wongarsu

That vacant land tends to be far away from electricity users. Not always - lots of desert around Las Vegas - but people historically like settling around arable land. Producing electricity close to where you need it is great, both because of transmission losses and because building more long-distance transmission lines is neither easy, cheap nor fast

We are right at the point where fully automated harvesting machines start to gain ground. And removing the operator (or moving the operator outside the machine) allows for designs that are less tall. The benefit of wider and wider machines also shrinks with full automation, even if it doesn't fully go away

The decade after the dotcom crash brought about as many if not more long-term successful products as the dotcom boom: Youtube, Reddit, Spotify, Google Maps come to mind

A bubble that funds all kind of crazy ideas and big investments into infrastructure work is great for laying the groundwork. But the end of the bubble won't mean the end of progress, or even necessarily a slowdown in progress

The issue with storing the memories in some encoded representation (e.g. in embedding space) is that this would turn that representation into something that has to be supported "forever". I have projects that I have worked on with Opus 4.6, 4.7, 4.8, 5, 5.5, as well as Fable 5 and 5.1, and obvious some Sonnet subagents along the way. And I expect to keep working on the same codebase when Opus 6 and Opus 7 release. Text is trivially compatible with all of those, but their embeddings all differ. And whatever other representation you come up with right now in Opus 5.5 might no longer make sense in the model architecture of Opus 7

Imho the interesting challenges with memory are about "what does this memory belong to" (attaching it to lines of code or similar), "what does it relate to" (webs of memories), "is this even still true" ("sleeping", "reconciliation", etc). But there's no real consensus what is even useful to handle, never mind how to handle it

Auto mode seems to have some reasonable restrictions baked in. For example if you give claude access to two k8s namespaces, seriousproject-qa and seriousproject-prod, auto-mode will grant `kubectl exec` on the qa environment while denying the same command for the production environment. Of course you should be enforcing that by restricting what kind of access tokens you give claude access to, but auto-mode does have opinions about this stuff too, even if it's not bulletproof

However I do share your desire for a better manual mode. Auto-mode is both sometimes surprisingly lax, and sometimes too restrictive. Maybe manual mode should inject a reminder "keep your commands as simple one-liners", the same way plan mode injects a reminder that you are in plan mode and don't want any code changes

My brain parses it as "communication" when reading quickly

Same first letters, the dots and the t roughly line up, apparently good enough for me

Big L when people do that

And some of the most common writing advise you will find, especially around getting out of writer's block, is to just write stuff down without correcting errors. Just get the first draft on paper. Some authors even choose to use typewriters or a pen because those don't allow them to easily go back to fix typos

Then in the second and third pass the author is obviously more concerned about fixing continuity, story and flow than about hunting typos.

Walmart is in a mature low-margin business. Their global expansion has stalled, and in the US competition is increasing

There are also many positive things about them. They are huge, and innovating. But their growth potential is somewhat limited

And you could just as easily frame it as Musk "gifting" something to the US government by offering cheaper launches than Roscosmos and Boeing/ULA, and developing new capabilities cheaper than Boeing

From my memory, a lot of the excitement was directed at what the design was trying to achieve: a structural body made from easy to make pressed steel panels. This was meant to enable a much cheaper car while also removing the bottlenecks plagueing Tesla's production lines at the time

None of this materialized. The truck that ended up in production was a traditional body construction, but with unpainted steel panels so it looks like the truck Musk presented. It ended up with all the drawbacks of a steel body, while still being expensive and hard to manufacture

Over the last 3 years we made a lot of progress in efficient training methods. On text models, you can now train GPT-2 levels of coherent English from scratch for about $50 (e.g. using Karpathy's nanochat [1])

1: https://github.com/karpathy/nanochat

Ranges are still somewhat useful. If they low-ball the upper bound, anyone expecting close or above the upper bound is unlikely to apply. So there is incentive to be honest about that end at least. But as an applicant (or current employee), knowing the upper bound the company is willing to pay for this position is useful information for salary negotiations (and to enable you to spend your limited time and resources in applying to well-paid jobs)

Obviously it's not perfect. But as you say, it's a difficult problem. And salary ranges are at least something

At the scale of a McDonalds ad campaign that seems easy to measure: go to the nearest Target parking lot and ask as many people as you can two questions: "are you vegetarian" and "did you see {description of ad here}". Then compare how which percentage of vegetarians saw the ad compared to the percentage of non-vegetarians saw the ad. There are even companies specialized in market research that will do this in a more representative and statistically rigorous way

The opening is fine, but when you click "Start the lessons" you get a six-pane interface that imho is pretty overwhelming

