AltDeep is a newsletter focused on mental models and microtrend-spotting in machine learning, and AI.
I perceive various manifestations of a persistent gap between AI products and the users they are intended for…
An entire cottage industry is devoted to helping design and manage campaigns on top ad platforms such as Google and Facebook, despite ads being these platforms’ primary source of revenue.
I’ve talked to leaders at AI startups that market tools based on machine learning for natural language understanding that help marketers analyzing and optimize marketing and advertising campaigns (AdSense, content SEO, email promotions, A/B content experimentation). A consistent complaint I’ve heard is that the individuals who have the sophistication and inclination to make use of such tools are rare. They are likely to be employed by companies large and sophisticated enough to have a custom tool or can get a marketing technology company to give them a custom implementation that doesn’t scale.
Heard from a VC… “The problem is that now when you ship AI, you have to ship a Ph.D. along with it.”
A few years ago, the probabilistic programming community was talking about democratizing AI. Now they are talking about program transformations and other advanced research topics.
> the probabilistic programming community was talking about democratizing AI. Now they are talking about ... advanced research topics.
Fair point, and thanks for reminding me of our community's goal. In defense of "advanced research topics": In my experience peddling probabilistic programming, the biggest impediment has been users complaining "its too slow" or "it's less accurate than my existing bespoke [monolithic hacky] solution". Users are metrics focused, and they won't switch paradigms if that means quantitatively worse solutions in the short term. I see "advanced research topics" as necessary means to make our PPL systems competitive with existing bespoke solutions and therefore able attract users.