merlindru

i think its a combination of these:

- OAIs models are much more token efficient. in some cases their new mythos-level model uses 1/4th of the tokens that mythos does.

- they reportedly have recently found another way to decrease compute needed "by half" in a breakthrough

- early on, they ensured they have much more compute ready, whereas Anthropic had to disable new users signing up because they ran out of compute

- OAI has fewer users (5mln codex users vs, what, 10mln claude code users?)

But those hacks and exploits have always existed. Just had to have the right people to find them / be sufficiently motivated.

The same models that can find these exploits can also help fix them, thus everyone will be better off.

Relying on the fact that nobody has found a security issue with a piece of software yet is not a great way to ensure safety

ah i see your point - my bad

right? the horrors!!

seems like the politicians are finally realizing what we've all been up to

this is basically trying to enforce security-by-obscurity, which is a terrible idea all around. it's just a model. the security issues still exist and are exploitable.

and after staking the economy on AI, you can't really put a cap on intelligence. if models are not allowed to be better than Opus 4.8, then the whole investment structure is about to unravel.

why invest billions and billions into AI if returns are artificially capped?

If you asked the same selection of people about Saudi Aramco or SK Hynix none of them would know what those are either though, right?

I do think 60B for Cursor is way overvalued. Just not sure how to quantify

They use intercom's Fin AI. Probably powered by a Sonnet or Opus model.

That said, it can't handle legal/refund/complicated requests and just forwards to a human for those

Unrelated, but while the tech of anthropic seems to get more impressive with every passing month, their support has taken a nosedive, sadly. Yet they continue to be the favorite. Model performance is deciding above all else.

I used to get a response within 24 hours back in the Claude 1 days.

In January 2026, it took 2 weeks.

For my latest support inquiry, I've been waiting for over 8 weeks for a response. Eight!

During early testing, Stripe reported that Fable 5, [...] in a 50-million-line Ruby codebase, the model performed a codebase-wide migration in a day that would otherwise have taken a whole team over two months by hand.

EDIT: I misread. This comment previously talked about 50 million lines being migrated. Instead, in a 50M LOC codebase, one specific codebase-wide migration was done.

Very impressive, but obviously not on the order of a whole-codebase migration

Probably. It just feels so out of character. Especially the fact that it can't be safeguarded and generate sexual innuendos, for example, which Apple usually hates.

I'm not even entirely against genmoji. It's just odd for Apple to do.

i dont agree at all. image generation feels very odd for Apple.

and usually they don't care about glaring omissions like this, either: iPadOS was lacking a calculator for yeeeeaaaars.

they repeatedly said they'd ship a calculator if they can do something special for its introduction, and only then.

so why did they lose their hesitancy to ship mediocre stuff here?

that's the most common criticism. can't say i disagree.

though monopolistic practices are unequivocally bad and used to get struck down. one may argue this is just another instance of disallowing monopolistic behavior.

the DOJ used to have sharper fangs than what the EU is doing now

i would say that that also doesn't change the essence of the moment displayed in the image, unless the image itself is about planes/projection/...

absolutely agree.

but Apple isn't known to make grand promises and then not keep them, is it..? usually they just deliver what they say they will

yet i've been reading about "well they promised AI Siri two years ago and Siri still can't set an alarm right" in every thread even remotely related to the topic

i don't remember reading this much about anything else. it seems to have soured people quite a bit, at least in my internet bubble

thats exactly what i'm saying they should be doing, but aren't. we're in agreement!!

why don't they just wait and not ship any AI junk at all? instead of promising a Siri AI rework, which then doesn't deliver? or Image Generation stuff that feels wildly put of character and generates tasteless and often downright creepy images?

not to mention that all of the new AI stuff they announced won't go live in China and the EU for a while.

why not do exactly what you proposed and wait it out? instead they seem to be trying to deliver AI stuff and just unable to.

there's also reports that apple execs held a secret emergency "oh shit what do we do about AI" type meeting.

