Published on 2024-09-29 13:02
I used ChatGPT to analyze its own future as OpenAI shifts to a for-profit model and key execs, like CTO Mira Murati, leave the company. This seemed like a good “Business Scenario Planning & Strategic Forecasting” use case for OpenAI’s latest model, ChatGPT o1-preview (seriously, who names these models?!)
o1-preview has better reasoning and math skills. It’s great for solving complex problems thinking through approaches step by step, like planning strategies.
So, I put it to the test. I asked it to evaluate future scenarios for OpenAI using a framework: criteria, ratings, weights, pros/cons, and reasons.
It took 27 seconds to think before giving its answer. Unlike earlier versions like ChatGPT 4o, o1-preview takes its time to give more thoughtful responses.
Below are the top three outcomes from each model. (See the ChatGPT conversations with analyses in the comments, https://lnkd.in/gGbWCEBR, including other scenarios evaluated like return to non-profit, full-profit shift, and internal collapse)
I was curious how o1-preview would react, so I shared GPT-4o’s analysis. It adjusted its top choices to align more closely with GPT-4o. I like that it can adjust based on new input.
Initially, I thought o1-preview’s improved reasoning made it the clear choice. But after seeing different results across multiple tests, I wonder if it lacks the consistency needed for strategic forecasting. Also, GPT-4o’s web browsing might give it an edge in staying current, especially in a space that’s moving fast.
I then asked Claude, Anthropic’s AI, since the ChatGPT models seemed to like Anthropic. Claude also leaned toward a merger with Anthropic, plus acquisitions by Microsoft and Google.
Looks like all three models prefer reuniting most original OpenAI execs, likely due to their focus on safety and reducing risks.
Of course, take all this with a grain of salt. These models have key limitations:
► Training data – o1-preview and GPT-4o are trained up to Oct 2023. Claude is trained up to Aug 2023. None have knowledge beyond these dates.
► No browsing – Of the models tested, only GPT-4o can browse the web.
► No specialized knowledge – None have insider knowledge or domain-specific expertise. (And neither do I in guiding them.)
► Lack of human context – Most importantly, they can’t account for human factors like desire, fear, trust, greed, and relationships, which influence real decisions.
This is just a demo of AI tackling complex problems using public data, not financial advice or predictions
Still, it shows how AI can help bring structure to decision-making. It gives us a starting point, but human judgment, expertise, and oversight are key.
In OpenAI’s case, I think the human factors will ultimately be the trump card.
What are your thoughts? Is a Microsoft acquisition inevitable for OpenAI? Or is there a better path to OpenAI’s mission of ensuring AGI benefits all of humanity?
#OpenAI #Microsoft #AI #AISafety #ChatGPTo1 #Anthropic GrowthPath Partners
