The articles on how over 90% of enterprise doesn't reach their return-on-investment goals with new AI models because their devs have to fix bugs and loops causing more token spend due to those developers "not constructing the correct prompts" is hilarious. Full on blame shift away from the AI providers and the underlying point. "Its a skill issue"..."Its a tooling issue"... While there can be issues with training when it comes to this... It has nothing to do with the new models using more internal reasoning tokens and increasing extended loops when it comes to unknown edge cases? There's a bottleneck and these enterprises, which already haven't proven ROI from AI models, aren't using new models due to these vast valleys in token overuse. If anything they're focusing on using free and open source models for a large majority of the work with cascading logic to bypass token spend. Which doesn't look good for further investment into new models if the majority of enterprise tasks simply do not require super compute models...and future valuations for these models therefor hits a financial wall. Compound that with the circular nature of keeping these AI companies afloat....yeah.... The more I research, the more I go...."wtf is happening here".... @mttaggart@infosec.exchange #ai #noai #tech #openai #anthropic #xai