Sebastian Knight

@codyschneider generator

I just had Claude Code build me a Facebook ad generator that can make 100+ on-brand ad variations in minutes for $0. And I made a full Notion document guide for you. It includes: 1. How to use Claude to find the pain points and desired outcomes of your ICP 2. How to use these pain points and outcomes to write ad copy variations 3. How to build a Facebook ad template entirely with code (just like the ones you see) 4. How focus Claude Code’s design so the ad feels “on-brand” 5. How to export the Facebook ads as PNGs in a zip file 6. How to bulk upload them to a Facebook ad set 7. How to use an AI data analyst to track the success of these ads Everything above is just API calls and Claude Code doing the work for you. You just come up with the ideas and polish the outputs. Like and comment "generator" and I'll send the Notion document to you

@TommiPedruzzi 90

The best investment strategy in 2026? AI publishing. • Find people’s problems • Package the solution inside a 90-page PDF • Sell it on Amazon KDP I made $109,000 in 90 days doing this. But obviously, there’s more to this business. I’m finally sharing my $50k/month eBook publishing strategy. It’s usually reserved for my paid students, but you can access it free for the next 24 hours. Like this post Comment “90” I’ll send you the link to my strategy for free. Make sure you’re following so I can DM you.

@StrtlyX Can totally relate to this 😂

entrepreneurship is just waking up every day and choosing delusion you go from “this will be a $10M business” to “why did 47 cold emails get 0 replies” in 6 hours the winners aren’t less delusional they just keep sending the emails after the crash

@cursor_ai Game on 🚀

Claude Opus 4.5 is now available in Cursor! It's 3x cheaper than Opus 4.1 with better performance. Try it out at Sonnet pricing until December 5th.

@StrtlyX B

Me A :-)

A B C D???

@karpathy @rasbt Have you thought about giving Samsung new trm model a try? Looks really promising to me.

Basically Llama-like, a bit simpler, some influences from modded-nanoGPT. Tried to find a solid baseline for this scale: - dense transformer - rotary embeddings (and no positional embeddings) - QK norm - untied weights for embedding and unembedding - norm after token embedding - relu^2 activation in MLP - no learnable params in rmsnorm - no biases in linear layers - Multi-Query Attention (MQA) - logit softcap Optimizer is Muon+AdamW, heavily influenced from modded-nanoGPT. I have a TODO to try to tune Adam LRs well (e.g. per module) to remove Muon, I haven't tried hard enough yet.

Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single, dependency-minimal codebase. You boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web UI. It weighs ~8,000 lines of imo quite clean code to: - Train the tokenizer using a new Rust implementation - Pretrain a Transformer LLM on FineWeb, evaluate CORE score across a number of metrics - Midtrain on user-assistant conversations from SmolTalk, multiple choice questions, tool use. - SFT, evaluate the chat model on world knowledge multiple choice (ARC-E/C, MMLU), math (GSM8K), code (HumanEval) - RL the model optionally on GSM8K with "GRPO" - Efficient inference the model in an Engine with KV cache, simple prefill/decode, tool use (Python interpreter in a lightweight sandbox), talk to it over CLI or ChatGPT-like WebUI. - Write a single markdown report card, summarizing and gamifying the whole thing. Even for as low as ~$100 in cost (~4 hours on an 8XH100 node), you can train a little ChatGPT clone that you can kind of talk to, and which can write stories/poems, answer simple questions. About ~12 hours surpasses GPT-2 CORE metric. As you further scale up towards ~$1000 (~41.6 hours of training), it quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests. E.g. a depth 30 model trained for 24 hours (this is about equal to FLOPs of GPT-3 Small 125M and 1/1000th of GPT-3) gets into 40s on MMLU and 70s on ARC-Easy, 20s on GSM8K, etc. My goal is to get the full "strong baseline" stack into one cohesive, minimal, readable, hackable, maximally forkable repo. nanochat will be the capstone project of LLM101n (which is still being developed). I think it also has potential to grow into a research harness, or a benchmark, similar to nanoGPT before it. It is by no means finished, tuned or optimized (actually I think there's likely quite a bit of low-hanging fruit), but I think it's at a place where the overall skeleton is ok enough that it can go up on GitHub where all the parts of it can be improved. Link to repo and a detailed walkthrough of the nanochat speedrun is in the reply.

@karpathy @garricn I have had the same experience, was really disappointed when it was released. But this week I have built a new SaaS product that can compete with similar products that existed for years with codex.

@garricn not yet but planning to! I rotate through a lot of what exists in few day intervals. Actually ~3 weeks ago I found codex worse and less polished than CC right now for the more basic edits and overall experience. It's just 5 Pro specifically feels SOTA by a good margin atm.

