breakwaterlabs

If you believe in things like Artificial Analysis, GPT-6.1-Sol is by far more economical than Opus 5.5, even before the alleged (observed) nerf last week: https://artificialanalysis.ai/?models=gpt-6-astra-high%2Cgpt...

Notable takeaways:

- Sol-low - cost-competitive with Haiku, while outperforming Sonnet on medium

- Sol-High - Competes with Astra medium and beats Opus 5.5-medium, at 20% the cost

- Sol-Max - Beats Opus 5.5 high, Astra high, and competes with Opus xhigh, at cheaper than Sonnet and a fraction of the Opus cost

And when you look at time spent, With the exception of Sol-max you get better results in less time for less money. My own experience is that this is true-- it is difficult to call Opus strictly better than Sol, but Sol gets where it's going far more efficiently.

It's rather hard to look at opus-- or any of anthropic's models-- and say "economical".

> Because on that specific benchmark (FrontierCode 2.0), Medium is Carbon's best effort level

I am referring to the chart that cross-compares multiple benchmarks, which Artificial Analysis clearly shows Carbon-high as superior on. Further, AA's FrontierCode chart places Carbon-high as having a higher score on Frontier 2.0 than Carbon-med (https://artificialanalysis.ai/evaluations/frontiercode-2-0?m...)

Where it gets very interesting is when you look at the Cognition "FrontierCode" publish date (2026.10.27), and the revision as of 10.28 in web archive (https://web.archive.org/web/20261028175237/https://cognition...). This revision has a VERY different Carbon effort / intelligence curve where high beats medium.

It is only with the latest revision, as of today, that all of a sudden Carbon's benchmark curve goes cone-shaped. Curious, indeed....

I'd encourage you to take a look at a third-party's take on this and see whether Anthropic's self-serving report looks "on the level" after doing so: https://artificialanalysis.ai/?models=claude-fable-5-2-high%...

Just to follow-up-- artificial Analysis shows no justification for that Carbon effort curve, but it does show Carbon on medium smashing Fable on things like Terminal-Bench and AA-Omniscience v1.1 Non-Hallucination Rate.

But I suppose the reality is rather obvious when you note the funny games they're playing with the graphs, and the things they won't talk about (where's Astra on that CVE chart?)

- Why are they stacking Fable xHigh against Carbon Med?

- How did they produce such a bizarre effort-intel curve for Carbon that goes against every available benchmark?

- Why did they leave Carbon-- positioned primarily as a cyber model-- off of the single cyber benchmark comparison?

- Why do they switch to comparing against Argon (rather than carbon) for robodojo?

> People know how the sites they visit are funded and need some way to justify cutting that off.

As the admin for various SMBs I cared only in a vague, way whether their browsing experience was made unpleasant by ads.

I cared enormously whether I had to spend valuable time imaging or de-rooting their PC because some 0-click infection delivered by ad-network ransomwared us.

I have never felt even the slightest obligation to the vague collective of websites that rely on the shockingly dangerous concept of injecting unvetted, advertiser-supplied code into trusted domains.

From the moral angle, paywalls are an option and were in fact heavily used 25 years ago. The decision to rely for revenue on whether and in what manner a browser renders certain elements is certainly a choice a site-operator can make but it creates neither legal nor moral obligations on me or my users. My obligations as a paid IT / security practitioner are rather clear, and they are to my employers and clients.

I do not know the latest numbers on "what percent of web-based (or all) infections are mitigated by adblocking", but based on historicals that it is well over 50%. In that context it is lunacy to ask "but what about the site operators who create this mess".

They should have just hired the paint.net guy and rolled it in.

The point of paint is a tiny pixel editor. If you want layers there are better programs.

Others have argued the point at depth, but its somewhat more simple than that.

Is it plausible that Kagi's financials require them to run a tighter ship than established players?

Conversely, is it plausible that they are so popular that they can afford to be deaf to the negative UX of the extra sub?

I would suggest that the answers to those, and a quick survey of the heavy costs agentic use has imposed on site operator over the past few year5, suggest that this is a reasonable compromise that keeps costs down for the average user and moves agentic users to the API-fee paradigm that is already rather common.

> Today it is load-bearing:

Please no.

You should be NATing outbound 53 to a local resolver that upstreams to DoH/DoT. Relying on such an opt-out is rather optimistic.

It's shocking to me that, sometime in the last 10 years, everyone forgot that adblocking's primary argument is security.

