Eran Mazur

Software Engineer | Ex Military Sniper | Dive Master

[Eran Mazur @EranMazur]

@zero_to_seed It never stops..

I got A6000 at first

Now i have 1xA6000 and 2x5080 and i can’t get enough

[Eran Mazur @EranMazur]

@lauriewired Why the fuck do people would need more than 8TB RAM per server? Slap your MF face and tell me what use case would require that instead of horizontal scaling?

[Eran Mazur @EranMazur]

@kyzoroXX @PeasantSmith Not exactly. It all depends on the harness and the configuration because if you need a shitload of work throughput then eventually you would reach a ceiling here. Though this is one hell of a setup which for 90% is a complete waste of money.

[Eran Mazur @EranMazur]

AI made building software trivial, but it didn't make human needs any easier to decode. When anyone can deploy an agent in 20 minutes, the hard part isn't making something that works. it's finding a problem that actually matters to the masses.

We have unlimited execution chasing a shortage of real value.

[Eran Mazur @EranMazur]

@bridgemindai Imagine running Claude Code all day trusting AI to write your production software but not knowing how to read a pricing table for 12 consecutive months.

[Eran Mazur @EranMazur]

If you’re paying $100 a month to get rate-limited before your coffee even cools down, you don't need Claude Code you need an IQ test

[Eran Mazur @EranMazur]

Claude Pro: More usage*
Claude Max: 5x more usage than Pro*

* The usage in question: 4 minutes of actual work, followed by 4 hours and 56 minutes of staring at a locked input box while your usage resets.

[Eran Mazur @EranMazur]

Unpopular opinion: people who pay $100 for a 5-hour quota are dumber than the execs who came up with the tier in the first place.

[Eran Mazur @EranMazur]

@matthias_mrc Who's dumber: the company that sets a 5-hour quota you can burn through in 4 minutes, or the person who keeps paying for it?

[Eran Mazur @EranMazur]

You are either ignorant or this post is just meant to get you some engagement from the community.

either way, i think you are missing the point of not being limited by a 3rd party usage/whatever rules they set.

the model people need varies by their task, most people today don't really know what model they really need and just sort out to what's available in a few clicks.

the truth is you don't have to pay thousands of $$ for a work that can be done for 100$ by a less generally smart model that excels at the work you need.

i'm not gonna mention all the side pros that you get with running a model locally because those together are not sufficient to justify the initial cost of getting a worthy hardware to run such models but rather act as a nice-to-haves though for some one of them is a crucial selling point for this decision.

[Eran Mazur @EranMazur]

@kapilansh_twt vibecoders who built some shit and showing their sourcecode be like

[Eran Mazur @EranMazur]

@kiri49x86 @Alibaba_Qwen I actually like the fact that their open source releases have a big leap from one release to another. feels like a cumbersome work now to just release another training checkpoint of the model.

Qwen4 gonna be big!

[Eran Mazur @EranMazur]

Now that everyone letting the agents do the work, people got more time on X @elonmusk.

[Eran Mazur @EranMazur]

@jun_song What about qwen? the leader of open source ai community.

[Eran Mazur @EranMazur]

Qwen3.8 27B being positioned 15th and at the same level as gpt5.6 luna which is presumed to be 500B MoE is insane.

Qwen team is really on to something other labs just struggle to understand while also being super committed to the open source community.

never thought i would see an open sourced consumer-level model that is an actual frontier level released from any lab.

I barely wrote any code with Qwen3.6 27B, now i don't write any code, i just steer the ship and design the flows i want. Qwen3.8 27B does the rest.

[Eran Mazur @EranMazur]

@DJSnM BO is so desperate for cash flow the marketing is insane

[Eran Mazur @EranMazur]

@edzart Amazing people still pay for claude when they can invest in their own hardware and run a good enough models for their task without any limits..

[Eran Mazur @EranMazur]

@AbhiCodes15 This is actually a very popular opinion and mainly among non technical vibe coding tweebs..

People be burning thousands of $$$ trying to build shit only to realize that its built like shit because they just don’t know anything about architecture and fundamentals..

[Eran Mazur @EranMazur]

@Alkhalas_Hadee @AmichaiChikli @BabakTaghvaee1 Shut the fuck up you piece of shit! You monsters are a cancer in this world.

[Eran Mazur @EranMazur]

@JoelDeTeves What is this kind of abomination?