ololobus

The new models are quite smart, so suffocating their context with dozens of MCPs and skills isn’t necessary like it used to be.

Genuinely curious, how MCPs can suffocate the context? And what exactly do you mean by this?

I have probably more than a dozen of MCPs enabled in my Claude Code (slack, jira, github, many internal ones), and I have never seen model calling into them unnecessarily unless it’s explicitly needed for a task. And in the latter case, well, it cannot do much without the right tools access (MCP in this case).

Skills and plugins are a bit of grey zone, yes, but even there it heavily depends on what you put there. Just plugin loading always takes infinitesimal portion of the context in my experience

Then use a sort of "reverse" parliamentary trick: the default is that this legislation is accepted. They needed an absolute majority - not of voting members, but of all members - to reject it.

No pun intended, but how is this legal? I mean, if you don’t have a quorum, then shouldn’t you just wait for an Autumn session? It feels like having a democratic parliament with backdoors like this kinda undermines the whole idea

Games are no longer about having fun, they're about how much money or activity can I extract from the player (yes, even the indie experience is tainted: buy off Steam and 30% of that purchase goes to the platform for just existing).

This is a HUGE overstatement, tbh. Without such big worldwide distributors like Steam, the only option an average indie dev would have is to distribute their game on their website that nobody knows about, so instead of hundreds of 1000s of sold copies they would have just some 1000s

Another non-obvious outcome of Steam is that the pirate scene largely just went away. It so niche that nobody really cares about it for real. It’s just so much easier and safer to buy

So I would say that Steam, Apple and so on have earned their 30%, they really created a huge market everyone can get a share in. Surely it should be regulated to avoid hostile monopoly behavior, but that’s a different story

It upsets me I never got to live the "glory days" of cars, technology, outdoor activities, music, entertainment, etc.

I’m getting closer to 40, but I sometimes feel the same. Like people were buying houses for a fair amount of money in 90s-00s, when I was going to school. And now it’s like you need a fortune to do the same

But generally, I think it’s more about human psychology. For many people, the past may feel more attractive. It’s a mix of nostalgia, current day-to-day problems, and the fact that you cannot really get there to check if it really was that great :)

Human history has known much worse times than now, and I think we all need to remember that we shape the present and the future in one way or another. Yes, you and I have way smaller levers than D. Trump or Elon Musk or some average CEO, but we do still have. Every day we make decisions that influence our life and the lives of people around.

Don’t want to sound like that social media influencer, it’s just how I explain it to myself too

Idk, I actually got the opposite impression. Most of the info is just what I would expect everyone to see: date formats, languages, various webview kind of stuff, network info. This is already more than enough for fingerprinting

information such as apps installed

This is what surprised me too, but if you read their hint, it’s not like list API. They probe various ‘open URL in app’ to see what apps registered them, so are installed. I guess this i) won’t allow you to track apps that don’t have ‘open in app’ urls, and ii) probably hard to limit without affecting UX

number of copy actions

This is odd, yeah, not sure why is it exposed

last wipe

They deduce this from the volume creation date. Probably possible to hide, but also not really that important, at least to me. Fingerprinting will work with way fewer info anyway

To summarize, I think iOS is still very solid in terms of involuntary info exposure (if you trust Apple itself). Most of really sensitive info requires separate permissions. Yes, you can harden it further, but that will be more like a paranoid mode

You are right, but I think you miss the whole point of the agentic workflows that are being discussed in this post comments.

Yes, you surely can read man, docs, whatever, then DIY. The point is that in many areas people don’t really want to become an expert, like in ffmpeg cli arguments, they just want the work to be done. Above is an example of agent being able to do it locally, and I think it’s great

Just speculating and thinking out loud. I think this might be a good news for AI-skeptics. Going with IPO means that investors finally want to get some cash that they cannot get by any other means. There are good examples of private market companies staying like that for many years because they are profitable, they have plenty of cash, and they have a queue of investors eager to put more cash

So what does it mean in this particular case? The board and investors probably don’t see it being realistic to become profitable soon, and maybe even worry about AI ceiling, so they want to profit now

Idk, with the One they already seem to try to claim too many things for a single device. Adding a keyboard and a bigger screen will be even bigger scope creep. As a Zero user, I really like the compact form-factor

And just personal imo, for coding on the go something like macbook air seems to be a way more comfortable option. I know that you wrote that you fit gpd in you pants, but man, you know that this use case is even more niche than flipper zero

The job market all over the world is ultimately changing. Wars, AI, energy crisis, etc. — it’s a combination of factors. Yet, the article is too shallow, so it doesn’t clarify much.

