rgbrenner

Even if we all switch to Chinese models, the west isn't going to be running the model on Chinese servers... and the majority of costs are from inference.

> cheaper yet equally good talent

China has tech talent, but this isn't a 3rd world developing nation. Chinese AI researchers are getting paid $10M+ USD/year salaries.

Also they're equally good, but somehow consistently behind?

and in this scenario, i’m supposed to root for musk who tried to use the court to harm a competitor who’s winning in the marketplace against xAI?

no thanks. if you can’t compete in the marketplace, the court isn’t your backup plan. there’s nothing. positive about the weaponization of the courts.

Doesn't change my point: the amount of code the agent can operate on is very large, if not unlimited, as long as you put even a little bit of thought into structuring things so it can be divided along a boundary.

If you let the codebase degrade into spaghetti, then the LLM is going to have the same problem any engineer would have with that. The rules for good code didn't disappear.

you can also execute larger tasks than this using subagents to divide the work so each segment doesn’t exceed the usable context window. i regular execute tasks that require hundreds of subagents, for example.

in practice the context window is effectively unlimited or at least exceptionally high — 100m+ tokens. it just requires you to structure the work so it can be done effectively — not so dissimilar to what you would do for a person

Just not accurate. WordPress doesn't prevent this.. It's up to hosting providers to work on their infra so it can run in a serverless fashion.

For example: https://www.agiler.io

That's serverless wordpress that scales to zero.. no changes to WordPress, plugins or anything else.. just platform infra.

So the story isn't really about firefox.. it's about Chrome's marketshare being high enough that some companies are happy to ignore every other browser.

On the other hand, if OpenClaw were structured as a SaaS, this entire project would have burned to the ground the first day it was launched.

So by releasing it as something you needed to run on your own hardware, the security requirement was reduced from essential, to a feature that some users would be happy to live without. If you were developing a competitor, security could be one feature you compete on--and it would increase the number of people willing to run your software and reduce the friction of setting up sandboxes/VMs to run it.

doesn’t that presume no value is being delivered by current models?

I can understand applying this logic to building a startup that solves today’s ai shortcomings… but value delivered today is still valuable even if it becomes more effective tomorrow.

The scary thing is that there is.. you should look up "sovereign immunity". The government has complete immunity, except where and how the law permits it to be held accountable. And while we have a constitution, defending those rights through the courts requires legislation to permit it. For the most part, federal law permits lawsuits against states that violate the constitution, but have permitted far less accountability for federal actions that violate the constitution.

For example, Section 1983 of the Civil Rights Act only permits individuals to sue state and local governments for rights violations. It can't be used to sue the federal government.

There's many court cases, dating back decades, tossing out cases against the federal government for rights violations. Look how SCOTUS has limited the precedent set by Bivens over the years, basically neutering it entirely.

the entire thing — from the phrasing of errors as “hallucinations”, to the demand for safety regulations, to assigning intention to llm outputs — is all a giant show to drive the hype cycle. and the media is an integral part of that, working together with openai et al.

if you go put your car in drive and let it roll down the street.. the car has 0% of the blame for what happened.

this is a full grown educated adult using a tool, and then attempting to deflect blame for the damage caused by blaming the tool.

Edit: I think I know where our miscommunication is happening...

The "think"/"ultrathink" series of magic words are a claudecode specific feature used to control the max thinking tokens in the request. For example, in claude code, saying "ultrathink" sets the max thinking tokens to 32k.

On other clients these keywords do nothing. In Roo, max thinking tokens is a setting. You can just set it to 32k, and then that's the same as saying "ultrathink" in every prompt in claudecode. But in Roo, I can also setup different settings profiles to use for each mode (with different max thinking token settings), configure the mode prompt, system prompt, etc. No magic keywords needed.. and you have full control over the request.

Claude Code doesn't expose that level of control.

also no idea why he thinks roo is handicapped when claude code nerfs the thinking output and requires typing “think”/think hard/think harder/ultrathink just to expand the max thinking tokens.. which on ultrathink only sets it at 32k… when the max in roo is 51200 and it’s just a setting.

You should just set aside some time to try out different tools and see if you agree there's an improvement.

For trying models, OpenRouter is a big time saver.

Traditional code editing -> autocomplete -> file editing -> agent mode

This is basically a gradient of AI output sizes. Initially, with the ability to generate small snippets (autocomplete), and moving up to larger and larger edits across the codebase.

Cursor represents the initial step of AI-assisted traditional coding... but agent mode is reliable now, and can be directed fairly consistently to produce decent output, even in monorepos (IME). Once the output is produced by the agent, Ive found I prefer minimal to no AI for refining it and cleaning it up.

The development techniques are different. In agent mode, there's far more focus on structuring the project, context, and prompts.. which doesn't happen as much in the ai-autocomplete development flow. Once this process shift happened in my workflow, the autocomplete became virtually unused.

So I think this shift toward larger outputs favors agent-focused tools like CC, Aider, Cline, and RooCode (my personal favorite).. over more traditional interfaces with ai-assistance.

It's also missing quite a few features still (like checkpoints, stopping the agent mid-task to redirect, etc). But the core feature set is mostly there.

also english wikipedia is actually for english speakers.. so it includes countries that aren’t america. there’s a reason they didn’t name it american wikipedia.