yazin (@yazins) header
yazin (@yazins) profile picture

messing with AI / founder @amalinvest (YC, W22)

Joined March 2012
from 107 mistakes, down to 69 overnight context: i'm training a small model to correct Quranic recitation mistakes people make, 100% offline that process was manual up until yesterday. then, i asked opus-5.5 to take the wheel how it works: - opus writes the experiment cards (targeted improvements on the listener, tracker, or the checker) - grok 4.7 compares to past attempts; drops dupes - card execution is a/b tested between grok 4.7 & astra; builder can't see exam data or access modal - grok 4.7 checks for benchmark special-casing; rejects overfitted attempts - tests run on modal, scored on 100 practice sessions - experiments merge only if > 3 mistakes fixed what i like most is that i'm seeing truly interesting 'big bets' -- not just quick fixes! (i don't have an ML background, so just asked opus to eli5 all the attempts, which you see bleow)
auto-research v2: - takes big bets (I approve requests > $100/training) - learns from mistakes (hopefully) - runs 24/7
6
3
45
1,313
example attempts
3
130
auto-research v2: - takes big bets (I approve requests > $100/training) - learns from mistakes (hopefully) - runs 24/7
inching closer 👀 won't be long now before we have a great Quran recitation correction model that is fast & accurate, works fully offline and is under 200MB in size
3
1
12
2,009
party ended earlier than expected (turns out "Expires Nov 1" meant at midnight Oct 31, not end of day) ps: /ultrafast was *really, really* nice -- that rush from getting frontier intelligence ~instantaneously... you really need to experience it for yourself! i really dont know if i can go back to waiting for prompts after.
$1,000 in OpenAI credits, 24 hours to use them what would you build? i'm giving $1k per person to the ideas I like most. credits expire Nov 1. drop yours below 👇
1
534
✅ you get $1k; 24 hrs to go, looking forward to seeing this game come to life!
Replying to @yazins
I wrote a Game design doc for a game i used to play back in the day. Id burn it all into that.
1
5
561
✅ you get $1k; please follow so i can DM the API key 24 hrs to go, make it count! @kdegeek
Replying to @yazins
I am working on an open-source agentic streaming box. Experimented building it on linux, now I'm rebasing on AndroidTV. Your own AI subs/tokens Home control No spying Composable, modifiable UI. No crap, just a TV option that doesn't suck, and isn't AppleTV
2
355
✅ you get $1k; please follow so i can DM the API key 24 hrs to go, make it count!
Replying to @yazins
Create an engine that properly finds rhymes in text and highlights each scheme/rhyme a different color. Creating this is extremely difficult and needs iterations and verifiable testing. Determining rhymes via text without a voice to guide it, is nearly impossible, but I hope to make it 90-95% accurate
2
8
951
✅ you get $1k; please follow so i can DM the API key 24 hrs to go, make it count! @TedHw8ng
Replying to @yazins
im building multiplayer ai, basically slack for people and ai agents themselves to communicate and work with each other efficiently. You can steer someone elses ai agent and make sure ai agents from different people work together
2
6
529
$1,000 in OpenAI credits, 24 hours to use them what would you build? i'm giving $1k per person to the ideas I like most. credits expire Nov 1. drop yours below 👇
🚨 i've got ~$10k of OpenAI credits expiring this weekend; what should I do??
47
2
40
6,025
🚨 i've got ~$10k of OpenAI credits expiring this weekend; what should I do??
16
1
19
7,563
ps: tried /ultrafast already; it's still not burning fast enough!
3
323
i've recently become obsessed with figuring out storytelling via video there's one question in particular that I want to answer: can i create a video people would actually enjoy watching, with just a single prompt? starting this thread to document my attempts 👇
6
1
11
1,720
2/ tried a different approach; this time, pulling out the big guns. used grok-imagine-2.0 with 11labs for the voiceover + effects. instead of picking a random story, i used Surah Al Fil for inspiration. i started by having GPT assemble the context for the storyboard, from the Quran, related hadiths and the reasons for revelation. then, i had it create a storyboard in JSON format. i then fed that storyboard to opengrok (via the new CLI i developed), which it used to generate the clips. i inspect the output and was. .. disappointed! the clips individually were okkk-ish, but they just felt assembled. they didn't really look like they belonged together. i figured there were 2 problems: 1. the audio: it didn't follow across clips, adding to the disconnectedness 2. the cuts: they were all these subtle zoom-ins, which grew old and overused after like 2 uses i told opus 5.5 about this, and it did the first take i actually thought was ... decent? here have a look (to the right: the open-grok editor, completely controlled via cli): ps: thanks you @noornet, @Mamoon89 for the tips
1
2
162
3/ omg, what a difference eleven labs v4 makes!
95
guys i finally figured out how to keep coding agents running EVEN with the laptop lid closed it's called "herdr" just run 𝚖𝚘𝚜𝚑 𝚖𝚊𝚌-𝚖𝚒𝚗𝚒 -- 𝚑𝚎𝚛𝚍𝚛 on the laptop, and... close the lid at will! 🏃 agents continue running on my mac-mini 🛜 seamlessly reconnects when wifi is off/flaky ✅ even localhost links work directly on my laptop! turns the laptop to a 'thin client', with all the *actual* work done on my mac (added bonus is my mac stays up to date!)
20
2
46
7,928
if you're sitting on premium domains, unload them ASAP. should be obvious that ai makes fancy (esp one word ultra premium dictionary) domains worth less
2
13
2,797
Alhamdulillah, Rusūkh is live 🌱 The hifz planner that fits your week. Recite from memory and the flow checker follows you word by word, fully offline. Built on tilawa.dev by @yazins, jazakallahu khayran 🤍 apps.apple.com/app/id6784409… nitter.cf/yazins/status/21080736…
66
cat's out of the bag! long time in the works but im so happy its finally out. what a journey this past year has been. no better time to come out of stealth than to say we got acquired by @askchapter! at ridgeway, we helped companies get the most out of AI and make their AI workflows and agents faster, more performant, and more cost-efficient across their internal and external business. @askchapter was one of our first customers and partners, and together we helped more seniors get the right access and help with their medicare faster than ever! today, we finally announce that we are taking this to a whole other level together! none of this would have been possible without our friends, investors, and partners. i am immensely grateful for your trust, mentorship, and most importantly, the time you have invested in me and the team! @SusaVentures @chadbyers @shaheersan @emergencecap @YazanElBaba @kabirsial @richzou @ChrisZarou @ItzSuds @_samirism @henrythe9ths @omnibusofomerta @DennwsLee @kylejeong @vincentzhu @svathsasa @dojideon @tomzheng @jaegpark @advith_c @MahapatraChirag @yazins @aymancooks @calixo888 @anaiskkillian @rowans_planet @0xRaduan @risashimmura @babykeem (bunch of whom I can't find your twitter handles but don't want to forget: Scott, Alan, Peter, Chonghao, Akhil, Jason, and my brother Rehan; if I miss you, plz do kick me next time you see me) just the beginning!! excited for what's next :) P.S. typing this from ridgeway plaza with @CobiBGantz
Excited to welcome @hamostaf04 and crew to our team. We need the best engineers to build the best tech products for seniors.
105
16
4
274
13,361