Using a coding plan, haven't noticed any throttling and very happy with the performance. They publish the quotas for each of their plans on their website [1]:
Previously I built a jar/docker and deployed to a common server. Fine.
Can this service deploy this to AWS free tier? Will it "know" how to avoid any paid services? (I don't necessarily need a docker or even permanent storage yet)
EDIT: Also - charging a % of AWS bill is pretty wild...
The cheapest way to deploy a container accessible over HTTP is to use web-service resource (which uses AWS ECS loadbalanced with HTTP-Api-Gateway, which is pay-per-use) and use EC2 launch type (cheapest instances are only a few $). Not sure if it's still under AWS free tier, but definitely not expensive.
Regarding charing % of AWS bill - yes, we're changing that soon. It's going to be flat fee + 10% of AWS bill.
Our free tier covers 90% of vibecoder, freelancer and side-project use-cases though. So probably nothing to worry about unless you're a 2+ person business. And at that point, it's still far less expensive than having a DevOps person or learning all of the 150,000 AWS loopholes yourself.
there plenty of published preflop charts and GTO ranges
in fact, a fun project would be take a non-reasoning model, play on a lesser known game format, and see if it learns an "a ha" moment or explicitly simulate moves ahead
> food waste from various stores brought to a facility to be fed to insects
a. how does that solve the transmission problem?
b. amazing work by EU bureaucrats to regulate businesses that dont exist yet
c. they can export the feed to fish farms or china or whatever. the question is do the economics work. US soy bean is just incredibly productive (and subsidized)
This is less about masked modelling and more about reverse-curriculum.
e.g. DeepCubeA 2019 (!) paper to solve Rubik cube.
Start with solved state and teach the network successively harder states. This is so "obvious" and "unhelpful in real domains" that perhaps they havent heard of this paper.
The original report by Ember [1] is decent but clearly biased.
They assume each battery cycles entirely EVERY day - even in winter. They also assume PV is never curtailed - not even in summer. They of course ignore multi-day weather anomalies. Like wise for weekend/holiday demand variations. etc.
The best part of the report are real world bids of 2025 ESS projects.
Or did he "resign" since Elon insists on camera-only and Karpathy says i cant do it?
reply