nuanced approach to academic writing going forward. I’m not sure if it will work but it’s a direct outcome of the arguments I developed in this book. It was only when I really confronted the rapid increase in the quantity of my (potential) output that I was able to commit myself in a much deeper way to the quality of what I wanted to write in future.

https://www.youtube.com/watch?v=6IytEOXamsk

And this is how we rise - by taking a fall
Survive another winter on straight to the thaw
One day you'll learn to strain the tea through your teeth
And maybe find the strength to proceed to the peak
You press on into the thin again and cannot breathe
Swallow so much of my damn pride that it chokes me
The real risk is not a slipped grip at the edge of the peak
The real danger is just to linger at the base of the thing

This is a follow up to the 23 part series I did last summer on How To Enjoy Writing. In fact it emerged directly from “I have something to say here” to “I should write another book”, which is exactly the transition I’m now questioning in myself

Be rigorous about capturing your fringe thoughts
Placing limits on your writing practice
Being realistic about how long you can spend writing
Embracing creative non-linearity
Keep trying to say what you’re trying to say
Procrastination is your friend, not your enemy
Knowing when (and why) to stop writing
Initial reflections from my AI collaborator
Identifying and valuing your encounters with ideas
A poetic interlude from Claude
Cultivating an ecology of ideas
Claude’s ecology of ideas self-assessment tool
Only ideas won by walking have any value
Using generative AI as an interlocutor
Word acrobatics performed with both harness and net
Don’t impose a shape on things too quickly
Creative confidence means accepting the tensions in how you think
Understand where the ideas which influence you come from
Not everything you write has to become something
Being a writer means being good at AI
Make your peace with the fact you don’t have creative freedom
Confront the creepiness of LLMs head on
Be clear about why you are writing

#academicWriting #claude #LLMs #reflexivity #scholarship #technologicalReflexivity #writing