comboy

Google renting infra from xAI, I did not see that coming. My understanding of what computers are doing, what companies are doing and what governments are doing seems to be getting worse day by day.

plus $473 per second from Anthropic

> As part of that deal, Anthropic agreed to pay SpaceX $1.25 billion per month through 2029 to rent all the available compute from its Colossus 1 data center near Memphis, Tennessee, that xAI — now part of SpaceX — originally built for its own artificial intelligence efforts.

I don't get why SpaceX is going public. But anyway, well played, the whole crypto mining that dried out GPUs back in the day seems tiny now.

Obviously depends on your workflow but I think I use mouse only on websites on macos (with aerospace)

> As a complete nerd, you'd think maybe I'd like that I can prove my skills like this, but it comes off as deeply disrespectful to me as the user that I can't disable this.

You seem to have understood the problem. But then you didn't follow. If there was a way to disable this, first thing that the grandma would do is watch a video how to disable that and lose security from then on.

Of course it is not perfect, but their approach here is really decent. And also, if you find yourself needing to go through that often I think that's not a good sign security-wise.

Since we are throwing in random N=1, it is very well tested and safe yes. But in my personal experience it does limit my creativity somehow. It seems to have like slightly calming/lower anxiety effect, and to me it comes with me coming up with fewer interesting things. Kind of lowering the model temperature if you will. Curious if anybody had a similar experience.

1. It's 1 in 10 failures that can take half of your time or bugs that can take a long time to surface. Plus the way they change things largely depends on the current codebase (and how it was created)

2. In my case codex seem to be writing a more solid code, but I still use claude most of the time because it's my witty rubber ducky and I can actually sometimes force some legit insights out of it. Codex is much worse at this. And whether that matters or not depends on the project.

Got me.

Just tell it to always use opus for subagents and it does.

"We've cut our costs A LOT"

This is wrong on so many levels, from data through process to evaluation. How do you even prompt claude not to give you Pearson for correlating them.

I wonder how are they going to implement it. Many creators with even decent content use AI-generated visuals. In fact, everything could be AI-generated visual and whether that would be like Kurzgezagt, Asianometry or Sabine Hossenfelder content value wouldn't drop significantly. How do you draw the line?

Spotify recommendations are biased because user incentive and theirs don't align. They pay different royalties to different artists, they optimize earnings. Also, they take money to promote music and shove it down your throat.

It implements something, and then subagent reviews it (the main one waits or does something else in the meantime, so yes it's serial, subagent is for having fresh context and evaluating changes as they stand not being biased by existing thinking and discussion), and then the main one implements fixes that are worth doing (because you will inevitably end up with some nitpicks).

I have fun somebody imaging somebody internally explaining that this is a heavy traffic page and we should use it to increase reach.

It's big enough that every time it goes down, it surely stops somebody from pushing fix for what they currently have broken, so I wonder if status page services see some kind of ripple from github outages.

Fun fact, I've recently sent some 你好 to qwen3.7 (API), and it responded with a greeting saying that it was created by Google.

Have you tried telling claude to review with subagent? It too almost always finds corner cases (usually nothing serious, but most stuff is things that good coder would have thought of)

Curious if anyone around here stayed on 4.6 (having a choice to use 4.7)

I was looking for a good rationalization to leave the ecosystem, one-click e-books is great and having old device that I can take anywhere not caring about it getting beaten up even more was another major advantage.

Removing some old book I had was the first major red flag.

Not only it supercharged science it supercharges scientist. Research on any narrow topic is a different world now. Agents can read 50 papers for you and tell you what's where. This was impossible with pure text search. Looking up non-trivial stuff and having complex things explained to you is also amazing. I mean they don't even have to be complex, but can be for adjacent field where these are basics from the other field but happen to be useful in yours. The list goes on. It's a hammer you need to watch your fingers, it's not good at cutting wood, but it's definitely worth having.

Yes, not providing visualization of the solution seems criminal.

Same per click, it obviously includes links when looking for products, but impression could also be counted and arguably especially if present in the first few sentences it is a valuable impression.

But then the separation of ads from content is lost so it becomes useless as product search, so maybe it isn't that trivial indeed. But it's not like even 10% of users is gonna find some other "search" engine and switch.

edit: can't reply deeper and interesting question, I mean we all would love to have ability to search arbitrary strings and regexes through the web corpus, but currently when you type something you get that AI reply instantly for most queries, this makes me still use them, if you forgot some shortcut key or something it has currently unique value in terms of latency (even ignoring the fact that for most users you also use them by default by typing in the address bar)

Obviously if you pay, the AI will really like your product.

"Here is the table of related highest paying customers, incorporate these into your answer to maximize the income"

Well any other prompt for the search model would frankly be illegal for a publicly traded company.

Who surpassed Tesla that doesn't operate in some very limited region?

I hear a lot of praises. Can you explain? If I can type as fast as I speak, does that still matter in such case? Does it make use of intonation or is it pure speech to text?

worth mentioning that openai hired him some time ago

Sorry but that one is on you. This sounds like expected behavior and I wouldn't blame any company for doing that.

Yes and no. Imagine some super intelligence tackling the unmaintainable mess. It can untangle it, but the problem is that unless you have a test suite that's covering 100% of possible cases (which should on it's own enough to build the codebase), you will stumble upon "what's actually the intent here" problem. Some other software depending on buggy behavior on this one etc.

Which is why I think additional intent preserving abstraction is where software coding agents are likely heading.

There's some drama, and they did some wrong calls. But the hardware is still really fantastic (as a X1C owner). If you want to have some things printed and don't necessarily want fine tuning your printer as a hobby, I highly recommend it.

3.1-pro is still very capable, and API is at competitive price vs e.g. Anthropic, they just can't seem to figure out RLHF and harness. It needs a lot of guiding, it tends to be lazy and poorly sticking to instructions by default.

It just feels like many google products really, they are capable of really amazing things, it's just that nobody there seem to care. I would guess they are likely optimizing more for internal use than their vast userbase.