mesmertech

I think its just a visual/default thing, cause Opus 4.0 isn't offered on claude code anymore. And opus 4.7 is on their official docs as a model you can change to, on claude code

Just ask it what model it is(even in new chat).

what model are you?

I'm Claude Opus 4 (model ID: claude-opus-4-7).

https://support.claude.com/en/articles/11940350-claude-code-...

Not showing up in claude code by default on the latest version. Apparently this is how to set it:

/model claude-opus-4-7

Coming from anthropic's support page, so hopefully they did't hallucinate the docs, cause the model name on claude code says:

/model claude-opus-4-7 ⎿ Set model to Opus 4

what model are you?

I'm Claude Opus 4 (model ID: claude-opus-4-7).

We went from "Peak hours" meaning 2x usage plus slower to now it just does 500 error

https://mesmer.tools/random/is-it-peak-hours

Was playing around with it earlier and the results were actually good with it being opensource and for how cheap the model is to run.

The tool page itself is smth I vibecoded in like an hour. Feel free to give it a try

turns out its for an April fools tomorrow: https://x.com/mesmerlord/status/2038938888178135223

Also saw this on twitter earlier, thought someone was just making a fake hype post thing. But turns out to be an actual prompt for capybara huh: https://github.com/chatgptprojects/claude-code/blob/642c7f94...

Was searching for the rumored Mythos/Capybara release, and what even is this file? https://github.com/chatgptprojects/claude-code/blob/642c7f94...

" let r = ["Acme Corp", "CloudVault", "DataSync Pro", "NexGen AI", "SecureStack", "TrustLayer", "Vanta", "ComplianceIQ", "InfraSec", "ByteShield", "PipelineOps", "CyberNova", "TokenGuard", "ZeroTrust Labs", "Aether Security", "PrismData", "CloudArmor", "RiskLens", "AuditTrail", "ShieldIO"] , n = ["just checked", "searched for", "ran a scan on", "verified"] , a = ["San Francisco, CA", "New York, NY", "Austin, TX", "London, UK", "Berlin, DE", "Toronto, CA", "Seattle, WA", "Chicago, IL", "Denver, CO", "Boston, MA", "Singapore", "Sydney, AU"]; "

fake popups, xyz domain, recent zeitgeist, 100% straight vibecoded. good hustle I have to say. a domain that'll now get ranked on google for SOC 2 compliance which likely has a high CPC and good DR to piggyback off.

I've found that there is just not really any "general" advice for doing FB ads. Most of the advice out there is usually for dropshipping, or some sort of physical product.

So just going through my journey from the beginning with FB ads. Still new to YT so pls bear with my (non-existent) editing but I wanted to go into as much detail as possible because a lot of the content online is very deceiving in terms of how doing ads actually work. It gives you a feeling that all you need is one viral thing with low CPC and somehow magically even if your product's price is low, churn is high etc, everything will be fixed and you'll be a millionaire if ROAS = >2x.

No one really goes into detail about sucking on their first few ads, how to improve, how long you should wait for FB to optimize before pausing an ad, how much you should wait before killing a concept if its not bringing you sales, if you should keep an ad running with negative ROAS(if your product is a SaaS) and so on.

tldr; start with image ads. target immediate pain points nothing generic, calculate your breakeven beforehand. and finally your first 200-400 USD will likely be wasted on learning, both for FB and your own skills.

As a solo indiehacker in Europe, its crazy that I have to be so worried about VAT related things and big tech just goes around the whole thing and doesn't even expect to be charged just fined

Indiehacking so couple of serious things and then some just starting out and some mostly for myself.

Serious ones making over 2k/mo https://bestphoto.ai/ - AI Img/Video tools https://aieasypic.com - Original version of above, still making some money

New things not making money yet https://admakeai.com - AI ads generator for facebook ads. Made mostly so I have a easy way to prompt for stuff and keep track of good AI ad prompts. Have been using it to make static image ads for BestPhoto and actually have over 40 conversions at $~45/conv. Pretty good considering my previous attempts with my own handmade ads using canva was like $80-100/conv. but most of the time FB wouldn't even spend anything from my budget before

Fun stuff/for myself:

https://xhdr.org/ - Made like a few months back when twitter was allowing HDR images in your profile pictures, they patched it like a week later, but was fun while it lasted. Still works for facebook, keeping it up for FB video ads, I've noticed people abusing this a lot when scrolling on FB recently searching for good ads, so I'm guessing it works? Good way to get attention of people on iphone quick

https://framecall.com - Saw people making cool AI motion videos using the claude code skill with remotion, so packaged it into an actual web app you can use as a chat instead of through terminal. Harder than expected to get all the tool calling(and auto continue) stuff to work, similar to how claude code works

