4 ms·

We don’t really need to know their whole product strategy, I just can’t even come up with plausible uses for AI that meaningfully change the company enough to r

by soared 3y ago

We don’t really need to know their whole product strategy, I just can’t even come up with plausible uses for AI that meaningfully change the company enough to require a layoff. I disagree on tour last point, organizations that have data useful for bespoke models store it in aws/etc whereas organizations with random files that they need to share across teams use Dropbox.

mieubrisse 3y ago
There's a thing I saw on LessWrong a while ago where he says, "Before declaring a thing impossible, ask, 'Have you put 5 minutes on the clock and just brainstormed?'" In that spirit, here's me doing this about AI at Dropbox: - Get consent from users and train a personalized AI on user files, to answer questions like, "Where is my meeting with Thiago?" - Provide built-in support for number crunching on unstructured meeting notes: "How much revenue did we make, from last quarter's notes?" - Automatically fine-tune a model using your own writing so that ChatGPT talks more like you - Apply an AI to recognize security threats in your files("Hey you might have stored a secret in plaintext in file Y. Do you want to encrypt it?")
- bayindirh 3y ago
  Even being able to train a personal model on the user’s handwriting to digitize the notes will be a big hit. I would love to have that. I’m this close to train my own Tessaract instance to do it.
- ilyt 3y ago
  YOu forgot: * find someone that will actually want to pay extra for their subscription on that.
  - lotsofpulp 3y ago
    And deliver it at a sufficiently better performance/utility to price ratio than Apple/Microsoft/Alphabet/Amazon/Meta can, and they might price it at zero.
- lostlogin 3y ago
  You’re thinking about AI from a Dropbox customer POV. What if it’s more about using customer data to Dropbox’s advantage? Combined with the telemetry you can’t turn off, they can mine the shit out of users. https://news.ycombinator.com/item?id=35724939#35726399 https://news.ycombinator.com/item?id=35724939#35726399
- cwilkes 3y ago
  Those are great applications of AI. Still doesn’t explain about how DropBox canning 1/6th of the company will help them do that. I can see DropBox making some interesting features for their products around AI. That’s going to require manpower to do it. Manpower that’s probably a little nervous about staying at a company that just did layoffs.
rco8786 3y ago
> organizations that have data useful for bespoke models store it in aws/etc whereas organizations with random files that they need to share across teams use Dropbox. This is genuinely blind to anything that's happened in AI over the last 6-12 months. The data that is the most valuable to AI is exactly all of the random, unstructured data (product requirements, decisioning documents, engineering designs, process documentation, etc etc) that has been shared across teams. > I just can’t even come up with plausible uses for AI that meaningfully change the company enough to require a layoff. requiring a layoff no, certainly me neither. There is definitely a bit of "compliment sandwich" happening in this letter. The layoffs were likely just a long time coming for a company that has been rather stagnant for quit some time.

by soared 3y ago

mieubrisse 3y ago

There's a thing I saw on LessWrong a while ago where he says, "Before declaring a thing impossible, ask, 'Have you put 5 minutes on the clock and just brainstormed?'" In that spirit, here's me doing this about AI at Dropbox: - Get consent from users and train a personalized AI on user files, to answer questions like, "Where is my meeting with Thiago?" - Provide built-in support for number crunching on unstructured meeting notes: "How much revenue did we make, from last quarter's notes?" - Automatically fine-tune a model using your own writing so that ChatGPT talks more like you - Apply an AI to recognize security threats in your files("Hey you might have stored a secret in plaintext in file Y. Do you want to encrypt it?")

bayindirh 3y ago
Even being able to train a personal model on the user’s handwriting to digitize the notes will be a big hit. I would love to have that. I’m this close to train my own Tessaract instance to do it.
ilyt 3y ago
YOu forgot: * find someone that will actually want to pay extra for their subscription on that.
- lotsofpulp 3y ago
  And deliver it at a sufficiently better performance/utility to price ratio than Apple/Microsoft/Alphabet/Amazon/Meta can, and they might price it at zero.
lostlogin 3y ago
You’re thinking about AI from a Dropbox customer POV. What if it’s more about using customer data to Dropbox’s advantage? Combined with the telemetry you can’t turn off, they can mine the shit out of users. https://news.ycombinator.com/item?id=35724939#35726399 https://news.ycombinator.com/item?id=35724939#35726399
cwilkes 3y ago
Those are great applications of AI. Still doesn’t explain about how DropBox canning 1/6th of the company will help them do that. I can see DropBox making some interesting features for their products around AI. That’s going to require manpower to do it. Manpower that’s probably a little nervous about staying at a company that just did layoffs.

rco8786 3y ago

> organizations that have data useful for bespoke models store it in aws/etc whereas organizations with random files that they need to share across teams use Dropbox. This is genuinely blind to anything that's happened in AI over the last 6-12 months. The data that is the most valuable to AI is exactly all of the random, unstructured data (product requirements, decisioning documents, engineering designs, process documentation, etc etc) that has been shared across teams. > I just can’t even come up with plausible uses for AI that meaningfully change the company enough to require a layoff. requiring a layoff no, certainly me neither. There is definitely a bit of "compliment sandwich" happening in this letter. The layoffs were likely just a long time coming for a company that has been rather stagnant for quit some time.