harshitaneja

We don't know that. We don't know if this particular kind of tool can do "useful" things without also developing the ability to write a sonnet. And they are absolutely a frontier. I am not suggesting we should continue building them just because they are, there are many technologies which could have been built had we thrown the amount of resources we have here and there is a case to be made for not doing it at the pace we are or if at all. But we can do that without diminishing what exists.

I don't have the same rosy memory of the history as you do. I remember very regularly having to wrestle with flash or java applets reinstalling and doing all kinds of voodoo I could to get sites to run, sites I used fine the day prior. Figuring it all out on my oh so slow connection and hoping no one calls the landline lest it stops all the progress. Not saying it wasn't a fun period of chaos and I everyday wish for internet to become the wild west it once was. It was acceptable in the sense that I didn't know it could be better, I was used to OS hanging and crashing regularly on my oh-so-fancy dual core system as well regularly so it was all acceptable.

Today I find myself having less patience for it all, because things just tend to work a lot more often than they don't.

With AI, I find so many small businesses and others I deal with which have their own websites have started to have better design, better performance and while I don't know if they are really accessible but they tend to have more prevalence of aria labels.

Not to take away from your experiences, just sharing my anecdotes.

Maybe we come from different cultures and context is harder to grasp just in text so maybe for those reasons your response feels ruder than I hope it was intended to be.

I am sorry for not being clear in my response but I didn't intend to twist your words. I am not sure where I did so. My response was intended to be a more general remark on the kind of discourse on this topic I see and that I think both sides are right from the context they are looking in with and also why I think both sides come out of this discussion exhausted of the other. Not discounting presence of bad actors but generally I think there are most engaging in good faith like you are probably.

Coming specifically to respond your last response, I don't think one needs all of these prerequisites to get value out of LLMs. In fact LLMs have helped me untangle some very messy ball of muds on projects where we previously deemed it not worth the effort and basically carried some codebases as legacy. Now we can write enough tests to feel confidence and do a port against those tests all in a span of few days, which we found impressive.

Now having said all this, I think I understand your perspective a bit better on your original comment.

While it's a very versatile hammer, if it doesn't work for your use case that's all great. I just think that a bit more patience though with honing it maybe could help you find areas where it could work for you. If not, cheers!

I think it's because both sides are talking about different things. If you go in expecting it is good enough to make developers obsolete today(reasonable impression to get from the way a lot of people hype it) you would be disappointed and after first couple of tries every few months you would probably not try it much with next generations. Reasonable if it's considered a dichotomy.

But a lot of people excited about new generations(including me, now) are not seeing it as a dichotomy but rather a spectrum where models are getting better and indeed once a year or even 6 months at times there comes a sudden growth which feels like an inflection point from what came before. Practically, it's a tool like any other, you evaluate it based on if it's worth the effort and cost for the benefit you get from it and if it is and has a good DX you use it. If the calculation doesn't work for you, it doesn't. For me, it has gone from a novelty, to good for some kind of quick manual search, to I guess it can debug some kind of errors at times in very specific conditions, to hey I think I am getting a bit addicted to autocomplete in IDE provided by them even if I don't use them for anything intelligent but it's becoming indispensable now but only this part, to it's good for areas I lack expertise in, to agentic sucks I will stick with discussing algorithms and architecture with it on greenfield projects, to holy shit it can do agentic decently well now, I am skeptic to give it access more than in limited cases, to now I am getting close to letting it run free on my device in not so distant future I guess. Some of these were big jumps, at each point I was skeptical of growth. Everytime I thought now the growth will slow down from days 2k context window to millions now. From basic chat completion to working on complex adaptive systems, game theoretic modelling, heurestics and constraint modelling and other things I throw at it. I am still needed in the loop, it can be so smart at times and then will do something so stupid, but the frequency of stupidity is rapidly decreasing. I am still needed, I don't think it could accomplish alone all that it has done for me. But I do at times at night remain awake reflecting on my self worth for the potential day when I don't add that value. When I have a harder time keeping up.

Also had someone told me not in even 2019 that in 2026 we could have NLP models do what they do today, I would have posited it all as sci-fi and here I am waking up in awe of the world we live in and how quickly we adapt.