Back in the bad old days of adobe and flash plugins in Internet explorer, malicious ads targetting those were the malware infection vector in something like 80% of cases. Browsers have gotten better, but it remains the case that ad networks are the single easiest way to inject your mal-content into another website.

In those days, any time I handled a client call about malware, I replaced everything with Chrome (built in pdf viewer, sandboxed/autoupdated flash) and adblock plus. And then, my calls from those users stopped being about malware or slow internet.

We see headlines about this-- about the Israeli firm that was abusing this, for instance, to hit targets-- and I am left scratching my head. Of course...?

IMHO it's the duty of any techie (with the social capital to influence digital hygiene) to ensure they and theirs are set up with at least ublock lite.

It may shock you to hear that there are other coding agents than Claude.

Comcast and/or verizon did it in the 2010s, there was an opt-out in the account dashboard.

It's sort of an absurd question, the prevalence of such things by ISPs was the literal impetus for DoH and DoT.

> Shouldn't this problem show up in a much larger portion of the population?

Casual awareness of studies on this suggest that it DOES show up but the symptoms (lowered immune function, seasonal depression, bone issues) are generic and subtle enough to fly under the radar.

It is not helped that, in my experience, many US health plans do not cover the cost of vitamin D testing without a diagnosis of deficiency (which generally requires a test). While that cost is usually low it does mean that it tends not to be offered because practices tend to avoid non-covered procedures for administrative and billing reasons-- you have to actually ask your physician to do the test.

My understanding is that it is somewhat difficult to end up at serum levels of vitamin D that create actual problems, so in general my not-medical-advice suggestion would be to ask your physician about taking a supplement: they're cheap and if everyone took them it is likely population-level health would improve, and it may dramatically improve your QoL particularly in winter months.

Do you have a cell phone? Do you use Wifi? Do you use an ISP who, in the last few decades, has performed any form of interception or traffic-tampering?

I hazard that the answer to all three of those is "yes".

The near-universal use of TLS and the absurd ceremony we go through to ship root CAs everywhere is all premised on the idea that MITMs are not terribly difficult.

The entire reason one can rest easy as they connect to their coffee shop's WPA2-PSK network, is the security provided by TLS.

And if you don't think an ISP or agency would abuse this if it could be done undetectably then you haven't been paying attention these last 20 years (hence why DoH is now a thing).

If your argument is that we should stick rather closer to the RFCs and wire-level truths when explaining TLS and the eMuhdra situation to laypeople, I wish you all the best in your endeavor.

I find that communicating just enough to allow the formation of an roughly accurate threat model better retains attention; and in this particular situation I was being somewhat glib to emphasize the point of how insane our PKI system is.

> IF they first spook DNS

Or arp poison, or BGP hijack, or get in-path via tenancy shenanigans... TLS exists, premised on the literal assumption that such things are possible. If your argument is that we don't need to worry about a TLS weakness because such things are hard, I'd ask why we even bother then. CA compromise is the literal worst-case in PKI-world.

The private key is a red herring, it is not used for "encryption" as such.

The certs are used to validate website identity, whereupon key agreement is performed to establish a symmetric key that has nothing whatsoever to do with your certificate.

All that matters here-- all that we use the keypair for-- is to prove that each party "is who they say they are", and that is something your CA can lie about.

If a layperson understands,

    able to issue arbitrary certificates that an arbitrary browser will trust 

as

    has the secret codes to everyone’s encryption

...then they're close enough, without digging into the RFCs, codicils, addendums, etc. They can perform a MITM, which is the literal thing that our PKI/TLS system is supposed to guarantee cannot happen.

> they are trying to maximize screen space.

Ironic to say on hackernews which shows massive whitespace in the margins-- and if I try to widen my tab sidebar it actually indents the margin rather than eating into it.

One can look across the computing space as a whole and see that, it is surely not the case that we're "maximizing screen space".

It's hard to explain partly because when you explain this thing we've done with TLS and PKI, it sounds a little insane.

> "We set up a system where every government and a bunch of random banks and IT companies in the world holds the secret codes to everyone's encryption. Then we make them pinky swear not to abuse it, with stern talkings to if they do. And now it's all on fire, because one of them was irresponsible!"

You can know the intractability of the problem that PKI is solving, and the (worse) alternatives that have been tried, but that doesn't make this system not-insane to a normal person, and its at least a little bit the fault of the collective "us".

