barefeg

In the case of the VanMoof s3 and above the app is used to unlock the bike. It has a “kick lock” that basically locks the rear wheel. When the phone approaches the bike then they communicate via Bluetooth and allows the “kick lock” to release either by pressing the button on the handlebar or tapping on the app’s button (or watch app companion).

Apart from that it allows to change some settings that are not configurable on the bike. The bike has only 2 buttons and an LED display. Useful settings are for example the speed profile for the automatic gear shifting.

Ghost writers are not always credited in books. Is it because they receive money in exchange?

I often start reviewing in GHs UI but depending on the complexity of the code I might need to go to vscode for better readability and intellisense. Does Visibly have an extension for that too?

Kind of off topic, but has tiling managers for macOS improved in the last few years? I remember I tried one solution (I can’t remember the name now) which worked OK but it was a little buggy some times. Ever since I just use the standard window manager and try to manage with cmd + tab, cmd + `, and gestures to change spaces.

I’m curious to know the real reason for the change. Is it related to the data they used for training?

Are there alternatives to the database service with similar API or is it leaving a gap in the market?

What would be the way to do yaml?

That’s an interesting combination of WeWork and Airbnb. It makes sense; people go to WeWork to not be alone, and remote work forces people to use their spare rooms to create home offices. So now they can’t rent out the spare room in Airbnb but they could rent a shared desk for work.

I also write in the code what I was doing so the code doesn’t compile. My only issue is that I don’t trust my computer, for me, if it’s not pushed in Git then it can disappear at any moment. If the work is not too important and I could easily re-create it from scratch the next day then I’ll leave it in my local Git. But if it’s important then I also want to push it.

FastAPI doesn’t come with an ORM out of the box. There’s https://sqlmodel.tiangolo.com/ which wraps SQLAlchemy and allows defining the database models with pydantic-like classes.

For me neither Django nor SQLAlcjemy hit the right spot when trying to define “pure” domain models that have no dependency (I.e. no ORM information). For that I created a repository pattern approach that converts “domain mode operations” into SQLModel query operations via filter specifications (https://martinfowler.com/apsupp/spec.pdf).

The result is that I always need to define 2 models: one is my pure domain model and another is a “table model” (defined in SQLAlchemy) along with a mapping between the two (for cases where there’s no 1-to-1 trivial mapping between their fields). Also making some of the SQLAlchemy magic (for example back propagating relational attributes) is messy.

But if you want to produce some quick applications I think SQLModel+FastAPI works quite well. I don’t think is as mature as Django though.

Right, so sort of like https://github.com/bigscience-workshop/petals but for the training phase. I suppose different training runs could be proposed via a RFC type of procedure. Then it’s not only the open source model maintainers that put the effort, but also supporters of the project can “donate” their hardware resources.

But aren’t the weights still not for commercial use?

I see it from the other side. LLMs are so commoditized that it’s super easy to add GPT to anything you can think of.

> Now, you have no idea what ads or manipulative dark patterns your kids/teens are getting on their phone.

I’m not sure about teens. But I’m present when my kid plays games and he despises ads and micro transactions as much as I do. He’s also not allowed to buy anything so he’s more selective about what games to play (a few ones that we have subscriptions for)

Do you have any plans for automated acceptance testing?

Is there any way around it?

> call it a day, finish work early and go to the beach for sunset.

This is closer to Python’s philosophy if you ask me.

That’s interesting do you have more details about the runner farm? In my small scale experiment it was difficult to have so many available runners at all times. We run a monorepo and there are changes that will hit several apps that need testing. So for example opening a PR might need 40 runners. Meanwhile other devs might need 1 or 2, etc.

I followed the inverse path: first academia in non-CS and then tech engineering. And though it hasn’t been that many years after academia, I wouldn’t be able to do it again after my gap. Granted, academics will probably don’t care about my experiences doing something else, but personally I have forgotten so many things that I’d have to start with master-level courses to refresh my memory. And then spend months to catch up with the advancements in the field I was working on.

But then again, I might be comparing oranges to apples here.

Sounds like an interesting path you’ve taken. Do you have any memoirs or blog post about it?

Do you think pricing based on features would be more reasonable for people that don’t currently have the need for the other 3rd party systems? I’ve heard that they don’t want to make pricing more complex but it does seem like a steep pricing if you only need say what GitHub offers for $4/user.

What is the plan to solve injections using this lower level representation?

> I will straight up ignore and delete bug reports from non-customers.

Wouldn’t it be more helpful to leave the reports but not work on them actively? It might be that a paying customer finds the same bug anyway.

For B2C applications maybe. But I don’t know many enterprise users who would like to send any of their data to OpenAI. So “enterprise-readiness” would be another big contributor.

Do you have a blog post with your findings? (Curious)

Out of curiously what kind of advisory roles have you had?

Completely understand your point of view. In my case, growing up I was interested in computers, and back then (late 90’ early 00’) kids my age were not into that. So after school I would roam forums. I would some times exchange IM with people I shared interests with, and chat with them every day. Most of them were adults and young adults, I didn’t know where they lived, how they looked, or their names. However I never felt the need to meet them in person. And my interactions felt as real (or even more) than interactions in the physical world.

Fast forwarding to now, I find myself mentally fine when working from home. Of course things are different since I have family to interact with at home, but for me, not engaging with colleagues about non-work related things is what makes remote work not as fun as the olden times. This is also harder to arrange because my colleagues would rather go out or do something else rather that keep chatting with me about something else.

Do you have some examples of situations where it’s ok for them to make mistakes while figuring things out?

We recently released GPT powered features for our research assistant https://www.youtube.com/watch?v=FcqL5l8Illw

Thanks for the link! This looks super useful. Do you happen to know any real life service that is using these techniques?