Idempotence is a semantically overloaded term in computer science where in functional programming it refers to the same concept as mathematical idempotence it refers to any function leading to the same state in multiple calls as the first.

And yes, in real machines we can't ever have true same states between multiple calls as system time, heat and other effects will differ but we define the state over the abstracted system model of whatever we are modelling and we define idempotency as the same state over multiple calls in that system.

Just adding for context that I use Gemini Ultra and across all models from Gemini 3.1 Pro to Claude Opus 4.6, I have never hit 429s as well as hitting model quota limits is incredibly rare and only happens if I am trying to run 3 projects at once. While not the biggest agentic coding fan, I have been toying with them and have been running it for at least 7-8 hours a day if not longer.

The amount of goal post shifting is so amusing to see. Yes, sure this was probably not an "important" or a particularly "challenging" problem which had been open for a while. Sure, maybe it remained open because it didn't get enough eyeballs from the right people to care about spending time on it. Yes, there is too much overhyping and we are all tired of it somewhat. I still think if someone 10 years ago told me we would get "AI" to a stage where it can solve olympiad level problems and getting gold medals in IMO on top of doing so with input not in a very structured input but rather our complex, messy human natural language and being able to do so while interpreting, to various degrees of meaning what interpreting means, image and video data and doing so in almost real time I would have called you nuts and this thing in such a duration sci-fi. So some part of me feels crazy how quickly we have normalized to this new reality.

A reality where we are talking about if the problem solved by the automated model using formal verification was too easy.

Don't get me wrong, I am not saying any of this means we get AGI or something or even if we continue to see improvements. We can still appreciate things. It doesn't need to be a binary. What a time to be alive!

I am working on something similar but with an AST for legal documents. So far, it seems promising but still rudimentary.

My memory of the recent history of the web isn't as cynical. I am not denying the perverse incentives involved but at least for Facebook & Instagram(maybe) we had the Graph API which was accessible and provided a decent amount of functionality and access but was curtailed post public backlash from misuse like with Cambridge Analytica. Similar for twitter where it provided decent APIs which wouldn't exhaust easily for end consumers and even the year(or two) before the Musk purchase they even provided academia with API access equivalent to their enterprise offerings for free. But there too bots and all were at least used as the public justification for curtailing many features.

Now none of this is meant to excuse the behaviour of all these large platforms for all the terrible practices they engage in. But at the same time, we never figured out how to safely deal with the power exposed by these APIs.

x86 Thinkpads is what I am gravitating towards as well.

Regarding system76 I have heard really good things about their workstations but not about laptops. Have you used them? I recommend PopOS to anyone getting started with linux as the first distro though.

I like Framework devices but yeah sadly haven't used one with battery life that works for me. I can get a bit more than 5hrs on macbook under Asahi fedora with my usual usage.

Framework is so close to having it all though.

Will try to procure one for testing. It's a bit on the larger side for me though. How much battery life do you get on your average daily workload and what kind of workload is it? If you feel comfortable sharing.

Agreed about macs feeling bloated. I will not get overdramatic by calling latency in many functions unbearable but it is certainly quite high.

I use chromium based browser and applications so don't get to enjoy the benefit of the power efficiency gains of safari sadly.

Geat to know this about your experience with thinkpads. Due to your familiarity with multiple such devices would you be able to recommend a starting point for someone who wants something light with budget not being a constraint?

It has been a while since I daily drove one but my old laptop used to have an nvidia hybrid setup and it was possible to get power management to work decently with it but that might have been me being lucky with the configuration. Thanks for the headsup.

I have tried multiple framework devices as a partner firm uses them. Good devices, want to support what they are doing but yeah battery life is really lacking for my use case.

Yes, it's just open weights however they do allow commercial use but with stipulations to do no harm, follow the law and not circumvent the guardrails in place. I would worry more about the revokable nature of it more than anything for commercial use.

That seems to be the conclusion I have been avoiding to reach. With graviton and other arm based linux server machines being a good bulk of my work I hoped I wouldn't have to worry about multi architecture docker builds. Ah well.

Any suggestions for something well built but lightweight and that one could figure out how to get 8+ hours of actual daily usage battery life on?

