@SpartacusNole56 "Just pack your whole life up and buy a house halfway across the continent to a dying shithole town with no high income jobs where you don't know anyone" for only $220k? dude sign me the fuck up!
English
Max 🇺🇸
396 posts
Don’t know what child needs to hear this but if you can’t buy a $220k starter home and a $5-7k used car and maintain both that’s your problem not mine.
Acquire some skills and a spine instead of crying on social fucking media.
English
I really don't get the appeal of codex 👀
How is having to press "yes continue" 100 times to complete a project better than Cursor just completing the project?
English
The Swiss grid is almost completely carbon free, yet it is largely illegal to install air conditioning in a house in Geneva, without a doctor's note and heat pumps have the a/c mode disabled and fines issued if used. A mean spirited degrowth mindset that transcends class and hides behind an environmental narrative even when it has nothing to do with the environment, like here. This is emblematic of a pathological hatred of the future and progress, that is a threat to Europe.
Atticus@redl3tters
One of my business school friends comes from a wealthy Swiss banking family and lives in the ritzy part of Geneva where the average house is in the tens-of-millions range. When I visit, I stay in his guest room and sweat profusely at night because, for all their wealth, these bozos can’t understand the unmatched pleasure that is a nice, cool room to sleep in at night. Bunch of uncultured rubes.
English
@JamesTimmins These tools were built for a time when wrapping /v1/chat/completions was difficult. Tools like Pi prove that you can do so much more with much less.
Pydantic is the only defendable one in today's landscape. LangChain is completely unusable outside of a pet project with no users
English
Is anyone actually using tools like Langgraph/crew ai/pydantic ai?
Feel like I should learn one (vs rolling my own), but Reddit basically calls them all unnecessary.
English
The harness matters more than the model.
Models have gotten really good. Great reasoning, large context windows, better instruction following.
But, what makes *use* of those capabilities is actually the harness. It's what provides tools, memory, skills and context to the model.
ChatGPT is a harness. Claude Cowork is a harness.
Without the harness, the model is just an engine with no car. You don't get anywhere.
English
@AskYoshik @NoemiTitarenco have you ever heard of "Software AI" or "Llama.cpp AI" or "Owning a computer AI"
English
@NoemiTitarenco I'm bullish on cerebras for now. Never heard of Local AI
English
The AI pricing war is about to get very ugly for closed-source AI companies.
DeepSeek V4 Pro: $0.87
Grok 4.3: $2.50
GLM 1.5: $3.08
Qwen 3.7 Max: $7.50
Opus 4.7: $25
GPT-5.5: $30
And people still think these margins survive long term?
Open source models are getting intelligent, cheaper, faster, and enterprises care way more about cost than benchmarks.
This is exactly why I keep saying the AI math feels weird. The tech is real. The current economics around it? not so sure.
Yoshik@AskYoshik
People are acting like questioning AI economics means denying AI itself. Some of the biggest financial analysts and investors in the world are already debating this openly: Michael Burry comparing it to dot-com, Jeremy Grantham warning about bubble dynamics, WSJ questioning whether current AI infra spending will ever justify returns. Even during the internet boom, the technology was real. The bubble was also real. Both can exist together.
English
Things Codex Likes to Say, Translated:
- smoke test (test, which is not clearly designed and organized as a unit, integration, or regression test)
- seam (module interface)
- gates (conditional check in code, or as a defined workflow development milestone condition)
- signal (test result, log value, or runtime input)
- surface (UI or API interface)
- knob (any configurable value)
- load-bearing (causally critical)
- runbook (instructions)
- bring up (initialize and start)
what am I missing?
Chris 🇨🇦@llm_wizard
It's crazy how you can learn when people start using AI assisted programming more these days by how often they say "smoke test".
English
What’s your experience with deepseek like?
Thomas - @sourcetms@sourcetms
There are two sides to the debate
English
@MicroLaunchHQ > Resell $2M worth of tokens for $1.99M
> $1.99M MRR
> $24M ARR
> Raise $500M
> Start a neocloud
> NVIDIA invests $1B
> Sell the company
English
> Join YC
> Get $2M in tokens grant from OpenAI
> Hire zero engineers
> Run 400 agents 24/7
> Half build apps
> Half do GTM
That’s the billionaire solopreneur starter pack.
English
@thatguybg llama.cpp & MLX
GPT-OSS-120B-Q8
Qwen 3.6 27b
Gemma 4 31b
Qwen 3.6 35b a3b
MTP if possible. throw them behind a local router that dynamically loads weights onto mem per request
English
ok new laptop again
maxed it out: M5 Max, 128GB mem, 4TB storage, liquid retina XDR display
what should I load up on this baby to put it to the test?
English
"you're absolutely right! *riffs on your wacky idea*" is psychologically healthy for a 4 year old to hear, just not for a 34 year old
did people who crave sycophantic AI not receive enough affirmation as kids?
if @OpenAI cared about public benefit they'd publish their research
English
- half open laptop <- ridiculous
- notion <- shit
- vercel <- great for frontends
- next.js <- has a lot of support & docs
- ai sdk <- generally goated
- supabase <- dodg shit
- cursor <- fell off, but is now the workhorse
- mcp <- great for enterprise
- shadcn <- goated
- embedding structured data <- hilarious
- “deterministic” <- new age. vs probabilistic
- drizzle <- funny how bad agents are at this
-“my agents” <- AI pilled
- context engineering <- Q3 2025
English
very easy to tell who learned “tech” through vibe code twitter
- half open laptop
- notion
- vercel
- next.js
- ai sdk
- supabase
- cursor
- mcp
- shadcn
- embedding structured data
- “deterministic”
- drizzle
- “my agents”
- context engineering
not saying it’s wrong just interesting the significant group think
English
@naturedotcom Opus for taste
Codex for Big brain autismo power
Composer 2.5 for production implementation
Qwen for turbo-powered research
But 5.5 is the GOAT for day to day. not even close.
English
for like 3 weeks, claude was better codex, everyone pivoted to TUIs. entire companies formed around the premise. Then codex became better, and now clearly the UI is better than being terminal based.
People just chasing the ball week to week man. have some long term conviction!
English
@jun_song @outsource_ It's like buying a Rolex. The guy at the apple store congratulated me when i picked mine up.
English
@outsource_ Mac Studio goes up more than average people’s portfolio 😂
English
UPDATE I relisted the Mac Studio M1 Max 64GB Ram for $2200
Someone just offered me $2,000
That’s $600 profit in 11 days….
Should I do it??
I don’t necessarily need $600 but flipping stuff is always 🔥
Vote in the comments 👇🏻 FLIP OR KEEP
Eric ⚡️ Building...@outsource_
UPDATE WE GOT EM BOYS We picked up a Mac Studio 64GB RAM!!!! Now I’ll pair this with my M1 Pro MacBook Pro 32GB Combined Ram now = 96GB + my 4090 etc.
English