jayk56.bsky.social

Also, attachments dont work 😅

new blog out with first impressions on codex app. gist is there's potential, but coming from the cli you'll need to relearn some things jkerschner.com/posts/codex-...

wtaf..

the answer was not quite, and >90%...times they are a changing

what does it mean when ALL sota models score 95%+ with different cli's? (codex, claude code, etc) aider seems to be a drag on their performance now, but we're back to the bigger question of can these be 'plugged in' to software workflows and what percentage of code will they be writing by EOY

as a proud American I'm disgusted by the cowardice the republican party in congress has shown, it's up to us to remind them who they represent. republicans have shown they are worse than those they hate, they have: NO budget NO beliefs NO backbone saturday we remind them we all have #NOKINGS

lots of places across the country this weekend where you can grab a bite, listen to some music, and use your voice with no kings to be found. hope to see you out there!

what does it mean when ALL sota models score 95%+ with different cli's? (codex, claude code, etc) aider seems to be a drag on their performance now, but we're back to the bigger question of can these be 'plugged in' to software workflows and what percentage of code will they be writing by EOY

please help fill in some blanks! openai: really gets the ergonomics of an ai chat interface (ChatGPT) anthropic: really gets the ergonomics of ai agents (Claude Code, MCP) google: really gets the ergonomics of __________ deepseek: really gets the ergonomics of ___________ #ai #llms

using claude with artifacts in the web client, I'm not surprised claude code gets this nickname

@anthropic.com what are you guys feeding him?? this model is cracked.. claude.ai/public/artif...

help! i just asked claude opus to build bloxorz and it's begun optimizing it in a dev loop. it hasn't let me test since version 4...

Another fun fact is the new Gemini 2.5 Flash (non-thinking) preview model performs on par with Claude 3.7 Sonnet (non-thinking) on this benchmark for 1/10th the price...

Gemini 2.5 Pro is a beast for the price and, given 8 tries per problem, scores 93% on the Aider benchmark. o4-mini (high) was not far behind, but cost about 30% more to score 91%. Sonnet 3.7 without thinking is able to average 87% but cost about $45 per run (more than double the cost of gemini)

joelgrus.com/2016/05/23/f...

thankfully, the people and USA continue to show it is strong enough to rebuff the would be dictators and totalitarians. building your communities only strengthens that defense, and it helps remind us that we're all in this together.

Mind blown moment: Apple's unified logging system captures EVERY framework and system component your app touches. 🤯 I had no idea! In my new post, I reveal how to tap into this goldmine of debug context – it's all there waiting for you. 👇 www.fline.dev/better-error... #Swift #iOSDev #SwiftUI

"it's just statistics and orthogonal projection.." i mumble as the text predictor doing my work continues to confuse me

Almost... follow here to get the next installment in this saga

what do we think? line go up? #aider

what does it mean when a model can score 95%+ on aider's polyglot benchmark? how does it change what you can do with it vs the 70% gemini pro is at now?

Things I didn't have to think about 6 months ago: - Do I need to pull up expected purchases to avoid paying 30% more? - Will I be harassed coming back from a vacation out of the country? - Will congressional republicans grow a spine before it's too late? what are some new worries you didn't have?

why I'm protesting: we have a government official trying to claim king like powers to punish the out groups and playing chicken with our allies. I reject this. team trump: RESPECT our people RESPECT our constitution RESPECT our allies #handsoff #handsoff2025

Hundreds of SacTown locals showing up today to protect veterans, retirees, and science and protest oligarchs carving up the American public's wealth. let's turn those honks into action with the Hands OFF! protest April 5th. #SacTeslaTakedown #TeslaTakedown #StandUpForScience #HandsOff

wondering today-if LLM are the computers-when are we going to see the 'office suite' of AI get launched, is this going to be limited to Gemini/Apple Intelligence integration? Is it what OpenAI is building out to serve the existing chat customers? operator, search, deep search, canvas/projects, etc.

just told a barista that AI is going to take my job in a few years. doing my part to accelerate the hype machine 🫡

VSCode + AI = ultimate text calculator! I'm sure this has been a feature for a few releases, but I wanted to merge a file from my clipboard into an existing file, BUT only change a couple classes. Using the 'Compare File with Clipboard' I was able to select the individual sections to apply. ezpz

How these new reasoning models are prompted is still so important that domain expertise still gives you a massive boost over the average person in getting a 'good' answer from the models. The next phase is probably looking at how to get better results with less specifics in the prompts.

My favorite part about this is the code at the end where it implements BroNumberCruncher()! minimaxir.com/2025/01/writ...

OpenAI's Realtime API feels heads and shoulders above Gemini 2.0 Flash live, but it's also possible it's just easier to get working examples going. I was comparing an example react component with 4o-mini against Google's example python cookbook with gemini 2.0 flash and flash just felt shallow..