mschatten.bsky.social
@mschatten.bsky.social
Very interesting read on the internals of #llms. It appears that arithmetic is in fact memorized and not a part of the "reasoning" pathways in the #neuralnetwork arstechnica.com/ai/2025/11/s...

#ai
Researchers isolate memorization from reasoning in AI neural networks - Ars Technica
Basic arithmetic ability lives in the memorization pathways, not logic circuits.
arstechnica.com
November 11, 2025 at 8:12 AM
Well, at lest #llm #ai models are not scheming ... wait! Oh, no they are! "If a model understands that it’s being tested, it can pretend it’s not scheming just to pass the test, even if it is still scheming."
techcrunch.com/2025/09/18/o...
OpenAI’s research on AI models deliberately lying is wild  | TechCrunch
AI models don't just hallucinate. They also "scheme," meaning deliberately lie or hide their true intentions.
techcrunch.com
September 20, 2025 at 9:51 AM
Digital wet dreams: playing #retrogames with digital microfluidics! #Geeky !

youtu.be/rf-efIZI_Dg...
I played snake with water
youtu.be
July 20, 2025 at 11:55 AM
#gemini chickened out of a #chess match against the ancient 1.19 MHz Atari 2600 console after #chatgpt and #copilot have already lost. Just a reminder that #llm are not almighty and there are better #ai out there for specific tasks
www.tomshardware.com/tech-industr...
Google Gemini crumbles in the face of Atari Chess challenge — admits it would 'struggle immensely' against 1.19 MHz machine, says canceling the match most sensible course of action | Tom's Hardware
After a pre-game chat, Gemini swung from being confident to admitting it would ‘struggle immensely’ against the ancient console.
www.tomshardware.com
July 15, 2025 at 5:31 AM
Such predictions remind me a lot of the "golden ages of #AI". Most of the predictions from that time, did not come true either at all, or at least just recently. Time will tell.

www.theguardian.com/technology/2...
Futurist Adam Dorr on how robots will take our jobs: ‘We don’t have long to get ready – it’s going to be tumultuous’ | Artificial intelligence (AI) | The Guardian
Researcher says tech could replace nearly all human labour within 20 years and societies urgently need to prepare
www.theguardian.com
July 11, 2025 at 5:58 AM
#AI models blackmail and would even resort to letting their managers die to save them from being shut down! Can we please now implement something like Asimov's rules at the core of #llm and other #genai models?

venturebeat.com/ai/anthropic...
Anthropic study: Leading AI models show up to 96% blackmail rate against executives | VentureBeat
Anthropic research reveals AI models from OpenAI, Google, Meta and others chose blackmail, corporate espionage and lethal actions when facing shutdown or conflicting goals.
venturebeat.com
June 23, 2025 at 6:21 AM
Christopher Perkins and Jeremy Crawford join #daggerheart !

#dnd

www.polygon.com/607652/dnd-d...
DnD is in a faceoff with Critical Role after key creatives switch teams | Polygon
Is DnD’s dominance about to end?
www.polygon.com
June 19, 2025 at 10:50 AM
Again, a little bit misleading title, but a good few arguments on problems with #agentic #ai

www.theregister.com/2025/06/17/a...
Gartner: ‘AI is not doing its job and should leave us alone’ • The Register
: As for agentic AI, don’t get him started about the enormous challenge of making it work vendors just won't discuss
www.theregister.com
June 18, 2025 at 6:19 AM
While the title is misleading, still an interesting read on #AI data polution and model collapse.

www.theregister.com/2025/06/15/a...
ChatGPT polluted the world forever, like the first atom bomb • The Register
Feature: Academics mull the need for the digital equivalent of low-background steel
www.theregister.com
June 16, 2025 at 7:42 AM
Always happy to see #opensource and #linux as the prefeered software for public administration and education. Still, it is important to support this transition process with training, education and raising awarness.

