I've taken careful steps to rid my speech of statements couched in terms of percentages and use qualitative terms as accurately as possible instead. Unless, of course I'm reporting well-documented percentages, such as in my own or other's research.
Thank you Azeem for this! Exponential view doesn’t just uncover the insights the world is missing, but now fact checks what the world thinks it knows. Love it.
Personally, as an individual, not an organization, I would say that in June of 2025, results were so unreliable (hallucinations included), that the AIs I used non-professionally did not produce much helpful work. Now the opposite is true. I spent 10 days "all day, every day" vibe coding" with Claude (successful), ChatGPT (laggy, not successful), and Gemini (very successful with smooth workflow). This was a vast improvement in productivity over myself at my very peak--probably on the order of 8 times more productive at least.
No study prior to mid-year last year is likely to be at all conclusive, and things are moving so fast that they can barely even be suggestive.
Thank you for checking this stat. It has been on my list to do as I was super skeptical when I saw it everywhere.
"60% of the time, it works every time"
Really cool you reached out for the clarifications. Great work
I've been saying for at least the past 8-10 years that 95% of all statistics are made up on the spot.
And often for clicks!
I've taken careful steps to rid my speech of statements couched in terms of percentages and use qualitative terms as accurately as possible instead. Unless, of course I'm reporting well-documented percentages, such as in my own or other's research.
Thank you Azeem for this! Exponential view doesn’t just uncover the insights the world is missing, but now fact checks what the world thinks it knows. Love it.
Personally, as an individual, not an organization, I would say that in June of 2025, results were so unreliable (hallucinations included), that the AIs I used non-professionally did not produce much helpful work. Now the opposite is true. I spent 10 days "all day, every day" vibe coding" with Claude (successful), ChatGPT (laggy, not successful), and Gemini (very successful with smooth workflow). This was a vast improvement in productivity over myself at my very peak--probably on the order of 8 times more productive at least.
No study prior to mid-year last year is likely to be at all conclusive, and things are moving so fast that they can barely even be suggestive.
Of course you personal experience is valid. But you aren't branding it as a population wide statistic!