Aaron Burton

@_devJNS They aren't paying for code production, they re paying for accountability and general common sense that Ai does not have right now. Ai is not good at solving/preventing future problems. You'd know this if you ever worked with Ai directly on any serious project.

why can't they just use claude?

conversationally Claude is amazing, but this is also its downside. I don't always want a human response especially if it has errors / unverified. But they are doing a damned good job of balancing this, really hard to critique Anthropic. The issue is when you are chatting about topics that aren't covered at scale. So it has to infer, guess, and "vibe" with what you think the answer is.

After Claude is Claude. Is there any AI tool that is better than Claude?

GPT is just behind. Something isn't right about its outputs. Not as natural as Claude. Gemini is nice because its a google product, doesn't chug, 2 clicks to access, no creating new chats. Tends to give based answers if you question is based. Gives bad super general answer for when your searching technical stuff. But will correct itself once it knows what you need. Microsft is just hard to access. Not native, dont use bing. But its pretty good, the memory is I think better than the top models. Deepseek will probably become top tier for cheap models as token bloat increases. Perplexity is an honorable mention. Could even replace most daily drivers since it integrates models. Of course all these companies are coming out with their own. ie Sonar, Composer, etc

Dear developers, do you all agree??

@TechByTaraa cursor is just an IDE but id choose it in a heart beat. Full project coverage is how you ship consistently. coding in the browser is a no go.

If you have $20, which one do you choose??

@TechByTaraa what do you men disappear? lol.

Be honest If coding disappears tomorrow what’s your plan B?

Had to build a load balancer yesterday, wasn't actually that scary. You'll know when you need one. Started with a round robin approach, didn't make sense, upgraded to a "least loaded" model. This is likely "good enough" for now. Don't need complex serverless technology. Just a few lines of code and some common sense.

I do not like the number 10 lol. I get it now.

@TTrimoreau Cloudflare hands down. workers are nutty bro. But its simple tech. Can see the same results on a VPS if you know how to set it up.

Best hosting platform for startups? - Render - AWS - Railway - Vercel - Microsoft Azure

@TTrimoreau No excuse not to live while your ai buddy is making money in the cloud.

> 2022 - Student > 2023 - Developer > 2024 - Prompt Engineer > 2025 - Vibe coder > 2026 - AI agent > 2027 - Farmer

@elonmusk Prob the best diffusion model tech out right now.

@brianshinsh rounded edge around 30 with a 22.6 degree angle. sorry im a design nut haha

new app is finally live! was soo nervous that it wouldn’t be ready by demo day but apple thankfully approved it on the last day 🥲 goal is to reach $10k/month in 100 days and i’ll be sharing my lessons on the way! (super happy that the first photos ever taken on the app are my friends from HRG ❤️)

@brianshinsh Lightwork! you have good design taste

Exactly

The reads are whats burning your tokens. Very ridiculous when you actually stop and look at the numbers. Cache Read 272,954 Cache Write: 1,212 Input: 5 Output: 625 Total: 274,796 Cache Read: 561,679 Cache Write: 3,656 Input: 8 Output: 2,142 Total: 567,485

@yaroslavvb @karpathy damn who wants to have a cookout!

Greg and @karpathy attemping to use grill at OpenAI 2017 offsite

ai doesn't touch files directly, have it create shadow copies you can observe. Not realistic for huge files but smaller text files are the best use case. System settings are even more dangerous. One small mistake can be irreversible.

@VadimStrizheus Too early for our own good haha. My recommendation is for people to learn all they can but dont try to build some big thing. Solve local problems that align with your own industry. .

We are so early into AI, and 99.99% don’t even know it.

@elonmusk Yea no bueno haha. I built my own from scratch, its not that impressive. And its dangerous if the user/company doesn't know what the are doing. You also don't need ai to run sub processes.

People giving OpenClaw root access to their entire life

@MaximeMB_ No surprise haha. Bell curve distribution for top 1% of performers holding CEO roles.

Just realized Anthropic's CEO was an ex-OpenAI employee 🤯

Noticing a systems design issue with models calling tools unprompted. Currently Google Gemini and Chat GPT load image generation models without warning. Whats likely happening is the Ai is allowed to call tools based on pattern matching alone, this isn't correct. They Should always require manual approval. Anthropic is doing it correctly.