There are a couple of decent approaches to having a planning/reviewer model set (eg. claude, codex, gemini) and an execution model (eg. glm 4.6, flash models, etc) workflow that I've tried. All three of these will let you live in a single coding cli but swap in different models for different tasks easily.
- claude code router - basically allows you to swap in other models using the real claude code cli and set up some triggers for when to use which one (eg. plan mode use real claude, non plan or with keywords use glm)
- opencode - this is what im mostly using now. similar to ccr but i find it a lot more reliable against alt models. thinking tasks go to claude, gemini, codex and lesser execution tasks go to glm 4.6 (on ceberas).
- sub-agent mcp - Another cool way is to use an mcp (or a skill or custom /command) that runs another agent cli for certain tasks. The mcp approach is neat because then your thinker agent like claude can decide when to call the execution agents, when to call in another smart model for a review of it's own thinking, etc instead of it being explicit choice from you. So you end up with the mcp + an AGENTS.md that instructs it to aggressively use the sub-agent mcp when it's a basic execution task, review, ...
I also find that with this setup just being able to tap in an alt model when one is stuck, or get review from an alt model can help keep things unstuck and moving.
RooCode and KiloCode also have an Orchestrator mode that can create sub-tasks and you can specify which model to use for what - and since they report their results back after finishing a task (implement X, fix Y), the context of the more expensive model doesn’t get as polluted. Probably one of the most user friendly ways to do that.
A simpler approach without subtasks would be to just use the smart model for Ask/Plan/whatever mode and the dumb but cheap one for the Code one, so the smart model can review the results as well and suggest improvements or fixes.
That's a nice idea, so nice in fact that it already existed as 18F until they closed it under the guise of efficiency earlier this year and are now starting over.
And USDS, which was specifically 2 year terms just like this, for actual experienced engineers at the top of GS15 pay. They destroyed USDS to pretend to reinvent it but with new worse humans.
This is just their attempt to destroy USDS, started by Obama, and rebrand with a Trump-created effort. Just like everything. He wants credit for everything. I am shocked he isn’t calling this Trump Digital Government Services or something.
Its not hard to distugush individual pictures that contain trackable attributes like a license plate number from building a large scale database of them for sale. Or making such a database not legal to sell access to without removing that information, etc. It doesn't need to center on the contents of a single photo.
It's pretty sad watching all the comments here calling this suggestion mccarthyism, witch hunt, etc. Police departments should be doing this and holding their employees to a high standard that's not a witch hunt that's common sense.
If you are a member of a violent gang or are too racist to police justly you are too dangerous to employ as a cop. Not looking for and removing these people is intentionally turning a blind eye to the extreme danger they present to people in their communities.
Because just calling for it does nothing. It’s clear that a top-down approach would be much more effective at solving systemic racism in law enforcement than a bottom-up one. The racists get witch-hunted and fired? Good. People being scared to be overtly racist is a good thing.
I do agree. It can work and when you've put the time into designing it to work then it can feel fairly magical, but there's a point at which you feel like you're almost pre-creating all the queries for users.
this here. I've spent the last couple of years working on a similar data discovery style product and after a lot of playing around with concepts I think semi-natural language descriptions typed and also generated based on your manual data selection can be really useful.
If I'm speaking i have to finish the whole thought and deal with excluding all my "uuhms" and half thoughts. If I'm typing i can intellisense prompt for relevant things. Correlate Sales with _[Discounts, ...]. I think terse natural langage descriptions of data views are really useful aside from voice.
Incidentally nothing like trying to play around with this stuff to make you super self conscious about uuh how you speak.
My experience with this is it's more of a gimmick. It's cool when it works but most of the time I've found it simpler and more accurate to just select the data you want.
Your hypothetical there assumes that AMP actually works well. My experience is it's got painful usability issues at least on android chrome. The worst one I see all the time is scrolling down to actually read the AMP page frequently results in the page closing, returning you to results.
Fast is good, but at least it ought to be a good user experience. I used to use google news a lot, but i totally abandoned it after constant frustrating experiences with AMP. Plus most of the amp-ified pages don't seem significantly faster than the original page. I'm not seeing the utility for users, just for google.
AMP is also literally breaking internet. People share google.com/amp links,sometimes there'd be no preview and hard telling what the actual website being linked is.
On safari I get nothing except AMP links in google and cannot get myself routed to the actual site even with extensive searching. This disables sharing the link for me as well.
Have you considered switching search engines? I've been using https://duckduckgo.com for the last two years and this is the first time I'm hearing of Google AMP.
I really really really want to use DuckDuckGo.com but the search results are probably 10% as helpful as Google. As a developer it's nearly impossible to find a really specific answer using ddg whereas google will have 3 super-relevant links.