The two examples are not really representative, “press spokesman at a small industry association” and entry-level “apprenticeship in marketing communications and a bachelor’s degree in international management”. I don’t want to say that they are completely bs jobs but, well, these are quite niche. Both seem to be only ‘affordable’ for a strong economy, not during an economic instability.

What I’d like to get answers to is why if everyone says about shortages of nurses, doctors, teachers, plumbers and other handymen, highly qualified engineers capable of making some complex stuff like rockets; I don’t really see any policy makers pushing to make such jobs more appealing, I don’t see people around talking about moving to any of such areas even if they struggle or lose their office/corporate jobs, or talking about their kids learning to do one of them

I’m also confused by their examples. All of them seem to be perfect renders/exports from 3D models — this is not the use case where I see it the most useful. Making a parametrized CAD model out of a hand-drawn sketch — yes, please

I think this is true under assumption that you know the CAD tool well. From my recent experience (I have a 3D printer), I regularly find myself in a situation where I know what I want to do, I can do measurements and I can make a sketch on a paper. Yet, making it a proper 3D model in something like FreeCAD is super tedious. I know OpenSCAD relatively well, but when it comes to something more complex I struggle a lot. The recent example, I was making a water tap for Lego duplo kitchen sink for my little one :)

So I would really appreciate a good AI/LLM tool that I can feed my sketch and parameters and it can save me hours of searching web and watching tutorials on how to extrude a circle over a curve

BTW, any existing AI tools work really well with OpenSCAD, so if you want a parametrized model that can be made out of simple shapes, I highly recommend this flow

Does it require core patches or I can install it into the standard upstream Postgres? Asking because, afaik, it did, but it might that something has changed already.

Interesting point of view, didn't know about Jevon's paradox before. To me, the outcome still depends on whether AI can get superhuman [1] (and beyond) at some point. If it can, then, well, we will likely indeed see that suitable-for-human areas of the intellectual labor are shrinking. If it cannot, then it becomes an even more philosophical question similar to the agnosticism beliefs. Is the universe completely knowable? Because if it's not, then we might as well have an infinite more hard problems, and AI just rises a bar for what we can achieve by paring a human with AI compared to just human alone.

[1] I know it's a bit hard to define, but I'd vaguely say that it's significantly better in the majority of intelligence areas than the vast majority of the population. Also it should be scalable. If we can make it slightly better than human by burning the entire Earth's energy, then it doesn't make much sense.

I'd totally agree with this point if we assume that efficiency/performance growth will flatten at some point. For example, if it gets logarithmic soon, then the progress will grow slowly over the next decades. And then, yes, it will likely look like that current software developers, engineers, scientists, etc., just got an enormously powerful tool, which knows many languages almost perfectly and _briefly_ knows the entire internet.

Yet, if we trust all these VC-backed AI startups and assume that it will continue growing rapidly, e.g., at least linearly, over the next years, I'm afraid that it may indeed reach a superhuman _intelligence_ level (let's say p99 or maybe even p999 of the population) in most of the areas. And then why do you need this top of the notch smart-ass human biologist if you can as well buy a few racks of TPUs?

Love it. I do very occasional birdwatching, so I still don’t know most of the birds I meet. What I like about Bird ID is that when I see in binoculars a singing bird I can quickly identify it, check photos, and really confirm that it’s exactly that bird.

I’ve heard from more experienced birdwatchers that it can false identify in some cases, so I always try to confirm visually, but anyway, for my casual use it’s more than accurate enough.

The title and overall ‘take’ are very broad, it starts with

It’s well known that video games today are disposable pieces of slop.

But then it falls mostly into multiplayer games. For the latter, I will probably agree that old multiplayer games were more decentralized and self-sufficient just because distribution was also less centralized back then.

Yet, overall, I tend to disagree because of several reasons:

1. Video games market is vastly larger than 20-30 years ago. That’s why we see more crappy games, but there many-many good games as well

2. Back then there were bad games as well. YouTube is full of videos where gamers walkthrough some old games. And many of even popular titles are literally a broken piece of crappy tech demo with broken mechanics, soft locks, bugs, etc.

