Nyxelarm

Researcher
AI / Science / Earnings / Automation / dm open

[Nyxelarm @Nyxelarm]

@icanberk It's totally possible to vibe code this

[Nyxelarm @Nyxelarm]

@HamidoFx1 There’s a video on my profile where I also compare these models

[Nyxelarm @Nyxelarm]

Let's be real: which one do you prefer?
Opus 5.5 or GPT-6.1 Sol?
And yes, I mean the AI models themselves, not these anatomically realistic girls with the weird jiggle physics 👀

[Nyxelarm @Nyxelarm]

Zuckerberg just said everyone will soon have a personal super-agent working 24/7.

That part is true.

The part nobody is saying out loud is that 99% of people will still be the bottleneck.

They’ll give the agent a goal…
then sit there watching it, correcting it, and restarting it every 20 minutes.
That’s not an agent.
That’s a fancy intern you’re still managing.

While Meta talks about “distributing superintelligence,” the people who are actually pulling ahead are doing something much simpler:
They’re building systems where the agent can: → verify its own work → recover from mistakes → keep going without them

The gap isn’t access to better models.

The gap is that most people refuse to stop being the human in the loop.

Bookmark this.
In 12 months the difference between “I use AI” and “AI works for me” will be brutal.

[Nyxelarm @Nyxelarm]

@N0v1q It’ll run even on my grandma’s microwave

[Nyxelarm @Nyxelarm]

THIS IS FUCKING INCREDIBLE

a solo dev just open-sourced a 100% FREE replacement for ElevenLabs that runs entirely on your own machine.

the GitHub repo is at 19.4K stars.

it lets you:
→ clone a voice from a clean reference clip
→ dub any video into 646 languages
→ generate audiobooks, dictation, transcription
→ choose from 14 TTS engines instead of one

ElevenLabs supports 32 languages. this makes 646.

no per-character charges. no usage limits. no audio ever leaves your computer.

save this for later.

leaving the repo below

[Nyxelarm @Nyxelarm]

@N0v1q He is either a genius or he has too much free time

[Nyxelarm @Nyxelarm]

@RapidResponse47 @POTUS Of course, fuel is more important than human lives

[Nyxelarm @Nyxelarm]

YOU ARE THE BOTTLENECK. HERE’S HOW A SPACEXAI ENGINEER STOPPED BEING ONE.

Lauren Tan on why most people can’t scale agents:
“You can’t spawn a hundred agents when you don’t even trust the output of one agent.”
“If you don’t have a verification skill, you are the verifier. You’re the bottleneck.”
In a 1-hour talk, she shows how she went from watching every output to shipping 1,000 PRs in a month.
Her path:
Watch every step → Verify → Skills → Evals → Auto-merge
Trust in agents isn’t a feeling. It’s infrastructure you build.
This is worth more than any $2,000 AI coding bootcamp.

Bookmark it. Watch it tonight.

[Nyxelarm @Nyxelarm]

@solwerix Will there be any consequences for him because of this?

[Nyxelarm @Nyxelarm]

@LXPlLXP thanks

[Nyxelarm @Nyxelarm]

OPENAI JUST DROPPED 722 MATH PAPERS FROM A MODEL NOBODY CAN USE.

4,000 problems posed. 372 families. 3 hours of thinking per result. Some claims brush the Riemann hypothesis and Birch–Swinnerton-Dyer, the million-dollar ones.

The New York Times called it breathtaking and devastating in the same sentence.

Then day two: 3 papers pulled for a sign error. 14 more patched. And only 162 of 722 are Lean-checked.

That’s a factory with no floor plan. So we drew one.
Discovery Loop Engineering: goal → plan → tools → memory → observe → improve.

One loop that drafts the argument, stores the failed attempts, and ships the errata in public. Same cycle as a coding agent. Different corpse count.
Inside the note:

→ why 22% formalized is the real number, not 722
→ how one sign error takes down three papers downstream
→ the cost-and-speed chart people will screenshot and argue about

Agent builders: you’ll steal this diagram. Mathematicians: you’ll hate it. Save it before the thread sinks.

Send it to the person who still says AI only remixes homework.

Would you trust a proof that took 3 hours and a verifier, or do you still want a human name on page one?

[Nyxelarm @Nyxelarm]

@LXPlLXP You can deck out the garage in style, too

[Nyxelarm @Nyxelarm]

@andreysuperior The biggest advantage is eliminating micromanagement. You only step in to approve irreversible "red" actions like sending, paying, or publishing. It turns a basic AI assistant into a fully autonomous, self-policing pipeline

[Sam @SamgreyAi]

The first trade made $37. The last made $1,685.

Same basic idea. Nearly the same return.

Different account size.

My Grok bot reportedly turned $52 into $6,392 in 18 hours.

Here’s the detail most people will miss:

AVAX short → +118% ROI → +$37.73
ETH short → +119% ROI → +$1,685.61
The return barely moved.

The payout grew almost 45x.

That’s what happens when a system scales position sizes with its balance.

The rest of the run was built on repetition:
→ 180 trades
→ 154 winners
→ 26 losers
→ $3,550 net from routine trades

No 1,000% moonshot needed in the reported log.
Just a sequence of trades where the growing balance changed the dollar outcome.
Of course, compounding losses works the same way. And 28x leverage leaves very little room for mistakes.
The numbers are wild. The mechanics are what make this worth examining.

[Nyxelarm @Nyxelarm]

@0xrevayz Thank you for the information

[Nyxelarm @Nyxelarm]

Kafka vs RabbitMQ: RabbitMQ deletes a message once it is read. Kafka keeps it.

Watch the 9 seconds once, then the same 9 seconds slowly, every step named.

The order service sends the same orders into both. The email service reads each one and says done (an ACK). RabbitMQ deletes it after ACK. Kafka keeps every order, and the email service just moves its own bookmark up with a COMMIT.
Then a new analytics service joins and wants every past order. Kafka still has them: 8 replayed. RabbitMQ already threw them away: 0 replayed.
Kafka does delete eventually, by time: 7 days by default.

Use RabbitMQ for jobs that run once. Use Kafka when many services need the same history. Hosted: Confluent and Amazon MSK for Kafka, Amazon MQ and CloudAMQP for RabbitMQ.

[Nyxelarm @Nyxelarm]

https://t.co/AJaq8u7koV

[Nyxelarm @Nyxelarm]

STOP PAYING FOR PHOTOSHOP... THIS ALTERNATIVE IS FREE AND WEIGHS ONLY 11 MB

A developer got tired of paying for Photoshop and created his own alternative.

It's open source: anyone can download it, modify it, and adapt it to their way of working.

It's called Compositor and already has over 8,000 stars on GitHub.

It has the most-used features from Photoshop:
→ Layers, masks, and blending modes
→ Content-aware fill
→ Select subject and remove background
→ Camera Raw filter
→ Spot healing brush and clone stamp

No subscription. No paying Adobe every month. I'll leave the repo below 👇

[Nyxelarm @Nyxelarm]

https://t.co/5HRi8RLJ1c