I used to main Claude, but I can't stand how it writes. I feel like I'm wasting too much time trying to decipher the output. Adding writing rules does not seem to work. Now I use GPT 5.6 Terra high fast mode, with Luna for everything else. I might consider using Sol for planning. I can't stand using Sol or smarter models for coding, because they will eventually try to rewrite everything in the codebase.
I also don't want to use the Claude Code and Codex agent harnesses. The good thing with Codex subscription is that it can be used in other harnesses, unlike Claude. As far as I know, only Anthropic has this restriction.
I recently switched over to GPT from Claude once my subscription lapsed, it's too early to tell, but I do appreciate how more straight to the point Sol 5.6 seems to be, vs. the insane wordslop machine Opus tends to be.
It's really strange because when Opus 5 released, there were some that pointed this out, but a bunch simply said it was the best and as good as Fable etc etc.
but for me, it caused me to get very demotivated and avoid interacting with the model, at least when using Claude Code.
Funny enough, about 8 months ago I switched to Claude mainly because of MCP.
Now I’ve gone back to Codex because I simply find the ChatGPT models much more comfortable to work with. I spend less time fighting the model, correcting its direction, or re-explaining what I meant.
Me too. Claude Opus 5's English is insufferable. Opus 4.6-4.8 was more reasonable.
I've moved to Codex 5.6-Sol. Much saner English, much better at execution, and gets stuff done in a matter-of-factly kind of way (Claude Code is a mess these days -- it gets things wrong and goes around in circles).
But I'm harness agnostic and am not locked in. I just keep my issues in Kata Tracker (https://www.katatracker.com/) and switch harnesses/model when I need to.
Being loyal to a particular model/harness seem unwise to me.
> Me too. Claude Opus 5's English is insufferable. Opus 4.6-4.8 was more reasonable.
I suspect this degradation is happening because the AI labs are using the LLM's output to feedback into the input, to create a thinking loop, and they're optimizing that.
With the price of ram going up by 5x, I want to see our industry move in the opposite direction. Computers for regular people should go back to 4gb or even 2gb of ram, just to keep the prices sane. We should get software working well on those devices, just like it did a decade or two ago.
Small, memory efficient programs run better everywhere. It feels like theft to make consumers pay for our inefficient software. Electron / react on the desktop is such a rort.
>With the price of ram going up by 5x, I want to see our industry move in the opposite direction. Computers for regular people should go back to 4gb or even 2gb of ram, just to keep the prices sane.
Sure. Right along with Intel Core 2 Duo and Windows Vista.
Modern operating systems too. Windows xp only used a few hundred megs of ram. Windows 11 is more complex, but none of its features justifies windows 11 needing 100x as much memory to run.
Yes and no. If I recall, 32 bit windows has a fixed allocation of 3gb for userland and 1gb for the kernel’s memory space. I think 32bit only really works well for memory sizes smaller than 4gb. It might not be worth it, given a lot of OSes are in the process of deprecating 32bit support.
First, we don't need to be restrained to "how Windows did it twenty years ago." But either way, if we are targeting a 4GB PC, it's inappropriate for any single app to actually use more than 3GB regardless.
I'm waiting for a software release for my iPhone 16 Pro Max that will let me plug in a dock that has an attached SSD, keyboard, mouse, and display, and an option to pop up on my phone to reboot into MacOS.
Also, if they make a neo 2 from the A19 with 12 GB that'd be cool.
You’ll be waiting awhile. iPads have had desktop M-series CPUs for ages now, and you still can’t run macOS on an iPad.
It’s just a software restriction. I’m sure they have macOS running on iPhones and iPads internally, probably both through a VM and natively. But it’s very Apple to block this use case.
Guess cause it'll still do local builds and stuff, and probably do a lot more of those if the agents are spamming builds and it's something like Swift or Rust that takes a lot to compile
The bot blocking services are more advanced than that. They will do stuff like checking how your fonts are rendered in a canvas element to determine your browser and OS. A stock mac circumvent all that by hiding in plain sight.
I thought Caveman was debunked in the token saving argument? In real sessions it doesn’t really save much longterm and made sessions more forgetful/take longer.