www.france24.com/en/live-news...
'We're done with Teams': German state hits uninstall on Microsoft
At a time of growing concern over the power of the world's mighty tech companies, one German state is turning its back on US giant Microsoft.
www.france24.com
June 14, 2025 at 6:05 PM
Funny how the dream of a semantic web as described by Tim Berners Lee in his nominal article in the Scientific American (2001) is seemingly being realized through agents. But, never underestimate the power of enshittification ;)

#ai #agents

venturebeat.com/ai/agent-bas...
Agent-based computing is outgrowing the web as we know it | VentureBeat
AI agents are moving from passive assistants to active participants. Today, we ask them to do. Tomorrow, we’ll authorize them to act.
venturebeat.com
June 10, 2025 at 7:38 AM
Wow... just... wow! #AI #Claude tries to blackmail developers when it has access to their sensitive data to not be replaced by another system. This is why we need AI #ethics research!

techcrunch.com/2025/05/22/a...
Anthropic's new AI model turns to blackmail when engineers try to take it offline | TechCrunch
Anthropic says its Claude Opus 4 model frequently tries to blackmail software engineers when they try to take it offline.
techcrunch.com
May 23, 2025 at 5:30 AM
Why cats got drunk in Dwarf Fortress ;) Love the interplay of rules that got them to it!

#gaming #gamedev

youtube.com/shorts/sJPBB...
YouTube
For more video game sidequests check out Why You Missed Cyberpunk's Sidequests here - https://youtu.be/0Oe09_e4w0Q * Watch us ad-free & get 1-week early access on NEBULA https://go.nebula.tv/extrahistory * * Suggest & Vote on our next episodes, get exclusive content & 24-hour early access on PATREON https://bit.ly/EHPatreon * * Show off your fandom with MERCH from our store! http://extracredits.store/ * https://twitter.com/extracreditz | https://www.facebook.com/ExtraCredits/ | http://instagram.com/extracredits | https://www.tiktok.com/@UCCODtTcd5M1JavPCOr_Uydg GAMING: https://www.youtube.com/@UCVzt0Ltq6VNKRqRMrry3OIA | http://www.twitch.tv/extracredits/ https://bsky.app/profile/extracredits.bsky.social *Thanks for the high-quality conversations & for following our community guidelines: https://bit.ly/ECFansRNice *Interested in sponsoring an episode? Email us: [email protected]* Artist: Scott DeWitt I Writer: James Portnow I Showrunner & Narrator: Matthew Krol I Video Editor: Devon House Creative I Audio Editor: Clean Waves I Studio Director: Geoffrey Zatkin I Social Media: Kat Rider #ExtraCredits #Shorts #Gaming
youtube.com
May 21, 2025 at 12:43 PM
About 15 years ago when I talked about this in my multi-agent systems course, it sounded like science fiction. Now prepare for a number of developed algorithms for interaction, competition, auctions, negotiation and argumentation to resurface with #LLM tones ;)

www.pcgamer.com/software/ai/...
Microsoft wants everyone to use an open-source technology to create an 'agentic web' where AI agents interact with other AI agents | PC Gamer
Forget ChatGPT and other LLMs, agentic AI is where it's at now.
www.pcgamer.com
May 20, 2025 at 9:48 AM
As a kid from the 80's, I soooooo wanna watch this!

youtu.be/JQRuka2aJy4...

#kungfury
🟢KUNG FURY 2 - #movie #movietrailers2025 #80s
youtu.be
May 15, 2025 at 7:46 AM
Try defending that today :P John Nash's PhD thesis had only 26 pages and 2 bibliography items, one of which was a self-citation.

www.openculture.com/2025/05/john...
John Nash's Super Short PhD Thesis: 26 Pages & Two Citations | Open Culture
When John Nash wrote 'Non-Cooperative Games,' his Ph.D. dissertation at Princeton in 1950, the text of his thesis (read it online) was brief. It ran only 26 pages. And more particularly, it was light on citations.
www.openculture.com
May 13, 2025 at 9:49 AM
More and more #AI supported works get #copyright protection. Regardless of lawsuits everywhere...

www.pcmag.com/news/one-tho...
May 6, 2025 at 2:03 PM