3. Outside of MMMO, F2P and multiplayer there numerous great games nowadays. Indie developers are very strong. Games like Buldur’s Gate 3 have a non-imaginable quality and amount of content for 2000s game industry. It’s a matter of personal choice, but I can name dozens of titles for the past 10 years or so, that are really great.

UPD: formatting

I was wondering how cartridges are designed and I think it’s a very simple and elegant design — just wire the standard microSD card. Otherwise, I love such cozy projects. Even if they are not that efficient, they solve the problem and bring joy into someone’s life, both author’s and a small user (in this case).

I can only second this. I have an old iPhone with a second sim-card, because I need it from time to time. And Apple introduced this auto-reboot a bit earlier, iirc last year. The problem is that after rebooting it also disconnects from wifi, so e.g. SMS/handoff synchronization stops working until you enter a passcode. This is very annoying because it was very convenient for me to receive calls/SMS to my main iPhone.

It’s a good and reasonable feature, especially if for some reason you are afraid of state or security agencies in a place where you live, or maybe during travel. It’s still questionable, because in some states you can indeed go to jail if you don’t unlock. Yet, I really want to be able to turn it off for use-cases like mine.

Parameterized queries.

Also happy to be wrong, but in Postges clients, parametrized queries are usually implemented via prepared statements, which do not work with DDL on the protocol level. This means that if you want to create a role or table which name is a user input, you have a bad time. At least I wasn’t able to find a way to escape DDL parameters with rust-postgres, for example.

And because this seems to be a protocol limitation, I guess the clients that do implement it, do it in some custom way on the client side.

I got mine at the end of 2021 and then used it till the mid-2023.

I know that it’s not a fair comparison, but I still compare it to macbooks because I’m a mac user for years.

Pros

- Linux support is amazing, basically you just install one of the popular distros and ‘it works’ (c). I used PopOS and was pretty happy. You also get all the Linux tools like eBPF out of the box, which is +1 compared to mac.

- Extensibility is a big deal. You can get 1 TB / 32 GB version for pennies compared to mac, where upgrades from the base are ridiculously expensive.

- Design and look is very neat.

- Keyboard is a classic one and also good.

Cons

- Battery life is really bad; same with cooling. At some point I started having more meetings at work and it gets extremely hot, noisy, and dies very quickly.

- Touchpad is just subpar to mac. Also chassis rigidity is meh. I know they improved the display cover design (switched to CNC), but I have the first revision.

- Display is 2K’ish. I don’t really understand, why they go with this resolution. Even their new display is around 2.5K. IMO, Linux works best either with 1080p/1K or 4K with x2 scaling (I prefer the latter) because fractional scaling is bad. I struggled a lot with external 4K monitor because it was nearly impossible to adjust all sizes so texts were good on both and especially when you disconnect and go portable. I know it’s Linux and you can DIY everything, but for me it was just too much of a headache.

I still fully support this company and wish them all the best, but since getting the MacBook Pro 14 with M2 (company’s, not personal) in the mid-2023 my Framework is waiting for two things: i) 4K display module; and ii) ARM main board. If they release these upgrades I will jump into Framework right away and give it another try.

So I recommend it if Pros are more important than Cons for you.

UPD: formatting and conclusion

I hear it all the time ‘coding interviews are useless’, ‘peer review in scientific journals is broken’ and so on and so on.

I’d say yes and no.

Yes, these are the problems that cannot be solved perfectly.

No, because in such areas any ‘reasonable’ filter is better than nothing. People say that these assignments don’t have anything with reality, but, well, we don’t have months to try to work with each other, we only have 3x1h.

I worked as individual contributor for years, but also had a chance to try a hiring manager role for the past 3 years a lot. We do standard leetcode-style interview (without hardcore) + system design. And I always consider both as a starter and bite to see how the candidate behaves; talks; do they ask questions to clarify something and how. And I always try to help if I see that candidate is stuck. By the end of all interviews you will have some signal, not a comprehensive personality profile. Do we do mistakes? I’m pretty sure, yes. But I think it just works statistically.

I use PiHole, it does break some stuff here and there, and sometimes useful things like Private Relay or iCloud in iOS; or once YouTube history stopped working for me (apparently they use a separate domain to track watched videos and progress!). It also depends on the block lists you upload. It’s pretty easy to unblock, especially web, as you just look on which domain cannot resolve in the browser dev tools and add it to the allow list.