TranslateVoice(name TBD) - An iphone voice translate app where you tap microphone button and it uses 4o realtime to translate between you and someone else. This was originally one of the biggest thing I was hyped about as a usecase when 4o released but when I tried it on their app, it just didn't do prompt following well at all it would randomly try to communicate with the user instead of strictly translating, just randomly cutting off someone before they finish speaking etc. Currently have 3 modes that work, that if no one else uses I will be using:

1. Interpreter mode: Basically User 1 speaks to phone in their language, phone talks to user 2 in their language, user 2 replies to phone in their language, phone replies to user 1 in their language etc. Just pure translation, with chat history transcribed in each other's language

2. "Friend mode": You tell it a general goal, "I want to get immigration documents, I need to know the requirements" it then basically acts as if its a friend you called to help with translation and gathers everything while talking back and forth with User 2 and then at the end goes back to user 1 with all the info.

3. Stealth mode: Airpods in, it will transcribe and translate everything being spoken and tell you what to say based on initial goal or you can also write extra instructions in the chat. This is currently the only non-working/buggy one I'm trying to figure out before releasing this, since both will be speaking in same language its hard for model to know who is user 1 and who is user 2 automatically.

First time I'm creating an Iphone app fully vibecoded(react native so I understand whats going on frontend wise at least cause of react). Not looking forward to the app store review process. And I know this is likely something someone already made but its probably paid and I have like $25k in Azure credits I can burn anyways.

You can click on the videos to copy the prompt for each one, made all of them(5 variations per) in one go with a script. Only one was made custom, the "The party told you to reject the evidence" one, which still took under 5 prompts to get there after some fixes using prompting.

Usually I'd say I'd say people are just being superstitious or got used to a model's intelligence. But recently I've had Opus just repeating things over and over a few times and just have lower code quality in general.

The latter is harder to "prove" as its mostly vibes but the former I've been noticing a few times.

Got two websites but the second one is basically a clone of the first with better visuals and better tech stack that I actually want to work on

https://aieasypic.com - 3k per month (declining cause not working on it a lot, just maintenance) https://bestphoto.ai - 2k per month (increasing cause of better SEO)

Now trying my hand at an actual non-consumer product, not that b2b but something to make making ads easy because that’s where I find myself getting stuck on when doing fb ads or TikTok organic stuff : https://admakeai.com

Was curious what it looks like now, and yea, not a fan of the fake hacker "we don't do CSS or styling". But then again maybe I was just used to their old design

Just some observations and experience from an indiehacker who has been primarily using Django for the last 4-5 years

So just to start, I've been suffering from backpain in the recent few months due to sciatica and it has basically killed my motivation for coding after I'd already been suffering mild burnout, with my usual workflow before being just sitting for long periods of time and focusing on coding.

As a recovery thing I've been walking for long periods of time, ~2-3 hours per day which ends up making me feeling unproductive because my focus time is at mornings and if I don't start coding first thing in the morning I'm not motivated enough to start in the afternoon.

All this preamble to explain why I had the thought of coding on an ipad in the first place. With Claude code I've noticed that most of the time I'm "coding" is just me inputting text and waiting for it to actually implement the change. Which is why I had a thought, if simple sentences are all I'm inputting, would it not be possible to use dictation on ipad and work on my side projects that way instead of typing it out. I got into detail about my setup in the video itself. Apologies for the bad audio, didn't remember that it was supposed to be just a test video to check my webcam's mic and forgot to use my actual external mic.

tldr; use vscode dev on browser, superwhisper for dictation, roo code(claude code doesn't work well with dictation cause terminal)

Sure man, any product you don't like is just "another wrapper". I guess every website is just a wrapper over postgres or wordpress too. I run my own serverless GPU containers on runpod with a combination of comfy and my own fastapi servers using diffusers, not that it'd even matter if I just used some third party APIs. It originally even started as smth that was hacked together using 4x 4070ti supers in my basement that I then moved to runpod. Indiehacking is mostly marketing, nobody cares if you built some technically beautiful thing.