And it's not like it is an impossible problem; we rely on TOFU for SSH, we have remotely-attestable TPMs everywhere, ZKPPs / PAKEs exist, session keys and federation protocols are widespread. Somewhere in the set of "possible solutions" are options that don't sound unhinged when you explain how we got to this eMuhdra situation.

EC, not RSA. RSA would *never* allow something like that.

It would need to be a 3, 5, or 7, since a prime is required.

> It is a presentational symbol through which forms of felt life become publicly perceptible.

It is utterly baffling to me that this is a thing someone felt needed expressing. Surely, intuitively knowing this is part of being human?

How else do we explain the blues, love songs... anything?

Of course it's AI generated, its a benign, beige opinion being drummed up as some incredible insight. You can use Awk on plaintext? Incredible! No one has ever in the last 15 years of journald debates mentioned this.

And every such burst of inspiration is accompanied by bullet-points to flog the dead horse further.

Lest you think it wasn't mind-numbing enough, now we're turning it into a narrative arc worthy of a screenplay: text, coming of age, has become markdown (which the author certainly loves). Completing the arc, we're going to acknowledge its limitations ("be unbiased / balanced"), and then come to a denouement where we tie it up with a bland bow.

If it were easy to see what's a number and whats a string, type confusions a la YAML's fun wouldn't exist: Is "no" a string, a country, a boolean? Is 0 a boolean, a number, or char[48]?

This is specifically an area that "text" is quite bad and is the source of much woe. Your provided example would be parsed by most parsers today as a string, since it mixes spaces (non-numeric) and surrounds it with quotes.

In fact, given well-defined binary, it is rather simpler to turn it into "text" than the reverse. It seems like you've forgotten that "0x32 0x20 0x37 0x20 0x2A..." is literally what is in that file you would view. Plain text is, after all, just one example of a binary format.

In almost all of these cases, it is both: the labs are operating the agents, while developing them, and committing their CFAA felonies in the process.

"The law" already exists, civil suits are a thing, and we no more need targetted AI regulation than we need targetted truck-rebar-incident regulation.

These incumbent labs are angling for regulatory capture by stirring up hysteria and suggesting that existing laws are insufficient. Don't do their job for them, first test whether there's actually a legal gap here.

By what yardstick should we then measure the DoEd's performance-per-dollar, or determine whether it represents "waste"?

Because it seems to me that a chief indicator that waste likely exists is an inability to show ROI.

> There isn't really much waste in federal spending

I do not believe anyone who has consulted in federal IT or worked on contracts is capable of making such a statement. All of the motivations for all parties are heavily bent towards waste, and in fact any attempt to make things more efficient (reduce headcount, for instance) would be recieved poorly by department heads and contractors.

As one example, a department head who reduces staff by 20% has just reduced their own next year's budget AND kneecapped their resume.

> Old people don't have that because they

Aren't insane.

When did the industry put the onus on the user to understand how the computer works? What happened to the old days of Xerox PARC's HCI studies putting the user first? The computer is in service of the user, not the other way around!

If I need to build a mental turing machine to understand your application, it is a bad application. It is rather the engineer's job to build a mental model of the user and their needs, and if you can't do that you should not call yourself a software engineer.

> I dislike how it is 'developers first' and not 'users first',

There are user-centric and dev-centric Linux distros. Windows is "Microsoft cloud onboarding" centric, and the experience has been dramatically degrading for years.

If that were not the case, why would senior executives at Microsoft say things like "we've heard you" and "we intend to reverse the suck in the coming year"? Even their management knows users hate the Win11 experience, and have placed it on their backlog....

> I dislike how janky its various GUI desktop managers are...igh pixel density, different audio setups, multi-touch trackpad support

These things are objectively better on a modern KDE linux. Out of the box I can output youtube videos to a dual-Sonos / Airpod setup by... clicking the sound icon, which pulls up an interface reminescent of "Windows 7, when the mixer wasn't terrible".

The reasons not to use KDE these days are because you need Windows software (usually: edge, teams, Office), or especially because LibreOffice is terrible. The core desktop experience, however, is notably and demonstrably less jank than the mess that is Windows 11.

The arguments in this thread-- amounting to "it's a good general practice because I happen to like it" (rather than "it is a sane / discoverable / usable default") are precisely demonstrating why these issues exist.

UX design is treated as a subjective matter, as if it is equally valid to clearly label UI elements as it is to have magic, nondescript UI pixels that serve as vital control surfaces.

Go watch videos of the research Xerox did on UI/UX and HCI in general, and weep for what we have lost...