Yet, DNS-based blockers have a limited usefulness at this moment as some major ad-providers started using the same primary domain for serving ads. For example, YouTube, partially Google, Yandex. I guess they cover everything with top level load-balancer and then route internally to specific service ingresses

That's where the separation of storage and compute kicks in, I guess. Startup process of our Postgres instance (compute node) is a bit different from vanilla Postgres. We need to go to the network storage service (pageserver and safekeepers) to get the last known commit LSN, but we don't need to perform any sort of recovery on the compute node side. That way, compute is mostly stateless.

Basically, to start we need to know this LSN and to bootstrap the Postgres processes. This is really that quick. After that compute is ready to accept connections and serve requests, as it's able to get any missing pages from pageserver with GetPage@LSN request.

We do have the whole bunch of problems to solve: queries latency after cold start; startup after the unexpected exit of the heavily loaded Postgres instance could be slower; etc.

c) Heard the cold start is a second (IIRC), how does that value differ if one runs Neon on bare metal instead of k8s?

Yeah, as Nikita mentioned it's 2 seconds now. We did some tests and measurements and on bare metal, it's sub 500 ms usually, so the remaining part is the k8s (+ our own control plane) orchestration overhead. For example, with plain Docker (which we use in CI in addition to k8s) it's around 1 second already.

K8s provides a convenient abstraction layer, though. So I think that we'll continue using it and optimization will come with pods pool / over-provisioning and it'll be realistic to bring the startup time closer to bare-metal.

-- Cloud engineer @ Neon

We do scale compute part down to zero after 5 mins of inactivity now (no active transactions). This 5 mins threshold is a random pick, it could be 1 min or 30 mins later, or even customizable by the end-user. Storage part is heavily multi-tenant, so it's always running and our main objective is to make resource utilization as effective as possible.

It still has a significant latency on the first connection attempt after suspend (1-2 seconds), but we are working on that and it seems to be realistic to put the startup time under 1 sec.

Pricing model is still work-in-progress, so cannot say much about it. Yet, my personal intention is to make it cost-effective for both end-user and us. I'd prefer to don't build a service with claims like 'here is your free-tier serverless Postgres with zero-latency on connect', which actually means that under the hood there is an always-running compute burning the investors money. Hope it's realistic to achieve :)

-- Cloud engineer @ Neon

I would rephrase it to 'Do not build websites that work 100% fine and fast only in Google Chrome'. Unfortunately, this is the case and most of users just choose the most convenient tool, while trading their privacy.

The main thing about building all these private cloud setups, that bothered me most of time is security. It is not a big deal to take from GitHub and run all these bricks of your infrastructure, but how to maintain? Everything should be updated regularly, otherwise you risk to get your data dumped and leaked by some automatic crawler or home-grown hacker, once new vulnerability is discovered in any part of your tech stack.

The only easy solution I see is to hide everything in the private network and make accessible only under VPN. However, it is not that useful, when you need to get some file or read/reply email from some new device not owned by you.

I dare to promote one StackOverflow question [1] about pandas I have tried to investigate and answer [2] half a year ago. And I was rather horrified by its internal complexity after digging into pandas source :)

OP was wondering, why pandas facing a strange overhead after each 100th iteration in some very specific case. There was a proposal about Python's GC, but it was not clear at all.

Finally, I have dived into pandas and found that it has a hard-coded constant == 100 (!) of a number of internal data storage blocks. After reaching this value it runs some consolidation routines [3], and they consume a lot of memory even leading to crash with memory error.

What was much more wondering, is that after changing this constant to some large value (1000000, actually it disables consolidation at all) reduces memory consumption dramatically! This consolidation seems to reduce storage and memory consumption, so I still do not know why the opposite happens and why it works well in all other cases.

[1] https://stackoverflow.com/questions/56690909/python-is-facin... [2] https://stackoverflow.com/a/56705419/978424 [3] https://github.com/pandas-dev/pandas/blob/761bceb77d44aa63b7...

16 GB of RAM and I use Safari mostly and Firefox from time to time with less than 20 tabs usually (less than 10 in average, I guess).

I check Data written in the Activity monitor — Disk.

"As for writes, at 1 gb a day - far more than my current rate of data use - it would take the same 114 years to reach 40 tb."

Maybe it works for author's very specific system or use-case, but on my personal MBP laptop with very occasional usage pattern--some days I do not use it at all during the week--I end up with 10 GB per day of writes on average. That way it will be 11.4 years already, not so many. And I do not do something very disk-expensive on it like torrents downloading or database testing, only some general development task, watching online videos, web surfing, docs, etc.