Yeah, I can click around and follow what's going on. But I'm with GP: this is not making it easier to understand. It's great at being Hollywood UX: looks pretty, stuff moving everywhere. One click on "step" makes the image animate, and change the text in the left sidebar, top middle, and right sidebar, in addition to stepping the table in the bottom middle. But it's almost the opposite of distilling information

This is not "kind of close, but not as accurate as a ray traced image", it's just completely wrong. The whale itself is fine: sun comes from behind, and sun rays refract inside the whale to make it glow. Believable enough. But then the whale is somehow shining light on the clouds below it, as if the sun was high in the sky and slightly to the left. Which is decidedly not where the sun is. And then out of nowhere light from round church windows shines on the church - but nothing on the whale has that shape, and that side of the church isn't even facing the whale

I can only conclude that this image is taken on a world with two suns (one in frame, one out of frame to the top left), and somewhere out of frame in the same direction are floating church windows. Even then we might need a third sun to make this work, and the light on the clouds is still the wrong color

So you're claiming a Claude-shaped readability gate? /s

While sitting at the desk, I type about as fast as I talk. While unloading the dishwasher I'm a much faster talker than typer

Their last investment round in May had them valued at 965B. And the company has grown a lot since then, those investors obviously expect a profit

I agree. Their financials are fine. They look like a company that has solid foundations, with room to grow but close enough to profitable that they can survive the end of the AI boom

But are they still a 1.5T USD company if their growth is already slowing this much?

If you go just by operating income, they showed they can be profitable in Q2, and the operating loss in Q3 is fairly small (in relative terms. For every 32 dollars in revenue they lost one dollar). They haven't shown wild profits, but nobody expects them to. Any money they make has more value when it's reinvested into growing their market share and growing the market, making next quarter's numbers look better. And imho they have solid enough financials that they can make the bleeding stop if the money faucet turns off

All of that isn't very newsworthy though. We knew all that from the Q2 numbers. What is interesting is that reinvesting all their profits to make the number go up in the next quarter didn't work as well this time as it did all the other times

Interesting. Their revenue has "only" gone up 47% in Q3, compared to doubling revenue in most previous quarters. Meanwhile their "compute and infrastructure" cost has increased by 77% compared to Q2

Maybe they tried too hard to get good Q2 numbers, making revenue happen earlier and expenses later, and that now makes Q3 look worse? Either way, this does not look great

Direct link to the prospectus: https://www.sec.gov/Archives/edgar/data/2130357/000162828026...

Consolidated financials are on page 23, quarterly numbers on page 160

Imho the strength of Indie games was being able to try ideas too risky to bet AAA budget on, or to target niche audiences deemed too small to make AAA profit

Neither of these is going away. The marketing strategy of teasing your game for months or years to build a fan base is going away. But if you life by "put up or shut up" I believe there is still room to win on vision alone.

If you don't have a unique enough vision (like most AAA, visions are risky) then I agree that you have to migrate up and just outspend everyone else

Why would a benchmark that shows "JPEG decoding - 20000ms, Nibnib decoding - 13ms" on some microcontroller be pointless? That'd be a great benchmark, if it was provided

If the issue is memory required for decoding the author could offer a benchmark on the smallest hardware that can still do both, with memory figures provided. Or the size of the largest image that can be decoded on an RP2350 with each algorithm

Now just add a ranking

Obviously the dropbox comment would be the most cited. But I wonder what number two would be

The code it writes in the first attempt is at least as buggy as a human's. So the question is more whether it is as good at spotting bugs in its own code as it is at spotting bugs in human code

I'm tempted to say "no, it isn't". But I don't have numbers to back that up

Have on-call sysadmins or have them work in a three-shift schedule. That way there's always someone there to fix stuff as it happens, including deploying security patches or worst case taking affected systems offline until they can be patched

This used to be common before devops and cloud took over. Looks like it might make a comeback. Maybe with a new fancier job title

The argument for bitcoin being secure is more along the lines of "if you store your own bitcoins securely nobody can take them from you". Storing them securely is easy in principle (just keep them on an airgapped computer, or write the key down on paper, then "just" do physical security well) but hard in practice (major security-convenience tradeoffs). However, storing your bitcoins with a major provider like binance has a terrible track record and not something the hardcore bitcoin advocates actually recommend

As someone who was a beginner once, I found nano leaps and bounds more intuitive than vim. You can quite easily stumble your way through nano with minimal exerimentation, and once you figured out that ^ stands for control you can just use the handy always-visible command reference at the bottom

Sure, GUI software is much more intuitive. But if you start with the constraint of being a TUI editor with no mouse support, nano does extremely well