#llmfails — Public Fediverse posts

Whoa, hold onto your keyboards, folks! 🚀 We've got another groundbreaking, earth-shattering, totally-not-redundant benchmark that throws language models into the wild jungle of real-world #vulnerabilities. 🤖🔍 Good news: only 953 out of 1000 cases were skipped, proving that even AI needs a day off! 😂 #SecurityEvolution #LLMFails
https://ndaybench.winfunc.com #SecurityEvolution #LLMFails #AIResearch #MachineLearning #HackerNews #ngated

A lot of ebay descriptions lately are pretty obviously generated by AI.
Some are hilarious, sad that the vendors obviously trust AI and don't bother to read them.
A description for a corded woodwork router, for example, praised it's Lithium ion technology and the fact that it runs on 230V. #llmfails #aifails

Last talk for today: Münker in political bias in LLMs #LLMFails

Now, Tymbay on LLMs in language teaching #LLMFails

Now: Li/Fowlie on using GPT as a semantic parser #LLMFails

4th: Storment joining virtually to talk about pictorial constituents #emojis #LLMFails

3rd talk: Maarouf on fine-grained semantic analysis, based on her PhD project studying technical reports #LLMFails

Next: Skachkova on bridging resolution #LLMFails

Sebastian Reimann from @ruhr-uni-bochum.de is going first, presenting on the failures of LLMs on metaphor detection @sfb1475 #nlproc #LLMFails

Kickoff for #LLMFails ! 🦙

https://www.ids-mannheim.de/home/lexiktagungen/llm-fails/program/