1,520 karma · joined December 7, 2012
hiring@spikelabsinc.com
Can also use `JSDOM ---> element.textContent` depending on your needs. Useful for snagging all the text content or a specific element's.
The main costs were:
- gpu time for training
- gpu time for inference
- storage costs for the users' models
- egress fees to download model
I ended up using banana.dev and runpod.io for the serverless gpus. Both were great, easy to hook into, and highly customizable.
I spent a bunch of time trying to optimize download speed, egress fees, gpu spot pricing, gpu location, etc.
R2 is cheaper than s3 - free egress! But the download speeds were MUCH worse than s3 - enough that it ended up not even being competitive.
It was frequently cheaper to use more expensive GPUs w/ better location and network speeds. That factored more into the pricing than how long the actual inference took on each instance.
Likewise, if your most important metric is time from boot to starting inference then network access might be the limiting factor.
I got a few of them at aws reinvent in like 2012.
2nd favorite is a Cassandra one of similar make.
You have built a really great product that is easy to use. I've tested out the self-hosted option on render.com with neon postgres and the cloud version. Both are super easy to get up and running.
Now, it is no exaggeration to say that the most annoying part of adding a new integration is just navigating the API provider's portal to get the app outh creds.
It should lead to growth/revenue/whatever your north star is.
IRL if you can - they're awesome.
The one in Dallas is huge. Like a gamestop for dvds but the size of a warehouse. It is pretty close to a Half-priced books, same concept but books, and hitting the two of them back to back is a fun way to spend a day.
Sounds like it was just what was needed at the time and worked better/longer than most abstractions.
Rallying devs and their community of users is worthwhile. My advice was more for how to respond and handle the review process internally with google and Apple based on what has actually worked ime ¯\_(ツ)_/¯
Did they slip by a reviewer that didn't care? Did they use their "VC" illuminati connections? Are they such a big player the rules don't apply? It doesn't really matter or have anything to do with getting YOUR app released and the reviewer you are talking to will most likely ignore that completely.
Your best bet is to remove or change the mechanic if possible, resubmit, and then add it back in during a future update if you feel strongly about it.
It is possible those other apps have something that allows them to keep it... and by communicating nicely and earnestly with your reviewer you might find a way to skirt the rules, if a way exists. For us, it was frequently tweaking copy or TOS or other minor changes that allowed us to release an app with only superficial changes that the initial review made it seem like we were totally screwed. I do get the instinct to just say FU though.
We signed up before they had an admin dashboard and did everything over cli. It was easy then and even easier now. We have also contributed a couple of providers when they didn't have apis we wanted to use setup yet (Stripe, Zapier NLA). They have an open slack channel and we were able to get them shipped immediately.
Other than that not getting much utility out of it but that’s fine
Custom authorizer in front of aws lambda with serverless framework works great with amplify’s frontend api client.
We also use dynamodb serverless but not with app sync and no graphQL. Would I rather have a good and easy to use serverless Postgres db that played well with aws lambda. Yes. Am I going to switch away because I have to plan my access patterns and indices out carefully… no.
Google must be feeling the squeeze from all over.
Q&A and summarization it will be easy to see improvements as current recursive summarizing and embedding techniques are very "lossy" but outside of improving current use cases what will now be possible??
I've had success with a simple system method that is one sentence defining how we are extending chatgpt.
"You are AcmeCoBot an extension of ChatGPT which we have enabled <x, y, z features> to assist users <goal>."
Then a user message with the actual instructions on what you want/need - this is much closer prompting for gpt-3.
user msg 1: Translate the following to German for me... some text assistant msg 1: an example translation
new user msg: Translate the following to German for me...
The model completes these interactions well. We may run 1-10 of these completions before using chatgpt for the last mile message that actually gets sent to a user. It takes a while to wrap your head around using the chatgpt api for a non-chat completion.
It is almost hilarious to see these content creators jump ship so quickly after espousing crypto as the end all money printer.
The screenshotting uses puppeteer and chromium and a read-only session to impersonate the user and screenshot their dashboard.
It uses the old version of chromium and there were many gotchas that required a lot of extra scaffolding to actually render ours and other websites like they would on my laptop. This will hopefully make it easier for us to maintain once implemented.
It isn't clear the hierarchy or size of your team or the longterm strategy/goals of the product development. Someone on the team should be pushing back on these changes as they are being planned out or before they are getting merged in.
Ideally, you have a respected senior/staff/cto engineer that can mentor these "clever" programmers. This person needs to be able to grok all the code and see at a macro level how the whole codebase is evolving.
Re: reading code - I am the same way. I spent years writing objective-c when xcode had the best code completion out there so long names weren't an issue. Long function and variable names and tons of whitespace - it is how all of Apple's libraries were written and how the projects templates were setup ¯\_(ツ)_/¯.
Best recommendation is to adopt a strict style for your codebase and ruthlessly apply it to all new code. Linting should happen automatically. There are tools that will run the linter on just the edited files so you can adopt coding styles 1 file at a time.
If you answer and say hello and there's an extra second just hang up. They'll call back if it is important.
Recently, I was glad I used the right card to book an airbnb which we had to cancel day of check-in. Full refund on a non-refundable airbnb.