they very much didn't intend to be this behind

cropping doesn't alter the moment captured. or, rather, you're also cropping when taking a picture - no camera can capture all 360 degrees.

so what is being captured is altered from the get.

yet i don't think you'd argue that the act of taking a photo is the same as photoshopping a giant giraffe into a photo :P

i don't think arguing that taking a photo which is cropped/enhanced (as in sharpening, color correction, ...) is akin to changing what's displayed in it.

those distortions only serve to let us better focus on what truly happened in a moment, not change the moment in essence.

the EU feels like this falls under monopolistic practices, which it has deemed illegal. the buck stops with the politicians - there's no reason other than "the EU thinks it's bad for the economy and should thus not be allowed"

do they have to let Anthropic and Deepseek run their models on it too?

provided it gets big enough, yes. the EU's position roughly is "if this hurts an entire market just to benefit you, and lots of people use / rely on it, then you gotta allow it"

what's worrisome is that they continue to fail at it. it's one thing to say "we're still hashing things out". it's another to parade around Image Generation features that are obviously widely not-cared-for and oftentimes actively disliked

Apple cares greatly about their brand yet this has hurt their brand like nothing else in the past decade

Ironically that's often what ends up costing one the most, i feel like

Sent you a mail! This sounds incredibly interesting.

Location: Germany

Remote: Yes

Willing to relocate: Yes, especially for great fit

Technologies: Great with TypeScript. macOS development, especially around accessibility APIs e.g. built a platform to let agents interact with apps (getinvoke.com). Good with React, Svelte, Vue, Solid, all things frontend. Good with UI/UX, would consider myself a strong Design Engineer. Good with Rust. Some audio-related work. A little cryptography experience (wrote provably-fair gaming algorithms).

Résumé: https://cv.merlin.audio

Email: merlindruzinec@gmail.com (personal) or merlin@merlin.audio

SEEKING WORK | Germany | Remote (any time zone)

Technologies: Anything. Very strong with TypeScript, frontend and backend. Dove deep into every frontend library/framework under the sun. I love Bun. Very strong with Rust. Have used all mainstream languages.

I've built getinvoke.com - an open-source platform over the macOS accessibility DOM. It enables deterministic computer-use for agents: they explore an app, click around, etc. Then they note down the actual working steps for any UI task in simple JS. Then replay them.

Got good grips on how React, Svelte, Vue, Solid.js work, both internally, and in practical terms. Led the transition for a 350k MAU website from a jQuery mess to a rewrite in Vue + redesign of the component system.

I've built Ethereum/Web3 payment systems before.

Worked with realtime online gaming and implemented cryptographically provably-fair algorithms.

Came up with & implemented a pluggable AI workflow system (think n8n) in the GPT-3.5 days.

I love design and UX.

Email: merlindruzinec@gmail.com or merlin@merlin.audio

why would Google help a competitor like that, though?

Could make it disable-able only from the terminal in recovery mode. That one would be too hard / bothersome to fend off most cases I feel like

How can you be sure you weren't placebo'ing yourself into feeling it?

To my knowledge creatine has no significant effects until your levels rise after, say, a week of taking daily

Tosche also has a very well made "Comic Code" font with ligatures

Recursive is terrifically legible, especially with Casual all the way on.

I couldn't tell you why, but reading code feels much, much more natural with it than with most other fonts.

Perhaps it's the high degree of separation because every character looks meaningfully different?

The "kerning" (or whatever the visual space between letters is called in monospace fonts) is also among the best.

yes but the compensation part is the problem

An 80h week should not be compensated the same as two 40h weeks

it's just extrapolation.

it's

    (revenue of this month) * 12

that means, of course, if you make $1000 in january, your RRR is $12000.

...even if you end up making $0 every other month and thus only $1000 total that year.

thats why RRR is perhaps harmful. especially when it's not growing. it can be much bigger than the actual revenue. in anthropics case it's rapidly climbing, though, so it underestimates revenue if that growth keeps up