Google can (or refuse to) redirect you anywhere. They also get to know about it, in case the analytics on target page was not enough.
That said, I don’t actually agree that AMP’s link rewriting is necessarily the thing that breaks internet, because the original link is still contained in the modified URL, and can be recovered completely offline.
URL shorteneres are much more harmful in that aspect.
The real danger of AMP is giving Google that much (more) power. They don’t hide links today, but what stops them from turning evil(er) tomorrow.
Preferring AMP pages is a clear monopoly practice, and should be the straw that breaks the camel’s back (if you didn’t care about the censorship…). I’m really trying to have faith in free market sorting this out, but I don’t have much faith in the Facebook drones that brought us all this shit in the first place. Are there enough free people left?
I'm using an LG V20 device, with Chrome mobile, and I see a significant improvement with AMP'd pages over normal sites. So-much-so that I dread going to non-amped pages as they are slow and the ads are horrifically annoying.
I guess this is why anecdotal evidence is not as heavily weighed as a thoroughly tested and evidenced study.
If you have specific issues you can spell out, consider filing a bug on it[0]? The AMP team is tracking most of their work via Github as far as I'm aware.
As an iOS user, one bug that bothered me for a long time with AMP was scroll inertia, which seems to be an issue with Webkit rather than AMP[1]. If you poke around the tracker, you can see the team does what it can to fix issues across all platforms/browsers.
Funny that AMP just happened to use some sort of page structure that triggered an iOS bug that >99.9% of the rest of the web does not trigger...
What are the chances that Google would listen to big reports, supposing that anybody could find those links? Google is notorious for being inaccessible unless you happen to know an employee personally or have enough juice on twitter. They do not have a reputation as an open or accessible company.
> The worst one I see all the time is scrolling down to actually read the AMP page frequently results in the page closing, returning you to results.
Fast is good, but at least it ought to be a good user experience. I used to use google news a lot, but i totally abandoned it
I use "request desktop site" on google news to get that working properly on my android.
Although AMP is not a performance beast by any measure, your experience is coloured because you're using Chrome on Android. Google news works pretty well on Samsung browser. It still annoys me that I can't see the link to the actual website but that's it.
So you're saying that Google's official browser is slow on their official web technology on their official mobile platform? And Samsung, who is well known for their software quality issues, is a better choice?
Well, he is speaking from experience. While this isn't the place for speaking from experience but more about enabling a better experience for all users he cannot deny what he has observed. However the user above us had the correct answer with filing a bug / legitimate issue with devs. That is the only way to increase usability for all users.
Samsung's browser (and several other competitors) are faster on Android because they contain closed source rendering that's customised for their chipsets. Google can't integrate the closed source code into Chrome because they'd effectively have to fork it from Chromium to do so.
Doesn't Chrome already contain close source code? I thought it was chromium plus Google proprietary code. For example, some features are available only on Chrome (DRM etc. "features"). If they wanted, they could optimise for a selected few top selling devices.
I thought this was common knowledge. Chrome is horrible on Android. If you look at benchmarks, the only browser/phone combination that is close to iPhone Safari performance is the latest galaxy flagship running Samsung Internet.
The cost isn't the only issue though. You still have to somehow get to the DMV to get it and oops many Texas DMVs were closed in poorer Democrat leaning areas so it's now a time consuming and expensive journey to go get it. Might be an hour or two away but you can't miss your minimum wage job during business hours or no rent money so what are you going to do? Or say you're old and impoverished and don't have your birth certificate now its a huge problem to get that id. Might be you have to travel to another state and go through a DMV like process to get a copy first and again you don't have the means or time.
Texas' strict voter ID requirement was struck down in the courts last summer precisely because of these reasons. It was found that the law, despite free IDs significantly disadvantages black and latino voter's in the state who are much more likely to have issues like above. Didn't stop polling place workers in some areas though some of whom still turned people away based on the non-existent requirement.
- claude code router - basically allows you to swap in other models using the real claude code cli and set up some triggers for when to use which one (eg. plan mode use real claude, non plan or with keywords use glm)
- opencode - this is what im mostly using now. similar to ccr but i find it a lot more reliable against alt models. thinking tasks go to claude, gemini, codex and lesser execution tasks go to glm 4.6 (on ceberas).
- sub-agent mcp - Another cool way is to use an mcp (or a skill or custom /command) that runs another agent cli for certain tasks. The mcp approach is neat because then your thinker agent like claude can decide when to call the execution agents, when to call in another smart model for a review of it's own thinking, etc instead of it being explicit choice from you. So you end up with the mcp + an AGENTS.md that instructs it to aggressively use the sub-agent mcp when it's a basic execution task, review, ...
I also find that with this setup just being able to tap in an alt model when one is stuck, or get review from an alt model can help keep things unstuck and moving.
reply