TheDavidTai

Founder, hacker, stochastic tool.
Checkout my github actions emulator:

[TheDavidTai @TheDavidTai]

notdownhub, is a tool I started working on to make your mission critical CI flows work when Github doesn't.

notdownhub is a fully compatible Github actions server that hooks into any git repo or git server and runs your existing CI YAMLs completely free of Github.

Use your own hardware or a cloud vm that you trust. Just add your existing YAMLs to the dashboard or cli and copy over your environmental variables to your machine. Your secrets are stored in the OS keychain.

Get it at notdownhub . com

Please leave a like, repost, bookmark, or follow if you find it useful so others can find it and get the uptime they deserve.

[TheDavidTai @TheDavidTai]

@holytrinity Is this an ASIC?

[TheDavidTai @TheDavidTai]

@trikcode Its going to do a 911 according to the last ad.

[TheDavidTai @TheDavidTai]

This is like saying a band is good because it makes more songs.

[Wise @trikcode]

I can’t believe Anthropic is claiming that ‘lines of code’ is a good measure of developer productivity

[TheDavidTai @TheDavidTai]

Keep in mind this is global times but this is basically how I see Apple doing it and saving all their money unlike every other megacap. Commodity level AI has no moat besides cost.

[Global Times @globaltimesnews]

Apple Intelligence has been formally integrated with Alibaba's Qwen model on macOS, according to Apple's Chinese website on Saturday, enabling voice assistant and text-generation functions and marking a significant step forward in the rollout of Apple's AI services in the Chinese market. Global Times found no equivalent support yet on iPhone or iPad. Apple is also working with Baidu on AI search capabilities.
https://t.co/21i9c66AsQ

[TheDavidTai @TheDavidTai]

@0xSero

[Avi Chawla @_avichawla]

https://t.co/HTVp6zvP3v

[baris @GBminA]

@TechMDAI @mr_r0b0t @Blackwellboy @NeoAIForecast @Tech2Wild @MichaelGannotti @aijoey https://t.co/SUxlBPIOBD

Maybe the vision solution I implemented here could also be effective; I’m definitely using it without any issues and haven’t noticed any difference from the native vision system yet. The Qwen3-VL-30B-A3B model was used for the vision feature.

[TheDavidTai @TheDavidTai]

Shape is somewhat different than the old one so that's a little bit of a surprise

[TheDavidTai @TheDavidTai]

Time for a new MTPLX pull request.

https://t.co/kiomgDC5ml

[TheDavidTai @TheDavidTai]

@Blackwellboy @NVIDIAAI @ViC305 https://t.co/S0JkuEnTQx

[TheDavidTai @TheDavidTai]

@Blackwellboy @NVIDIAAI @ViC305 https://t.co/es01reuqc6

2TB ($18k) and I think its set up for a 6000 + 6x 3090 or something.

[TheDavidTai @TheDavidTai]

The difference between renting and owning the means of production. Even if its aspirational (which it isn't since we're seeing price coming down on the minimal viable setups especially since May), I think you should try because I don't want to deal with an API and metering for everything.

[TheDavidTai @TheDavidTai]

@Tech2Wild That'd be a fun setup since you can get so much TPS across 8 concurrent runners.

[TheDavidTai @TheDavidTai]

@Blackwellboy @NVIDIAAI @ViC305 How's ram offloading because I'm looking at the 2tb of ram board?

[TheDavidTai @TheDavidTai]

Sol, are we about to be cancelled?

[TheDavidTai @TheDavidTai]

Just use Zod.
https://t.co/cTVV7EXAH4

[TheDavidTai @TheDavidTai]

If you've never done runtime type checks on javascript, this is the biblically correct way to do it since typescript is not run time check. You don't really need to do this but technically its correct to do this with unverified inputs. You should probably just use Zod and hide all this stuff behind something human readable.

I'm QTing this because I'm getting way too much engagement on my original comment and I'm not sure if people actually understand what's going on.

[ben hylak @benhylak]

it's important not to forget that this is what AI written code looks like.

[TheDavidTai @TheDavidTai]

@Blackwellboy @NVIDIAAI @ViC305 Do they saturate the full 16 or can you use 8x splitters for 14 or 16 depending on board?

[TheDavidTai @TheDavidTai]

Time for a new MTPLX pull request.

https://t.co/kiomgDC5ml

[Sudo su @sudoingX]

the ai bubble is smaller than it looks, and shallower too. millions argue about which lab is winning, and almost nobody has gone one layer down: who owns the place where you think.

you don't use these models like software anymore, you think with them. and if the model is rented, your thinking has a landlord. rented minds get repriced, deprecated, updated into something else overnight. the one you spent a year calibrating can vanish with a changelog, and you never had a say.

owning your thinking starts with owning metal. weights on your own disk run identical in ten years, nobody can revoke, reprice, or rewrite them.

[TheDavidTai @TheDavidTai]

@loktar00 How does training work on these inference optimized setups? I was under the impression that training is the domain of GPUs only due to their insane memory bandwidth.

[TheDavidTai @TheDavidTai]

When you slide into her dms and she's already in a committed relationship.

[maria @maria_rcks]

Tibo out here striking fear into the other labs