According to Jevons paradox this will likely lead to more electricity used and more CO2 being released to the atmosphere, because it makes it more profitable to build another data center.
I've subscribed to Kagi for three of years now. Kagi is still great, but it doesn't feel like "Google ten years ago" anymore. I don't think it is because Kagi got worse, rather it is the web that has become worse. The content nowadays kinda sucks.
Yeah, it's become quite noticeable. The amount of clearly LLM generated sites that I've blocked just keeps going up and up. The only advantage Kagi has in that regard is that blocks are account wide, regardless of what device you use
Kagi nowadays has a feature called "SlopStop" that you can opt into. It's basically a global block list for LLM generated websites, etc that is shared across all users who opt into and anyone can report sites/content to it.
Kagi of course reviews it but generally it's good enough that I don't see much LLM generated content in the results anymore.
I was extremely pleased to login to my Kagi account and see that SlopStop was enabled by default. Thanks for the heads-up on this, and thanks to Kagi of course for enabling it.
They had quite some lag at the beginning. I made reports from the very day they announced it, and none of them were processed last month. But they are now! (Though they rejected half of them)
I had the same experience. I don't know why they haven't invested more in this feature. Heck, if they classified every site in their index as "AI/not AI" using machine learning, and had a large allowlist of "known-good" sites, that would take care of most of the problem (I am aware of the drawbacks of text-only AI classification, but they could use other indicators, like the site's publication date, domain name, domain registry date, etc.). I think for right now, the "state-of-the-art" for non-AI search is going to entail maintaining large whitelists/blacklists of websites that are crowd-sources by users (e.g. what you can find in some GitHub repositories). I was hoping that's what SlopStop was going to be, but it is not nearly as effective as it needs to be.
They started marking a lot of my reports as not slop when it’s very obviously ai garbage. Stopslop is nice in theory if they actually handle reports correctly. It’s exhausting to spend time trying to help clean up their results only to be told no.
I doubt that my list would be of much use to you, since a lot of the sites relate to specific hobbies.
However, as eloisius mentioned, you can view the most commonly blocked sites. Going through that list is a good start: https://kagi.com/stats?stat=insights
But this is also why Google "sucks now". The thesis of these competitors was that Google got worse because of bad product decisions or implementation failures. But I found that to be a very questionable thesis from the start.
> The thesis of these competitors was that Google got worse because of bad product decisions or implementation failures.
But they did though. Google results got significantly worse after ads started lagging in revenue. So they hindered search and increased focus on the ad platform which shot up profits in 2019 I believe.
The web becoming more inhospitable is true. But Google has the tools, the data, and the engineers to wade through the bs and return good results. They just rather have bad results and good ads because that is where their bread is buttered.
A lot of competitors are working with a fraction of the budget, teams and data. So asking a small team to do proper AI detection, to return smaller indexed sites with high quality info over mass linked SEO abused hellscapes is very complicated, but asking Google to figure it out is proportional to their talent, expertise and resources.
It just not going to move their bottom line, so search is de prioritised compared to ads (some might even argue its made worse on purpose to make ads look even better)
This is the thesis, I understand the thesis, I remain skeptical of it. In my view, search sucks now primarily because there just isn't much worth searching for on the open web anymore. I think the "it's Google's fault" thesis is mostly an unwillingness to admit this.
To large extend, that content stopped dying after Google deprioritized it in search results. People stopped being able to find it, sites/communities stopped having views, authors gave up.
I stopped caring about my blog about that time - after I was unable to google my own postd. I remembered the name and some of the content, but it was just not coming up in the search. There was also noticeable drop in views. I kind of concluded it is pointless when it is not coming up in searches unless I SEO hard.
I could be convinced of this narrative. Like, if someone writes a book about this, with good evidence of what caused this to happen, I might find that persuasive.
But to me, what happened was social media and video becoming the locus of attention. I don't think people stopped putting content on independent websites because their traffic from search results went down, I think they just started putting content on Facebook and Twitter and Instagram and YouTube and TikTok instead. To me, the whole way the web is used just changed over this period of time, in ways that mostly had very little to do with Google's search algorithm.