Also its easy to criticize from the sidelines but, do you have products that you made by yourself that are used by hundreds of thousands of people? I have 5 such sites, 2 of which I named above

Not the OP but this is smth I've vibecoded using cursor: https://bestphoto.ai/ MRR ~$150. It basically started as a clone of my other site: https://aieasypic.com (MRR 2.5k, 5-8k/mo rev) since I was having trouble keeping code context in mind and claude was pretty bad at doing full features with the tech stack I used for that site(Django BE, NextJS FE) making adding new features a pain, so I completely switched to a stack that claude is very good at NextJS fullstack(trpc BE) and now it can basically one-shot a feature request.

Just putting this here because a lot of times AI coding seems to be dismissed as smth that can't do actual work ie generate revenue, while its more like making money as a solo dev is already pretty rare and if you're working in a corp. instead you're not going to just post your company name when asked for examples on what you're using AI for.

Mostly Lora training not full finetunes. Eval for image gen is esp. hard because if you look at AI generated images of someone for too long you can start to miss which one looks more similar, even with your own images.

And yes model training is available on the site, the serverless pipeline itself was easy, making it fast(less than an hour) and have the most similarity while being flexible enough for a general user was the hard part.

Let me know if that answers your question

I mean if you're looking for a todo list, travelling to third world countries and tipping just like $5 for normal service and seeing people's eyes light up is fun.

Location: Slovakia, EU. OK with working in US timezones Technologies: Python, Django, NextJS, AWS, Docker Experience with platforms like AWS, Azure, Hetzner, Runpod In addition: Worked with frontend technologies such as Nextjs, React, Zustand, Mantine, Material UI, Tailwind

Email: mesmerpercy@gmail.com

Resume/CV: DM/Email

Some extra info: 6 years of experience as a fullstack dev with Django + React/NextJS. I've built and run a site that receives ~20 mill pageviews per month(private, can showcase on interview if needed). Currently interested in generative AI, and I have experience creating serverless pipelines for GPU tasks to reduce costs instead of using something like Replicate, huggingface etc. Recent public examples of my work:

https://aieasypic.com/ (800k views per month, 4k+ MRR)- Built my own serverless api for the dreambooth training and AI image inference to reduce costs. Also uses 4 of my own 4090s in my basement

looks like its not even available on EU, I'll call that a win

This specific link I shared is the quant running on a 4090 I rented on runpod, I have no affiliation with the repo itself

Thats what I thought as well, but FP8 is much faster on h100, like 2x-3x. You can check it/s here: https://github.com/aredden/flux-fp8-api

Its why fal, replicate, pretty much all big diffusion api providers use h100

tldr; 4090 is max 3.51 it/s even with all the current optimizations. h100 is 11.5it/s with all optimizations, and even without its 6.1 it/s

prolly make sure your host machine cuda is also 12.4 and if not, update the other cuda versions I have on the pastebin to the one you have. I don't think it works with cuda 11.8 tho, remember trying it once

but yea, can't help you outside of runpod, I haven't even tried this on my home PCs yet. for my usecase of serverless API, it seems to work

Yea its a pain, I'm trying to make an api endpoint for a website I own, and working on a docker image. This is what I have for now that "just" works:

the conda always yes thing makes sure that you can just paste the script and it all works instead of having to press "y" for each install. Also if you don't feel like installing a wheel from random person on the internet, replace that step with "pip install -e ." as the repo suggests. I compiled that one with cuda 12.4 cause that was the part takes the most time and is what most often seems to be breaking.

Also I'm not sure if this will work on Fedora, I tried this on a runpod machine with 4090(apparently it only works on few cards, 3090, 4090, a100 etc) with Cuda 12.4 on host machine and "runpod/pytorch:2.4.0-py3.11-cuda12.4.1-devel-ubuntu22.04" this image as base.

EDIT: using pastebin instead as HN doesn't seem to jive with code blocks: https://pastebin.com/zK1z0UdM

Demo on actual 4090 with flux schnell for next few hours: https://5jkdpo3rnipsem-3000.proxy.runpod.net/

Its basically H100 speeds with 4090, 4.80it/s. 1.1 sec for flux schenll(4 steps) and 5.5 seconds for flux dev(25 steps). Compared to normal speeds(comfyui fp8 with "--fast" optimization") which is 3 seconds for schnell and 11.5 seconds for dev

its that color for people who recently signed up

You can check out civitai's repo if you want smth for NextJS: https://github.com/civitai/civitai

Found it pretty useful for my usecase at least since most of my tech stack aligns with them, along with the component framework they use(Mantine)