one func man

"Now that I’m older, doing well personally is not enough to get me up and going in the morning. I’ve started to focus on service to my community." - @RukaiyahAdams

@JoeDudekJD @Jungroan youtube.com/watch?v=W0_KJq…

nimis being 200 divines makes me sad that my unique is so inaccessible for more casual poe players who just want to have a good time efficiently leveraging a clever use of intended mechanics :(

Numbers like this out of Quickwit don't surprise me anymore. We run it in Lambda functions and get sub-second on 10s of millions of records doing fairly complex aggregates on multi-field searches and filtering. It's wild.

From the Tortured Traffic Department: Release of #TaylorSwift album leads to music streaming surge, according to @kentikinc traffic stats. kentik.com/analysis/relea… #thetorturedpoetsdepartment

inside .git wizardzines.com/comics/inside-…

I realized chess pieces can be redesigned to be purely geometric attack directions

I just scored 177 in #FlapjackFwop! Can you beat me? cassidoo.github.io/flapjack-fwop via @cassidoo

@theburningmonk They have a waitlist / application process gating access to GPT4 via API.

@onefuncman Also, does that usage limit apply to API users? I imagine people building legitimate apps with their API would need more than 25 API per 3 hours as all.

Am I understanding it right that Alpaca is so cheap to train only because they can use a superior LLM to fine-tune it? So training an LLM hasn't become drastically cheaper, but copying an existing LLM and making a slightly worse version of it has? ie. a poor man's ChatGPT?

@theburningmonk @focusotter @MrConerMurphy In order for the lambdas to reach out to NewRelic — I was using a VPC endpoint to access DynamoDB up until then.

@theburningmonk @focusotter @MrConerMurphy I only needed VPC for Opensearch at first, then I had to deploy lambdas into that VPC, then used my 2nd OS cluster as a chance to prove out the extraction from my stateless stack, and THEN I extracted the 1st VPC to another stack when I realized I needed to add NAT to both…

I migrated one of my CDK repos away from using multiple stack (APIStack, DBStack, FileStorageStack, etc) to instead use a single stack What I found: - This is a simpler mental model - Multi-stage deploys are simpler with cdk.context.json + gh actions Any other tips/gotchas?

@theburningmonk They launched GPT4 with rate limits to prevent these attacks and keep access somewhat fair. It was 25 messages every 3 hours, I think you need tens or hundreds of thousands of messages to use this method

@onefuncman ah, right, combining multiple LLMs, and you might end up with a better LLM than those you used in the fine-tuning, now I get it. I wonder how soon would ChatGPT update their terms of use to say you can't do this, to protect their competitive advantage

@theburningmonk @focusotter @MrConerMurphy Fortunately I made my Opensearch environments completely ephemeral. Friends don’t let friends use search as the primary db. Still couldn’t get around deleting and recreating a vpc to fix nat issues though.

@onefuncman @focusotter @MrConerMurphy It takes time and practice and sometimes making mistakes, I've made many similar mistakes before I settled on an approach that works for me most of the time. And sometimes things change, and you still have to adapt! It's not you, this stuff ain't easy

@theburningmonk @focusotter @MrConerMurphy I’m still kicking myself over putting opensearch in my shared backend infra stack before I got the concepts a little more cemented in my thoughts I guess. Unresolved stack organization angst, ha ha

@onefuncman @focusotter @MrConerMurphy Come to think of it, I can probably do a decision tree, like this, which would hopefully clear things up? (super rough 1st draft)

@theburningmonk @focusotter @MrConerMurphy This follow-up article is what I meant by your caveats. Having worked with lots of AWS services, I can generally intuit which ones will cause trouble being commingled in a CDK project, but for newcomers I don't think it's clear between serverful, stateful and serverless services

@onefuncman @focusotter @MrConerMurphy Aha, that's why I wrote a follow-up post about serverful resources like RDS: theburningmonk.com/2023/02/how-to…

@focusotter I've been tying stuff together with ARNs as StackProps out of laziness, but SSM works better than CloudFormation exports.

@focusotter I've been using a single stack for each CDK project but I split out to separate projects pretty quickly. Current project ended up with separate CDK repos per domain and/or team ownership, with one central "backend" repo for easily managed stateful components like Dynamo tables

@theburningmonk @focusotter @MrConerMurphy I've had trouble reconciling what you say your position is vs the caveats you express around RDS, OpenSearch, etc. I separate stacks based on what things should change together or not. f/e, I never want my OpenSearch stack to change when I'm changing the lambdas that index docs

@focusotter @MrConerMurphy I do the opposite actually, I keep stateful and stateless resources in one stack as much as possible. theburningmonk.com/2023/01/this-i… Cognito might not be a fair example because users don't usually belong to one service, so it shouldn't be in a service stack to begin with.

@NatTowsen Weird Al Yankovic, full stop.

What's an example of a genre parody that ends up being a good execution of the genre?

@andrew_westwick @BrantAshe #sorrynotsorry