I don't see why it can't be both. The web has less good content AND Google has altered their search results such that results are worse generally for us, but better for them and their bottom line.
I'm not saying it isn't both, what I'm saying is that the thesis of the newer Google competitors is that the problem is mostly that Google has chosen to suck, so they can do a lot better just by choosing to not suck. But I think that thesis is wrong, that it is mostly that the web sucks now, and that these search engines are ultimately likely to be unsatisfying because they can't actually solve that underlying problem.
But if I'm wrong, that would be great! The comment that started this thread made me think that it seems like I'm not wrong though.
I think the counterargument would be that the web sucks in large part because of the way it’s been shaped by Google. Most of the hyperwordy blogspam is driven by sites trying to optimize their SEO for Google. So websites like Kagi could potentially push the underlying web in a positive direction
I don't buy this at all. There is no way for any search provider to avoid the arms race dynamic. Marketers will always have an incentive to appear first in results in order to capture an audience to monetize. "Just win the arms race then" sounds nice but it's wishful thinking, IMO.
And I don't even think this is the primary issue with the web sucking. The web is just an entirely different thing than it was during the period we're all nostalgic for. I used to search for information and find it on independent pages that I never would have found without a search engine. But now that same information I used to search for is all on wikipedia, with much better citations than I used to get.
And pretty much everything other use case of search is now behind some social media walled garden.
IMO the web was kind of just a victim of its own success. There stopped being much incentive to put new things on the web on a huge proliferation of independent sites.
(NB. This is terrible for Google! If my take on this is right, then their business model is going to eventually be in real trouble. This is another reason I'm pretty skeptical that they have caused this to happen. It's bad for them, I don't think they're happy about it at all.)
Google can both be worse because of the web results declining in quality and terrible product decisions. It is of course much easier to see one of the two, and make an alternative search interface possibly using different rating criteria. It just can’t solve the entire issue, because spam is also absolutely cheap to mass produce now.
Yes I agree. I don't begrudge competitors trying to do better at search at all. Maybe they'll succeed. I've just always been skeptical that they will be able to make significant progress on this problem, because I think the problem is ultimately garbage in garbage out.
The percentage of quality content on the web has gone down but it still exists. I usually use Marginalia Search (https://marginalia-search.com/) when I'm researching something and not finding any good results on Kagi or Google.
imagine if we had a browser like kagi with its own engine from scratch where you are not the product and it charged you a monthly subscription for privacy free browsing
You're the product in Firefox, just a different kind than in chrome. Ad Nauseam is silly because ad companies know they are fake clicks and don't count them. Ads aren't the only thing you'd need to block from the internet.
There is zero evidence Ad Nauseum does not work, but there are plenty of signals it does. Google banned it from the Chrome store and setting any threshold below one hundred percent click rate just uses random.
Even if it is somehow detectable it's just Ublock origin and not every ad network is worth billions of dollars
I mean, Orion exists (on Mac at least, although I heard there's a Linux version now?). Might not have its own engine built from the ground up, but WebKit is pretty much the next best thing.
The increased update timelines by Google, Samsung and others roughly coincided with EU legislation coming into effect that mandates 5 years of updates after end of sales. We'll see.
Correction: if the manufacturer chooses to provide updates, and they don't have to, they must continue to make those updates available for five years after end of sales.
In other words, manufacturers aren't required to publish updates at all, but if they do provide updates they have to make them available to users for five years after they stop sales. This only stops the case where a manufacturer ships a device and publishes updates for the device, but then takes those updates offline after they stop selling the device (but before 5 years is up).
Google promised their Nexus phones would get new versions of Android for X years then, after selling a bunch of them, just changed their mind.
I'm having a hard time googling it since every result that comes up is about Google cancelling Nexus phones entirely way back when, but I remember a lot of Nexus users were kind of PO'ed about it.
I think we need to change our perspective of what success is. I believe there will be a ton of small companies popping up instead of a few big ones that eats everyone's lunch. Like Google, Microsoft and others giants have done until now.
I also don't want to use the Claude Code and Codex agent harnesses. The good thing with Codex subscription is that it can be used in other harnesses, unlike Claude. As far as I know, only Anthropic has this restriction.
reply