The claim going viral right now: “LLMs get lost in real conversation.” It’s being used to suggest that AI systems fundamentally break down the moment you move beyond a single prompt. So I went and read the paper. And of course, that’s not what it says. The study (“LLMs Get Lost in Multi-Turn Conversation”) examines underspecified multi-turn task refinement, meaning: You give a task. Then you add constraints. Then you modify it again. Then you add edge cases. Over and over. That’s not “normal conversation.” That’s iterative requirements gathering. And here’s what the researchers actually found: • Performance drops primarily because reliability collapses across turns (best-case vs worst-case results diverge). • Models tend to anchor on early interpretations and don’t always fully re-evaluate when constraints change. • Even reasoning-enhanced models show the same structural reliability issue. This is a system design challenge. It’s about constraint tracking. State updating. Error accumulation across iterations. It is not evidence that LLMs “can’t handle real conversation.” If anything, it reinforces something anyone who works in project management, product design, or business analysis already knows: When requirements change midstream, you restate the scope. AI systems need that. Humans need that. So why are we framing structured iteration as a catastrophic failure just because AI is involved? The practical takeaway is: If you change the task, say what changed. If you add constraints, restate the objective. If you revise direction, signal the revision clearly. That’s structured iteration that's pretty darn normal. What do people think an LLM is? Mind-reading technology? I broke down the actual findings (with links to both the paper and the viral take) here: crispyrose.com/no-llms-dont-… Read the research, then decide whether the headline matches the data. #AI #LLMs #AIResearch #GenAI #PromptEngineering #AICommunication #CriticalThinking #TechDiscourse #MachineLearning #ArtificialIntelligence

ELMO IS OUT OF CONTROL AND TRYING TO SILENCE THE LEFT. LITERALLY. Luckily Cut/Paste still works. #EvilMusk #TechInfluence #InnovationDebate #AIEthics #CorporateResponsibility #SocialMediaImpact #DisruptiveTechnology #PublicPerception #TechGiants #DigitalCulture #LeadershipChallenges #FutureOfTech #VentureCapital #Entrepreneurship #MarketTrends #TechAccountability #InfluencerImpact #SustainabilityInTech #ConsumerAwareness #DigitalEthics #TechCritique #BusinessEthics #InnovationVsEthics #TechDiscourse #SocietalImpact #TechLeadership #CrisisManagement #PublicTrust #TransparencyInBusiness #EthicalTech

Fundamental and very interesting for Media. Thanks! #shumtech #shumedia #shumevo #techmythologies #techdiscourse #techevolution

Don’t believe the hype: the media are unwittingly selling us an AI fantasy | John Naughton #divinefuture #shumtech #ai #mediarepresentations #techrepresentation #techdiscourse #media theguardian.com/commentisfre…