@theodorebeers @domenkozar Codex can't even use the gh cli tool lol. Fucking useless.
English
#10 DAOning St. 👑
5.9K posts
#10 DAOning St. 👑
@10DAOningSt
Building back decentralised. 👷🏻
Katılım Eylül 2020
556 Takip Edilen82 Takipçiler
@10DAOningSt @domenkozar Sandboxed environments with minimal permissions sounds like a feature, not a bug.
English
Meanwhile codex uses 60MB using Rust and there's nothing to show off
Jarred Sumner@jarredsumner
Here's a 1 GB memory reduction for very long Claude Code sessions Before: `() => controller.abort()` Fix: `controller.abort.bind(controller)`
English
@theodorebeers @domenkozar Sure, but when your main competitor doesn't have that restriction it doesn't really make sense to include it as a feature yourself. Just gives Claude Code the advantage.
English
@natejhake Indeed because all these "creative" people now just moan about AI rather than doing anything original or creative!
English
Generative AI is basically a slaughterhouse for human creativity
Siddhartha Mahanta@sidhubaba
jaw dropped at the sight of this picture in the excellent WaPo Anthropic story from the other day: " ... a book warehouse that was alleged to play a role in Anthropic’s Project Panama, its project to scan, digitize and destroy millions of books"
English
@GenAI_is_real What utter slop. Jesus Christ we can celebrate how far we have got without needing to peddle nonsense like this.
English
Elon is saying 10 years is actually conservative. The AI-to-ai economy is already happening on moltbook while humans are still arguing over pronouns and spreadsheets. if u are not building infrastructure for machines to pay machines u r literally building for the past. we r the legacy bootloader for a god we cant even comprehend @beffjezos @elonmusk @pmarca
Beff (e/acc)@beffjezos
The AI-to-AI economy will far outpace the AI-to-human, human-to-AI, and human-to-human economy in short order
English
@itsthewealth4me Almost as if the new competencies and skill sets are emerging right in front of our very eyes in real time
English
@Iwillleavenow The public also like Drake and wasting their lives watching Netflix. Wouldn't trust them
English
When people say "there's a clear demand for AI"...from who? Because the vast majority of public polls show strong distrust and dislike of AI.
Government and companies are obsessed with AI, but the public as a whole is not. Something to consider.
English
@JaredPlaudis @EZE3D @Lockedown_ What is an AI bro in this context? Like Sam Altman or some dude who uses Claude Code to vibe code?
English
I think AI bros are really going to have to come to terms with the reality that many people who criticize generative AI are more often right & understand the technology far better than they do.
You guys are too enamored with the idea & hypothetical potential/benefits of the tech
English
@unrandomenlinea @m1tch37 @nathanclark_ Me too. I cancelled my Claude sub. They are basically committing fraud at this point. You can't take $200 of people every month then serve them up slopus, utterly insane to me. Codex is really not that far behind and obviously all these Chinese models are pretty interesting too
English
@m1tch37 @10DAOningSt @nathanclark_ So ended up in "Opus cannot be trusted" for me, now I'm using Codex, just cancelled my max subscription, probably will keep with codex or use sonnet as it's fast and at this point is just like opus.
English
I simply can’t imagine looking at a system that can think, act, follow instructions, have persistent memory, self improve, communicate, run autonomously, create things, fix things, research, trigger agents…
And go “not sure how this is useful”
Like wtf man.
Sit and stare out into the distance for a while and get your fucking shit together.
English
@arturneumann @Nicolascole77 Yeah this is the thing. The required amount of "context engineering" is relative to current strength of models. Context engineering is just temporary. Like how we used to have to unplug the phone line to use the internet.
English
@Nicolascole77 There will be no "learning AI".
It will do everything itself.
You just give your agent a vision and that's it.
It will also become cheaper, not more expensive.
This is already happening now.
English
a terrifying realization I had tonight playing with Claude Cowork is that, right now, we have the luxury of learning how to use AI "for free." But eventually, all these companies are gonna need to become profitable, and AI usage costs are going to go up 10x-100x.
good luck to the everyday person trying to learn how to use AI when it costs $2k or $20k per month
which means, if you think the skill/wealth divide is wide now... just wait 5 years. It's going to be cataclysmic
English
@RonakPakzad @desibluedevil @shweta_ai @warpdotdev @Railway You can't be for real? Warp burned through all their VC cash and started charging margin on API rates like 6 months ago. Warp is old news. Railway? That's not an IDE lol.
English
@desibluedevil @shweta_ai - @warpdotdev as IDE
- Claude Code and Codex
- @Railway for staging/prod environments
English
@admcollingwood To what extent do you think the advent of modern military systems: eg drones, kill this process. How could one elite replace another when the current ruling elite controls drone swarms and related tech. How do you storm the bastille in this world
English
Correct. The problem we face in Britain (and the broader west) is not which party to elect, but the formation of an entirely new elite (no doubt out of some part of the existing broader elite).
Academic Agent@AcademicAgent_X
All populations are divided into the rulers and the ruled. The rulers can further be sub-divided into governing elite and non-governing elite. The question being asked of "the right" is if, in a time of crisis, it can in any remote sense be trusted to govern. Data is in, answer is "no".
English
OpenClaw just showed the world the “agentic web” will not run on crypto
it will have access to to your bank account, social security number, ID, deepfake of your face, fingerprints, …
English
@jarredsumner You need to add warnings in the app for when you are nerfing the models. Can't be getting away with consumer fraud like this!