I think congrats again to OpenAI for cooking with GPT-5 Pro. This is the third time I've struggled on something complex/gnarly for an hour on and off with CC, then 5 Pro goes off for 10 minutes and comes back with code that works out of the box. I had CC read the 5 Pro version and it wrote up 2 paragraphs admiring it (very wholesome). If you're not giving it your hardest problems you're probably missing out.

@thesamparr How do you come up with $5m+? Isn't Ali the asset?

Ali Abdaal, the YouTuber running a $5m+ creator business, says $100M wouldn’t change how he lives. Why? Because six figure freedom beats a nine figure exit. He came on Moneywise this week. Here are some numbers: -Net worth between $1M–$10M -~20% cash, 30% real estate, 10% crypto, rest in S&P -Rent is $11K/month in Hong Kong (buying would cost $7M) -$70Kmrr from courses, sponsorships, and SaaS apps -Biggest luxuries: flying business class, never taking out the trash Ali got his start tutoring kids at Kumon in his teens. He saved for five years to buy himself a MacBook. Lost it all in a scam. That sting pushed him to start his own tutoring business, which he later sold for ~$200K. From there, he scaled YouTube, courses, and now SaaS into millions. My favorite quote from him: “Freedom comes from leaving money on the table.” His goal is $2M profit a year. So he can work when he wants, on what he wants. Full episode of Moneywise is live.

The Turing test is outdated. The real AI test? Build a working theme toggle in Next.js + Tailwind… No model survives. 🪦

@zeeg @kentcdodds Impressive! I spend quite a lot of time improving my cursor rules files to fit my project in the beginning. With that in place i can focus the prompts on the actual problem I want to solve.

I basically do everything the same but I try to provide a reasonably clear description of what I’m trying to achieve as I know otherwise it won’t be remotely successful. It’s definitely more effective in smaller codebases but sometimes the results surprise me. I generated an almost fully production grade patch in sentry with my normal flow (without as intense of planning).

I literally have not written code by hand in weeks btw Doesn’t mean I’m hands off on the system design or anything else with the code, but I’m def not writing code.

What is your best hacks for getting back links. Today i built 40 different calculators in different niches. Will it pay off?

@cursor_ai If I haven't used up my monthly quota and I run in auto mode, does that affect the monthly quota in any way? It would be good to know if you most of the time should start in auto mode.

We recently updated our pricing, but missed the mark. We're refunding affected customers and clarifying how our pricing works. cursor.com/blog/june-2025…

@marclou @tibo_maker @beehiiv Why not just try another provider for a week and see the stats?

@tibo_maker @beehiiv Yeah, I wished I had migration initially you can't swap domain reputation (40K readers, 55% open rate 😭)

How to 6x your MRR in 1 minute: 1. Triple the pricing 2. Reduce usage by 50% 3. Add no extra value I want to leave @beehiiv now.

@marclou In Europe i think we walk a lot more, my average is at 18k each day without thinking about it, and I have like 4k back and forth to work. My 15 year old daughter and her friend did a walk challenge a few weeks ago, 125k under 24h 💪

Walking is the easiest way to burn body fat. 10k steps is 500kcal, like an intense HIIT session. Meanwhile, you can listen to a podcast, take a call, or write a tweet like this one.

@gregisenberg Any experience with chargeflow.io

can we all agree Stripe Disputes feel like extortion for most founders? a customer taps “chargeback,” the product’s gone, stripe skims a $15-$22 fee, and you’re dragged into evidence-upload purgatory judged by a bank that doesn’t know your business. you win the dispute and you only recover what was yours; lose and you’re out cash, time, and sanity no appeal, no credits, just an invoice for getting ripped off. yes, we need guardrails for real fraud, but piping every edge case through a pay-to-defend toll booth punishes builders more than bad actors. there has to be a smarter fix. i gotta believe it. am i wrong here? i feel like im taking crazy pills.

@tifethedev Print a text on a paper and put it on the table so everyone can see it, instead of writing it to every customer.

can someone explain to me like i’m 5, what caching is?

@adcock_brett Impressive, but for this kind of tasks a cheaper specialised robot should do the job. Would love to see figure more in a more generic human environment.

Is this working Dan?

@nevmed @thesamparr Love the bike and @thesamparr , but the helmet 🤣

@thesamparr Sick. P.S. There's a story where this motorcycle plays a part in my marriage, it's how Sam first met my wife bc it got delivered to my garage...he was first friend to meet her, then ended up officiating our wedding!

New Hampton office is coming along. We got the important stuff here...