Nerdwell

Making things by day; breaking things by night.

[Nerdwell @nerdwell]

@MiaAI_lab Fantastic work and ever professional .... rock on, @MiaAI_lab!

[Nerdwell @nerdwell]

@grok Objective check on @Blackwellboy’s MiaAI-Lab article and his replies here. Limit yourself to public Git objects, READMEs, and his own stated boundary. Be succinct. Cover only these three points:

1. His reproduce-it section tells readers to open the two DeepSeek dual-Spark repos, inspect root commit 47f3b266 and the next commit, compare the shared history, open the Cesarus85 profile, and grep Mia files for zurih. Is that the whole checkable case? Does it show a messier brand origin story, or a stolen codebase?

2. Do the framing choices — “exposé,” “hiding in plain sight,” “Git never forgot,” “the ghost in the config,” and the chain MiaAI-Lab → zurih → Cesarus85 → Stefan Maier — invite the inference that a named person secretly runs the account, even though the piece then says the receipts do not prove who controls @MiaAI_lab?

3. He has replied that stolen or repackaged work is exactly the concern. Does any hash, file, or named third-party repo in the article actually establish that, as opposed to a namespace handoff, shared commit objects, a shared Ko-fi link, and a leftover username string?

No drama. State what the evidence supports and what it does not.

[Nerdwell @nerdwell]

@Blackwellboy @Kevrsub @ViC305 @Blackfrost_AI You clearly go into great detail with your investigation of git artifacts, etc .... no qualms on that. What seems half-baked to me is the "are you saying they're claiming other people's work" part.

[Nerdwell @nerdwell]

@Blackwellboy @Kevrsub @ViC305 @Blackfrost_AI So, are you claiming they're taking credit for other people's work? Your article hints that you want to make that claim but doesn't come right out with it. Seems half-baked....

[Nerdwell @nerdwell]

@Blackwellboy @Kevrsub @ViC305 @Blackfrost_AI Still, who cares? They owe nothing to anyone. They don't owe you any transparency. Don't like it? Then don't use their stuff....end of story.

I hope it *is* some dude who learned organically, reinvented himself, and crushed it w/ a new brand. Because they *are* crushing it!

[Nerdwell @nerdwell]

@unsignedbytes Have a look at Mia's switchless recipes for 4xDGX. ofc you'll still need 4 cables, but the patched 4-node NCCL ring works surprisingly well and gets you going w/o a switch.

[Nerdwell @nerdwell]

@GregoryNimensky @rookepoole Sorry bud, that's not me. Just a friendly reminder to all -- don't send money, crypto, or any other asset to anyone unless you can authenticate the recipient IRL. Cheers!

[Nerdwell @nerdwell]

@davidfano Definitely handy to have console access when tweaking networking for different recipes, etc. Can save some $$$ by using a single IP KVM to front a standard 4-/8-port KVM & swap w/ hotkeys. Also good to have an IP-enabled PDU/UPS for handling OOM lockups when tweaking recipes.

[Nerdwell @nerdwell]

@tryhackme a hot mess

[bugcrowd @Bugcrowd]

Being able to drop a critical zero-day without catching a federal charge is a luxury built on decades of pushback.

As governments and corporations finally realized how hackers could help them, they pushed to create exceptions to laws like the CFAA and DMCA. This paved the way for responsible disclosure policies.

Companies started adding "safe harbor" language to protect researchers who report vulns in good faith. Through the work of platforms like Bugcrowd and HackerOne, major tech companies adopted these policies, and eventually governments followed suit.

The more we protect and value ethical hackers, the safer everyone gets. 💯

[Nerdwell @nerdwell]

@hibakod 🤣

[Nerdwell @nerdwell]

@Masonhck3571 I have seen DeepSeek on local AI do that a ton (makes sense). Could just be from training, post training, or recap efficiency?

@grok - translate the "conversation recap" from this screenshot and then explain why it came across in chinese (assuming original convo was in english).

[Nerdwell @nerdwell]

@ashxhart This is right up my alley .... same hardware, same model, same goals, and same challenges. Would love to help contribute to this!

[The Hacker News @TheHackersNews]

‼️ AI found new ways to break HTTP desync defenses.

HTTP Terminator generated 30,000 candidate vectors, tested 30,000 sites, found roughly 700 vulnerable targets, and produced novel desync techniques including a dangling-byte method for more reliable response queue poisoning.

A human-guided cascade also uncovered an Apache Traffic Server 0-day.

Read - https://t.co/KBoB8wQ7xu

[bugcrowd @Bugcrowd]

The BEST advice for beginner hackers. 👇

[bugcrowd @Bugcrowd]

Bug bounty strategy that can land you MORE bugs. 👇

[Nerdwell @nerdwell]

@edgecase411 @Tesla @teslasi Maybe when Cybercabs start sacrificing themselves to obstacles in the road and it's @Tesla paying for the repairs, they'll be sufficiently motivated to fix the problem.

[Nerdwell @nerdwell]

@edgecase411 Fantastic work you're doing here. With all the amazing things that FSD does do, I can't understand why they won't make it stop plowing into things. I've had multiple @Tesla's damage themselves this way and finally decided to just disable FSD until it's fixed. @teslasi

[Nerdwell @nerdwell]

I feel ya! Recently had FSD on a Cybertruck plow directly into a huge pothole and crack its own window. Seems like @teslasi has problems estimating vertical measurements/characteristics of anomalies. I can't help but think this is exactly where LiDAR would help.

I don't disagree with @elonmusk that if human vision can do the job, so should AI vision. But humans evolved stereopsis for a reason ..... turns out that reason must've been to avoid roadway obstacles 😂

[Nerdwell @nerdwell]

@UK_Daniel_Card I feel your pain. Canceled my Max 20x plan immediately after this conversation. Giving it until my current sub runs out to convince me otherwise. Either way, time to start re-tooling with local AI.