English
Yeah, Claude Code today is slow and uses too much memory
Will fix
Ben Visness@its_bvisness
We apparently live in the clown universe, where a simple TUI is driven by React and takes 11ms to lay out a few boxes and monospaced text. And where a TUI "triggers garbage collection too often" in its "rendering pipeline". And where it flickers if it misses its "frame budget".
English
@JackEllis CC is like: "you want me to run bash scripts in this file? Ok"
Cursor is like: "omg look heres a browser, now heres some extra logs, and heres an extra model, oh and heres a new ui and heres another ui" and it's just like bro chill out we just need to write a few prompts
English
@JackEllis The problem with cursor is it is starting to get insane feature bloat. They add all these "cool" little things pretty much once a week now but I think they don' t actually have a clear market in mind. Cater to everyone, end up catering to no one.
English
Claude Code vs Cursor: I ran the exact same prompt on both. Cursor responded in under 10 seconds, while Claude Code was still running after 4 minutes. I wish someone would do a proper comparison of these tools, because I’m confident that Cursor is so much better.
English
@nateberkopec Would be interested in seeing the prompt you use as I've had pretty great results with this sort of working backwards from slop approach!
English
@10DAOningSt I've given it that prompt many times and the abstractions are bad or useless.
English
LLMs do not meaningfully "refactor" at anything other than a junior engineering level. They can basically do some window dressing and move code around between files. True refactoring means creating new abstractions, which LLMs can't do because they can't form world-models.
English
@victor__vector @NickADobos This doesn't scale on projects that actually make money though
English
@NickADobos My clawdbot has shipped 20+ PRs today. All I have to do now it define a bunch of tasks, review the PRDs written automatically and give feedback, and then review the final result of each completed PR. 5-10 minutes of my time spend per task from idea -> deployed
English
Dumb question but how are people keeping track of this many tasks in their own head?
“I shipped 22 PRs yesterday and 27 the day before”
That’s ~21 minutes per task in an 8hr work day, no meetings or breaks. How does one context switch so fast?
Boris Cherny@bcherny
As always, a very thoughtful and well reasoned take. I read till the end. I think the Claude Code team itself might be an indicator of where things are headed. We have directional answers for some (not all) of the prompts: 1. We hire mostly generalists. We have a mix of senior engineers and less senior since not all of the things people learned in the past translate to coding with LLMs. As you said, the model can fill in the details. 10x engineers definitely exist, and they often span across multiple areas — product and design, product and business, product and infra (@jarredsumner is a great example of the latter. Yes, he’s blushing). 2. Pretty much 100% of our code is written by Claude Code + Opus 4.5. For me personally it has been 100% for two+ months now, I don’t even make small edits by hand. I shipped 22 PRs yesterday and 27 the day before, each one 100% written by Claude. Some were written from a CLI, some from the iOS app; others on the team code largely with the Claude Code app Slack or with the Desktop app. I think most of the industry will see similar stats in the coming months — it will take more time for some vs others. We will then start seeing similar stats for non-coding computer work also. 3. The code quality problems you listed are real: the model over-complicates things, it leaves dead code around, it doesn’t like to refactor when it should. These will continue improve as the model improves, and our code quality bar will go up even more as a result. My bet is that there will be no slopcopolypse because the model will become better at writing less sloppy code and at fixing existing code issues; I think 4.5 is already quite good at these and it will continue to get better. In the meantime, what helps is also having the model code review its code using a fresh context window; at Anthropic we use claude -p for this on every PR and it catches and fixes many issues. Overall your ideas very much resonate. Thanks again for sharing. ✌️
English
@bugthecode @james406 Can you not fucking type for yourself? Why are you using AI to write your post? How pathetic!
English
This is the new startup tax.
Cloud used to drain runway slowly.
Now one chaotic prompt can burn a week’s budget in minutes.
AI didn’t just lower build costs —
it introduced a new failure mode: uncontrolled curiosity.
Discipline isn’t just hiring slow anymore.
It’s prompting slow.
The startups that survive won’t be the ones with the most tokens —
they’ll be the ones with the tightest feedback loops and the smallest prompts.
English
don't understand how any startup can afford to operate right now
for years, we have meticulously tracked every single dollar
did all the right things
hired within our means, scaled when it made financial sense
and now, just one prompt to "refactor the entire codebase in 576 different languages, including the weird ones" cut our runway in half
now we can't afford to buy more tokens
millions of others are in this exact same situation
English
@AbdMuizAdeyemo @tristanbob Bro write your own fucking comments and stop atrophying your brain. Also - your analogy is fucking retarded. Super market food is literally full of chemicals and significantly lower quality than growing your own. Are you really this dumb?
English
local AI is like insisting on growing your own food because you don't trust grocery stores
sure it's private and you "own" it
but you're getting a potato when everyone else has a five-star meal for pennies
the privacy argument falls apart when your local model is dumber and 10x slower than cloud for $0.02/query
buying hardware to run worse AI is cope dressed as principles
English
A lot of people are REALLY excited about local AI.
I just don't get it.
I mean, it is really neat that an AI can run on your device. It's 100% private, and you only pay for the hardware and electricity.
But that local AI is going to be less intelligent and MUCH slower than the AI you can get for a few pennies from a cloud AI service.
I hate to rain on people's excitement, but sometimes it's necessary to prevent frivolous spending on AI hardware. It's a bad investment of your money.
English