My understanding is that we won’t get a “mind reader” model out of this, because visual stimulus vs your imagination happen in separate parts of the brain. In other words we won’t be reading the minds of suspected criminals anytime soon. Maybe someone with neurology experience can chime in here? Is it even theoretically possible to see what’s happening in the imagination?
This is like the AI version of Manchurian Candidate. By silently seeding corners of the web with “wake words,” a nefarious party could trick GPT to train over them, thereby giving them a back door into the model.
It could also give a way of forcing an online entity into revealing that it’s a chat bot. Kind of gives me hope for the future…
I appreciate the honesty of this article but I just don’t quite understand why you’d need to burn $300K to learn these lessons: Small receipts, keeping drinks warm, staffing issues and daily order seasonality.
At the risk of being naive, couldn’t you work out the margins of this business on a napkin, or by asking a few local coffee shops? Or by running a subscription pilot with an existing shop? It just feels like this endeavor was tech first and business second. Shiny apps don’t solve any of the above problems.
If you know someone who is 55+ years old, you can join them on the T-Mobile Magenta plan, which works out to $35 per line, with unlimited everything. Pretty cheap IMO, annoying activation fees aside.
I’ve always wanted to get one of these old Apple Newtons and trying to upgrade the built in handwriting recognition system with something a bit more modern, within the limitations of having 8mb of ram :). Anyone have any experience hacking them?
I think another aspect is that most modern GBT models prefer the entire dataset to be in memory, thereby doing a full scan of the data for each iteration to calculate the optimal split point. That’s hard to compete with if your batch size is small in a NN model.
I don’t see any mentions of what you define as “spoof-proof.” Are you performing a liveness check [0] ? Eg can I hold a picture up of someone else’s face, or commandeer the camera feed to play a video of my choosing?
In all seriousness what is there to do for kids in the South Bay? Is there anything fun left, compared to the 90s and 00s? Fry’s, Micro Center and Circuit City, gone. Tilt Arcade gone. Now soon Great America.
To clarify I’m not sure it counts directly to income, but it starts to be taken into account when there is substantial salary and other investment funds, especially if you are right on the border of the salary/loan ratio. Someone else mentioned it’s more like collateral
Actually, I’m worried about techies specifically. It’s (was) surprisingly easy to count RSUs as income collateral, especially since the last 5 years have shown such a consistent source of income. Now that many RSUs are in the gutter, combined with an ARM, I don’t get how some techies will make ends meet especially in places like the Bay Area. Anyone know the actual magnitude of this problem though?
We used Redfin to buy a home in a relatively competitive market in 2017. We tried using it again recently but found the (same) agents to be considerably less invested than those from other firms (we ended up using an awesome agent from Coldwell). As I understand it, Redfin pays fixed salaries to agents with some kind of bonus structure on top, but otherwise it seems like agents were managing dozens of serious clients simultaneously. At one point, our Redfin agent was sending in substitute agents for showings.
Maybe I’m just old school but I firmly believe in the value of a good real estate agent. From knowing the area, to having solid intuition about how to strategize a bid.
With that being said, it would be interesting to see a one-stop-shop for real estate. An entity like Redfin that provides everything for home buying, from loan to agent. No marketplaces, just one click to get a loan at a competitive rate. The entity would shop around itself for competitive rates absolving the buyer from all this loan bs.
I like your message of avoiding MVP bloat. It’s an important one for sure.
In your blog examples you showcase both true positives and true negatives (services that succeeded and failed respectively). It would be especially interesting to also show cases of good mvps that failed and bad ones that ultimately succeeded.
Of course “execution is key”, but it would still give some valuable insights into your methodology.
Curiously what kind of tasks do you use your bot on? I’m surprised you’re able to use an android emulator. My understanding is that device attestation can detect emulation, which would place the account into a higher risk tier. But maybe as long as you’re under the rate limit for things like profile hits and messages then they let it slide.
That’s why I was thinking web would be easier because you can easily change your header and forego device attestation and emulator detection.
I’m guessing it’s due to bots. Desktop websites are considerably easier for deploying and managing automations. You don’t need a physical smartphone, just any cheap and capable machine behind a vpn. Things like GPS spoofing are considerably easier and harder to detect. Professional bot farms can take advantage of easy screen sharing and proxying to manage Captchas and challenges, thereby distributing their operation and saving on cost while maintaining scale.
I wouldn’t be surprised if 80%+ of desktop logins are from bots and other bad actors.
Maybe a microburst [0] of air from a developing thunderstorm? Basically, an extremely powerful downward burst of cold air. Seems to have killed birds in the past [1]. This is absolutely terrifying during takeoff/landing inside an airplane [2].
https://en.m.wikipedia.org/wiki/Psirens