drzoltar

This reminds me of the Psirens episode from Red Dwarf:

https://en.m.wikipedia.org/wiki/Psirens

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?

Bot: “I just learned the square root of 2 is 1.41”

Me: “what’s your least favorite movie?”

Bot: “Probably about math since I am good at it…”

Conversation bots are hard lol.

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.

For every 10kW they use, they’ll generate 2kW and claim they’re investing the remaining 8kW back into their company. The Uber of the power business.

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?

[0] https://www.liveness.com/

Are there any good 3rd party alternatives that work better?

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.

Sad to see Anker on the list. I used to think of them as a pretty reliable electronics manufacturer in a sea of low quality chargers and hubs.

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].

[0]: https://www.weather.gov/ama/microbursts

[1]: https://lancasteronline.com/news/local/mystery-solved-unusua...

[2]: https://youtu.be/Q3Sr2F_xxLo

Homer: Not a bear in sight. The Bear Patrol must be working like a charm.

Lisa: That’s specious reasoning, Dad.

Homer: Thank you, dear.

Lisa: By your logic I could claim that this rock keeps tigers away.

Homer: Oh, how does it work?

Lisa: It doesn’t work.

Homer: Uh-huh.

Lisa: It’s just a stupid rock.

Homer: Uh-huh.

Lisa: But I don’t see any tigers around, do you?

Homer: Lisa, I want to buy your rock.

[0]: https://youtu.be/xSVqLHghLpw

I have a dark-pattern theory: Gmail tracks retention metrics, one of whose proxies are app opens. If you don’t get a lot of email, or haven’t opened the gmail app in a while, then they purposely let some spam through to trick you into an app open. Also, people who are susceptible to click on spam are likely excellent targets for Google’s behemoth ad platform, so it probably leads to higher conversion (by collecting data points on you).

Are there any aspects of Wordle that can be patented? AFAIK, game mechanics cannot be patented unless they are demonstrably novel (whatever that means).

Can you speak about your experience building a startup with a newborn nearby?

I think the actual guessing space for these free response problems is much smaller, through simple priors over the question. For example:

“Richard, Jerry, and Robert are going to share 60 cherries. If Robert has 30 cherries, and has 10 more than Richard, how many more cherries does Robert have than Jerry?”

A rudimentary model will likely already know the answer is between 0-60.

Knowing that the answer involves addition and subtraction narrows it down to maybe 8 answers.

While SAT problems have only 4 answers, there’s usually one trick/trap answer, which I think might be be difficult for a model to not accidentally guess. The analogy I can think of is sometimes it’s better to cover up the answers first and work out a solution, to not get biased by any particular answer choice.

It’s frustrating how myopic these papers can be. It seems like the goal of the paper is to solely work within the GPT framework to test the theory of verifiers. Why not try verifiers out with other models? Perhaps it’s not a fair comparison but I remember a Kaggle competition [0] from six years ago which involved building models to solve grade school science multiple choice questions. A simple word2vec model already could achieve 50% accuracy. Despite multiple choice being (maybe?) easier than free response, I’m just skeptical that the way to solve these problems is to throw billions of weights at them. It’s also not convincing to me that this new dataset doesn’t suffer from a much smaller template space, in that the models still just memorize templates.

[0]: https://www.kaggle.com/c/the-allen-ai-science-challenge/over...

Have there been any publicized studies done on the reliability of aircraft engines, with and without lead? I’m guessing this is what made the transition take so long, but I can’t seem to find any references.

I think a big factor here is the sentiment of people who live near small airports. The noise is already an issue. At least now there won’t be lead raining down on the neighborhood, so maybe less petitions to close the airport.

As well, every time I test the fuel on a Cessna it inevitably sprays everywhere so there’s that worry too.

Could you recommend a beginner 3D-printer?

I really miss the addons for this, especially Rat Race. Me and my dad would bet candy on the rats and then yell at them when they turn around and start walking in the opposite direction.

Also the Totally Twisted addon was hilariously office-inappropriate, from lawyers being eaten by piranhas to a guy in a trenchcoat flashing a clock.

So much nostalgia from all this. They just don’t make 'em like they used to.

[0]: https://youtu.be/r81V4PAz_rg

[1]: https://youtu.be/ZsKYXftB68M