code_runner

I’ll do my best.

Number of params is the number of weights. Basically the number of learnable variables.

Number of tokens is how many tokens it saw during training.

Vocab size is the number of distinct tokens.

The relationship between params/tokens/compute power is something people have studied a good deal and how it affects model performance. https://arxiv.org/pdf/2203.15556.pdf

I also want to say that an industry doing things on paper and in person isn’t always in need of a high tech overhaul.

Why would a scuba shop want to take on the burdens and issues of your tech solution? They are running a shop already. I don’t imagine they planned on becoming billionaires. Maybe they just like scuba and want to facilitate other people enjoying it too.

Telling ann established business owner you have a “better way” to do something is so arrogant. Not everything has to be hyper technologically advanced.

They’ll only piss off the people paying attention to every drip.

Just to chime in for anyone wondering who Kipchoge is if you don’t follow marathoning.

He is hands down the best of all time and is totally in a league of his own. He trains constantly. His entire life revolves around it.

The ability for the robot to move in multiple directions at once… bending knees and arms and waist at the same time, threw me for a loop.

The only sane approach. If someone is on support they also work to lessen the support load permanently. Recurring issues with no root cause analysis and fixes are deadly to a team.

TLDR: support should absolutely tank velocity, make the org feel your pain. If they want better velocity they need to invest in shoring up issues.

One thing to consider… support tanking your velocity is a feature, not a bug.

Sounds like you need investment in support tooling etc that can lessen that burden. Velocity will improve when support is less burdensome.

This is obviously not one-size-fits-all advice but worked well for a previous org I was at. If the sprint work will have a side effect of fixing support or support is so intense it requires a majority of the team you HAVE to fix that first regardless if sprint/scrum/kanban, etc. if velocity was super high for a few sprints but regressions/bugs were introduced they could impact future velocity, so those high velocity sprints weren’t as productive as you thought. It’s never just one number.

The org I worked at with the absolute worst support/sprint structure also had the only code base I considered “unsalvageable”. They refused to do anything to improve existing processes and spent half of every team’s time on the same support issues on an endless loop. They never had the measurements needed to actually figure out where to improve clients experiences.

Just hit up laravel and get building something quick and easy. I’m sure a lot of the laravel core contributors actually have some good advice for general env setup outside laravel as well.

Because people can make whatever decisions they want based on their own list of tradeoffs. I can tell you from experience laravel is far easier to get going than any c# or Java framework. You can SPRINT through a project in laravel and then break it into pieces after the fact.

I would pick it if I wanted to launch something new without getting stuck in premature optimization hell. Bigger org I probably would not unless I was building an isolated POC.

+1 for laravel as well. I joined a php project for a fresh rebuild in laravel and had mostly .net experience. Laravel was great.

The documentation is an almost perfect blend of examples and nitty gritty detail. The documentation is almost engaging if you’re coming from other frameworks.

I found myself thinking “wow that’s almost no code, what about this feature or that” and the next section would be the exact topic I was wondering about. Had a blast and could learn more php details as I went

I think there is definitely value in both approaches and depending on your preference one may be more fun than the other. My point is not to get caught up in the framework of the day.

Every framework will teach you a different way of looking at things… can’t go wrong picking one and going really deep, then exploring some others. I wouldn’t recommend one specifically, but definitely start with one and stick with it for a bit

Yeah because other car companies aren’t selling “full self driving”

It can if they want it to. It’s a car.

I have a very good friend who is in tech recruiting. I feel for them a lot.

But I also can’t justify this ratio. When times are good they ramp up to support wild hiring.

Even if times are still good you hit some critical mass where the recruiter count just doesn’t make sense.

As much as they can be annoying at times, recruiters do get used and abused.

I believe the training is where the architecture of the model is most apparent. You can absolutely download plenty of pre-trained models.

You will also probably need to fine tune for a specific use case, so a common approach is downloading a pre-trained model and fine tuning.

I think including the “from scratch” tuning script is educational more than anything else.

I worked for a company where everything was cute names with a particular outdoors theme.

All of the conference room names had the same theme. Starting out it was impossible to navigate if we were even talking about a product or a conference room etc.

Obviously crypto companies can get by without any real regulatory considerations for a while…

If the smaller companies I’ve worked for are any indication… don’t get hung up on branding this app. Be able to swip swap a logo and call it a day.

My god the time wasted talking about branding and then a client can’t even provide a decent logo and turns out they didn’t care about the branding anyways.

Don’t waste your time on that stuff unless it’s hyper critical to the core product itself.

I know you’re coming back in a few minutes to cite your sources. Excited to see the results of this survey!

“One of the best things for society” is a very very bold claim. For the uninformed like myself how would psychedelic therapy be THIS beneficial?

Yeah I assumed we were going to get to some sort of types/inheritance based solution as well.

Stuff like this doesn’t bother me a ton unless it’s littered everywhere. If every function can run one of two paths that’s certainly annoying… if there is some high level function that builds one of two “pipelines” and I don’t think about userType again, I don’t care too much.

The shower after a workout is extremely satisfying. Only takes a couple minutes in any case

It’s gross to have a function floating off somewhere that has no true reuse value at all and doesn’t live close to where it is used.

I like Python but this is probably my biggest gripe.

Multi line anonymous functions. An anonymous function is just as good as a true function in JavaScript.

Lambdas are incredibly limited and encourage the declaration of a whole separate function that now lives physically far from its use.

List comprehensions are also really underpowered and I personally hate the map and filter functions and their goofy list-like types.

Check out linq in .net, streams in Java, array in JS, collections in scala to see the expressive power of a fluent interface on collections

Oh wow you should definitely give this a shot again. Type hints in Python save the entire language for me.

People do have bad habits of cramming everything into dictionaries but if you do some type hinting and use data classes heavily you’ll really have a good experience.

I could take or leave mypy personally.

You didn’t buy “crowdfunded yada yada”. You bought a product called “full self driving”.

You did end up paying for Tesla to figure out how to build the product that didn’t exist when sold. (all of which should’ve become a useable product “next year” by the way)

The fact you’re ok with that doesn’t mean everyone else is. It doesn’t mean it was marketed ethically.

I’m not even saying you were swindled. I think teslas seem like nice cars. You’re just making the wrong argument.

They sold everyone a snickers. Just because you’re happy to have nougat and a little Carmel doesn’t change the fact you’re missing peanuts and chocolate.

that may be why you paid, but it is not what you paid for.

It’s improved based on a heuristic for you based on whatever area you tend to drive in. (Maybe interventions is reported by the car after a session, but it’s still a small sample size.)

At the end of the day, you still don’t have the feature you paid for, you’ve paid to be part of a (very risky) beta test. I don’t think that’s how most people buying FSD planned on this going.

When I was delayed and flying delta they continually rescheduled me until I boarded one of the flights.

It was super convenient even if I was fuming over the multiple hours delay.