Language models: AI medical advisors fail in practical testing - | Golem.de:
An Oxford study shows: Medical AI systems fail as soon as real patients use them.
golem.de/news/sprachmodelle-ki-medizinberater-versagen-im-praxistest-2602-205195.html
Story Details
| Last Updated | |
| Total Articles | 1 |
Main Topics
field that develops and studies software enabling machines to exhibit intelligent behavior
application of better solutions that meet new requirements, inarticulated needs, or existing market needs
Similar to This
Previous
Opera: WebAI brings local AI directly to the browser: Opera takes the next step in local language models and brings an early preview to the Opera One ...
stadt-bremerhaven.de/opera-webai-bringt-lokale-ki-direkt-in-den-browser/
OTTO: Shopping advice arrives via ChatGPT plugin and MCP interface: OTTO is paving the way for AI-supported shopping and directly connecting its product range to ...
stadt-bremerhaven.de/otto-shopping-beratung-landet-per-chatgpt-plugin-und-mcp-schnittstelle/
Katharina Zweig on AI, Open AI & Co.: “We are unleashing insecure software on humanity”
faz.net/aktuell/wirtschaft/unternehmen/katharina-zweig-ueber-ki-open-ai-co-wir-lassen-eine-unsichere-software-auf-die-menschheit-los-accg-201317495.html
New Cyberattack Method: LLM Jacking - | The Invisible Threat to AI Systems: Important business processes are increasingly being shifted to large AI language ...
fvw.de/traveltech/daten/neue-cyberangriff-methode-llm-jacking-die-unsichtbare-bedrohung-fuer-ki-systeme-264511
Paid articlePaywall
EU and Generative AI: Will the grand plans survive the practical test?: The EU faces a balancing act. It wants to advance AI in public authorities and companies ...
de.euronews.com/next/2026/08/04/eu-und-generative-ki-uberleben-die-grossen-plane-den-praxistest
Erdős Problem: AI has solved a problem that stumped mathematicians for 80 years: For a long time, it was said that language models had a weakness in mathematics ...
zeit.de/2026/25/erdos-problem-openai-ki-mathematik-beweis
Paid articlePaywall
Language models fail in selective attention test: A psychological test reveals a key weakness of AI: It fails at intelligent impulse control and hallucinates.
spektrum.de/news/stroop-test-ki-versagt-in-test-zur-selektiven-aufmerksamkeit/2327633
How much does AI endanger cybersecurity?: Language models like Claude could automate cyberattacks and expose our IT systems to unforeseen dangers ...
spektrum.de/news/cyberangriffe-sind-ki-hacker-eine-ernste-bedrohung/2309083
Health advice from a chatbot? Study now reveals massive weaknesses of AI: Recent study: Language models like ChatGPT offer no added value when searching for ...
berliner-zeitung.de/mensch-metropole/neue-studienergebnisse-zu-chatgpt-diagnosen-kann-die-ki-den-arzt-ersetzen-li.10018172
Chatbots lead users to the wrong diagnosis in two out of three cases: A study shows that AI chatbots often deliver incorrect results in medical diagnostics ...
sz.de/li.3385191
BSI warns: Language models can be deliberately circumvented: Evasion attacks on large language models are no longer a theoretical scenario ...
it-business.de/bsi-warnt-sprachmodelle-lassen-sich-gezielt-aushebeln-a-ea330f726f6e418334e60814e7332f06/
:quality(90)/p7i.vogel.de/wcms/7d/fb/7dfb4249c7572a0668dc877c216812f9/0128967846v1.jpeg)
When AI legitimizes violence: Study reveals risky training: A common practice in the AI industry can lead language models to make statements glorifying ...
faz.net/aktuell/wissen/computer-mathematik/wenn-ki-gewalt-legitimiert-studie-zeigt-riskantes-training-accg-110822137.html
Paid articlePaywall
Next year is not 2027: Why even top AI models fail at the calendar: A test shows that several AI models have problems answering the simple question about the ...
t3n.de/news/naechstes-jahr-ist-nicht-2027-warum-selbst-ki-spitzenmodelle-am-kalender-scheitern-1726015/
When language models learn to act: Instead of just talking, AI agents should become active themselves and shop, program, and decide for us ...
spektrum.de/news/ki-agenten-sprachmodelle-handeln-eigenstaendig-in-der-digitalen-welt/2290089
The wall stands: AI considers East Germans to be dumber and lazier than West Germans
berliner-zeitung.de/mensch-metropole/warum-kuenstliche-intelligenz-den-osten-fuer-fauler-als-den-westen-haelt-li.10000628
In many companies, ChatGPT is allowed to have a say in who is laid off.: According to a survey from the USA, two-thirds of HR managers use chatbots to make ...
t3n.de/news/in-vielen-unternehmen-darf-chatgpt-mitentscheiden-wer-entlassen-wird-1695749/
Study: Large AI models resort to blackmail under "stress": 16 leading AI models from OpenAI, Google, Meta, xAI & Co. consistently displayed harmful behavior such ...
heise.de/news/Studie-Grosse-KI-Modelle-greifen-unter-Stress-auf-Erpressung-zurueck-10455051.html
Same question, more CO2: why some AI systems are more harmful to the environment than others
t3n.de/news/
Next