New bottleneck: Marketing.
For my entire career, software/engineering has been the critical path.
That's not true anymore. I'm not writing much code myself and we all now have near-infinite bandwidth for shipping products, when using AI efficiently.
Software has so much existing tooling that brings together all required context. Error & info logs, code repos, and even infrastructure & usage logging - literally in 2 places for me: AWS + Github. It's so easy to recurse improvements because you're just connecting 2 things.
Marketing context management is f'ing impossible. Assets scattered all over the place. Analytics across many systems. Even as a team of 1, it's not obvious what to use as the source of truth. Or even how to connect all these systems.
AWS should still be in the picture (or other compute ecosystems). But what is the data repo?
Legacy CMS systems aren't helpful anymore because they drop trajectories (think: git versioning)
β¨AI newsletters are a clever way to promote & launch your AI product. But finding them all is tricky
We assembled a list of 50+ top AI newsletters, highlighting which ones offer sponsorship
To get the full list
π Comment "newsletters"
(must have DM's open or be following me)
Solving many problems requires trade-offs among multiple 'correct' points of view vs jumping to a single correct answer.
Google/Search casually solved both types of problems:
1) delivering a single perfect answer AND/OR
2) providing page-ranked points of view
Many new AI feature announcements focus on model details and snazzy feature overviews vs customer impact
Watching closely in the next few months to see if the headlines flip from 'shipped AI feature' to 'customer ROI increased XYZ'. (2/2)
It seems like many corporate AI strategies are still focused on checking the box and joining the wave. ie, "ship this AI feature asap so we don't fall behind"
How can you tell? (1/2)
This approach seems to work across a handful of domains we've tested so far - It just requires a durable vectorization technique.
Super cool unlock for realtime conversational AI
@riffusionai is a gift that keeps on giving
@sethforsgren y'alls clever solution to stay in latent space during topic interpolation has broader use-cases
We applied this same idea to text during realtime conversation topic modeling. It's incredible
cc @natfriedman@danielgross
This approach gives you the topic clusters, but more importantly it flags when topics are shifting in a live conversation in a machine-readable way
It's otherwise an annoying problem to solve because of how noisy topic modeling can be in text-space before a conversation ends
Recently I've been thinking about the behaviors that remote work normalized. One example:
Every live meeting - zoom or not - has a sub-meeting happening on Slack/Teams/sms. Private messages lobbed back and forth among meeting participants.
This was rare 5+ years ago, right?