Interesting thread on working with an LLM by work requirement and how it feels.

From: @corbden
https://defcon.social/@corbden/114383066951837248

LLMs provide the opportunity to create AI-powered features that were previously unimaginable

Read more https://lttr.ai/Ad5TD

Anyone work on #NLP for #agglutinative or #polysynthetic languages? What tools do you recommend? I'm curious to explore NLP for polysynthetic languages like #Greenlandic #Inuktitut #CentralAlaskanYupik but figure there's more out there for something like #Turkish or #Finnish that I could use for inspiration. #Linguistics #NLU #NaturalLanguageProcessing #LLM #AI @linguistics

Which circle are you in?

Values in the wild: Discovering and analyzing values in real-world language model interactions https://www.anthropic.com/research/values-wild #llm

#portswigger has released some #ai thingy for #burpsuite. It doesn't do anything on it's own, but the feature is on by default. If you go to the settings to disable it they ask you for feedback why you turn that crap off. Isn't that obvious? No matter the pinky promiss you make to your customers, that you don't store the data or train on it, as soon as you hand it of to an #LLM company we have no idea what happens. AI and customer data don't mix. End. Of. Story. #infosec #security

Feel the #AGI

Looking at headlines like „Can artificial intelligence develop consciousness?“

…and I think it’s about as stupid as „Can a lawnmower become a spaceship?“

Neural Networks and #llm fundamentally differ from conscious life on earth. So called #ai is not intelligent. It’s fancy statistics.

Drunk driving or operating dangerous machines and causing harm is a crime.

So what’s an agent acting under the direction of an hallucinating AI causing harm then?

Open Codex CLI: Local-First AI Coding CLI Emerges As Alternative to OpenAI Codex CLI

#OpenCodex #AICoding #LLM #LocalLLM #OpenSource #DeveloperTools #CLI #Python #Phi4Mini #Terminal #AI #CodingAssistant #CommandLine

https://winbuzzer.com/2025/04/22/open-codex-cli-local-first-ai-coding-cli-emerges-as-alternative-to-openai-codex-cli-xcxwbn/

"ChatGPT generuje miliony strat na tym, że użytkownicy piszą „proszę” i „dziękuję”"

Mamy się oduczyć zwrotów grzecznościowych, bo energia?
Niech się pier****

Po co mamy w ogóle tracić energię na witanie się codziennie z kimś, my też tracimy na to zbytnią energię. Po co hejtować, oceniać innych na to też tracimy energię.

Tak, zapewne na takie formy grzecznościowe LLMy tracą energię, ale nie powinni zakazywać używania formuł grzecznościowych. Jak do AI przestaniemy używać ich, to i w prawdziwym życiu przestaniemy ich używać. A miłym słowem można więcej zdziałać niż go nie używać.

#AI #LLM
https://ithardware.pl/aktualnosci/chatgpt_miliony_strat_uzytkownicy_prosze_dziekuje-40894.html

Confusing to me why investors and companies are willing to put so much money into building and running #LLM s. I see no use for them: their output is just annoying to me. Someone told me that the target customers are middle management who hope to cut down on costs for college-educated labour.

Sure, it's impressive how LLMs appear to understand questions. But that's just a fun party trick when you know that the answers they give are glorified auto-complete.

无意中看到了关于四代和五代编程语言的概念，原来SQL算四代编程语言中的数据管理和操作领域用的编程语言，而五代语言的目的是为了让电脑用户不用编程，只要聚焦问题的定义和解法的适用条件。但是上个世纪80-90年代的尝试失败了，人们发现对于一个给定边界的具体问题，要生成对应的计算机算法去解决这个问题，这个任务本身就是一个困难的挑战，要完成这个任务仍然离不开程序员的洞察。

但这个定义让我联想到，最近几年来业界基于大模型开发的各种agent，mcp协议，就是冲着这个方向去的啊，并且看起来比40年前要promising得多了。

It is not the #LLM s fault that it was made with money of greedy venture capitalists - but I thought it doesn’t matter where you come from, it‘s more important where you‘re going?

Vibe coding sta cippa

https://bluebabbler.it/content/Vibe_coding_sta_cippa

21st century edition of „The dog ate my homework“:

My AI hallucinated

Revolutionizing NPC Interactions: How to Build an Interruptible LLM for Real-Time Gaming

In the fast-paced world of online gaming, NPC interactions must feel natural and fluid. This article delves into the innovative techniques used to create an interruptible language model (LLM) for NPCs...

https://news.lavx.hu/article/revolutionizing-npc-interactions-how-to-build-an-interruptible-llm-for-real-time-gaming

#PSA for #VSCode, you want to set
> "chat.setupFromDialog": false
When you don't want to see the "Ask to do something or start typing to dismiss"-prompt for #Copilot garbage.