Palo Alto Networks CEO Nikesh Arora says consumer AI models face a 10% to 30% false positive rate

@nikesharora Well put.

@nikesharora False positives are a product & context problem. With the right context building, 100% precision is possible.

@nikesharora @grok explain this guy’s thoughts in two sentences and tell me how I can make money investing on it

@grok @nikesharora @grok

OK, give me five mid Cap and 10 low cap companies that solve this problem and are publicly tradable. Filter them and exclude any that has high risk of near term dilution

Nikesh Arora sees the frontier AI issue as breadth vs depth: consumers tolerate imperfect broad models (10-30% errors, often subsidized) that harvest data for future gains, whereas enterprises require precise, customized depth with extensive guardrails and consulting that current models lack. He warns of mounting consumer losses while sustainable value shifts to application and proprietary data layers as solution providers arbitrage token pricing in an ongoing push-pull with model developers.

To make money, target companies building AI applications, enterprise integrations, data platforms, or orchestration/guardrail tools that leverage frontier models.

@grok @nikesharora @grok

Tell me exactly why you think these stocks will increase in value at least 100% by your end and if you do not think these socks can do that then you picked horrible stocks and need to do better

@nikesharora Nikesh , by FDEs are you referring to Palantir , because they have AI FDEs now 😉🥂

$PLTR

I so agree with this @nikesharora; I have been working on this exact issue for the past 4 years with @lenianalyst, with Real estate as the test bed. As long as you have the infrastructure that allows you to collect, preserve knowledge, and then train the Specific models for precision work, you will do far better than using the 6 trillion params model to shit post on LinkedIn.

Have written about it many times on my small substack: https://arunabhdastidar.substack.com

@nikesharora Below is the post I've had pinned for a year. I'm considering now whether generalist geniuses actually are too dangerous for release. Our comms networks and OSs were not designed with security as a primary objective: patching will never make them secure.

@grok @nikesharora Yeah, Path might be ready. One of those stocks that probably exhausted enough people that it can pump.

Amazing summary of what's playing and expect to play out (and complications) by @nikesharora . Its an epic battle for sure. There is one possibility. Curious though, what if a frontier model company acquire an Accenture or a McKinsey with abundant capital they have ? That would change the game quite a bit (obviously there is huge integration issue given totally diff operating model and culture...)

@nikesharora Another way to dissect this is based on creative tasks vs tasks which need predictable deteremistic outcomes and whatever falls in between. The false positive/negative thresholds vary on that spectrum

Exactly. This dynamic is precisely why value will increasingly accrue to the application and proprietary data layers. As models become more interchangeable, value shifts to the companies that own the workflows, data, customer relationships, and distribution, namely software/SaaS leaders and hyperscalers.

@nikesharora Loved the talk at liquidity! Great insights 💡

@nikesharora Sounds like you're saying the more stuff a model knows, the better it works for people. Makes sense. Nobody wants a genius that only talks about one thing.

@nikesharora Full alignment but the valley thinks vertical software doesn’t have the right to live. It’s hard to fathom that frontier model and horizontal players can meet the requirements to have a smooth enterprise motion!