I have been on an M1 macbook pro since launch and while I love the hardware, easily my favourite device I have ever owned but MacOS has just always been the thing to be the faustian bargain coming from being a linux person. I spend a lot of time SSHed into more GPU capable linux machines for most of my work and thus get an escape but after driving a friend's linux machine I started looking for a way to daily drive a linux machine. I tried Asahi Linux and also tried to find some non apple machines including with Snapdragon X Elite ones but so far I haven't found anything with good battery life and a decent linux driver support. So far Asahi linux with the reduced battery life seems to be the best bet. I don't mind tinkering. I love tinkering. I am not looking for "just works" but something which I could get to work after putting in the hours. If someone has suggestions please share. Edit: Sorry to go somewhat off topic.

I spent years fussing about getting all of my APIs to fit the definition of REST and to do HATEAOS properly. I spent way too much time trying to conform everything as an action on a resource. Now, don't get me wrong. It is quite helpful to try to model things at stateless resources with a limited set of actions on them and to think about idempotency for specific actions in ways I don't think we did it properly in the SOAP days(at least I didn't). And in many cases it led to less brittle interfaces which were easier to reason about.

I still like REST and try to use it as much as I can when developing interfaces but I am not beholden to it. There are many cases which are not resources or are not stateless and sure you can find some obtuse way to make them be resources but that at times either leads to bad abstractions that don't convey the vocabulary of the underlying system and thus over time creates this rift in context between the interface and the underlying logic or we expose underlying implementation details as they could be easier to model as resources.

RPi is a good option if one has less RAM requirements especially if you take into account the quality of the drivers and software support in general.

RPi can be a compelling option if you need lower power draw. It does take some effort to squeeze out power efficiency but if the requirement can't be handled by a microcontroller then it is the most convenient of-the-shelf option.

For everything, RPi isn't a very compelling option. Even for GPIO, during RPi shortage I started experimenting with just using STM32 dev boards connected via USB to a NUC or an old PC and it worked well. But I just prefer to use ESP8266 or ESP32 for those tasks most of the time. Bandwidth and latency of USB communication/wifi to the main device has been low enough for it not to be a concern for me and I recon outside of very specific robotics cases it won't be for most.

CSI port is quite nice though and not many great alternatives.

After hoc, therefore something else hoc...

I am curious as to how are you envisioning such a platform be funded and operated? Is it government(which country? or a coalition?) funded and operated by an "independent" board. Or something more akin to Wikimedia Foundation where the public from around the world fund the endowment and that combined with volunteers help run the platform?

Everyone should obviously be able to make decisions for their own content on their own websites/blog and more power to you for exercising your agency and choice here. I browsed through your blog and with your content on vintage computing and history, I can consume the content without having to care about a date.

However, I don't find that to be the case for most of the written content I consume and knowing where some content was created temporally is important context more often than not(at least in my experience). Now, I can't nor would I want to dictate what others do on their own hosted content(where they generously share their hard meticulous work).

Comments on HN and other places lamenting lack of dates however should be fair game to desire such additional context. But then again, you commenting back is just participating in the same dialogue so I guess my entire message has been just meandering.

I help hire for a few clients as well as for my own small organization. We are already seeing impact of these tools on our hiring. For the same responsibilities and tasks we are already requiring lesser resources. For clients with less complex problems we are able to manage similar work with 60% of the resources planned. And that's when most of our work is mathematical modelling, heuristics, constraint programming and such. However, I don't foresee at least for the next few years we would ever get to a scenario where we don't hire developers. Given that most hiring has shifted to only senior developers.

I think you are referring to Pranav Mistry's Sixth Sense(https://www.pranavmistry.com/archived/projects/sixthsense/). It was part of his research at MIT media labs. I don't think he took it any further. Maybe got incorporated in some ways at Samsung where he headed research.

Assistant was surprisingly useful. In a couple of occasions I had it find me research papers I didn't know existed for a problem I had been working on for long. However the realization for me was that I shouldn't neglect search abilities of the LLMs I was already paying for. I tried to replace my Claude and ChatGPT subscriptions with Assistant but the usage limits were too low for me. And for the cases where I wanted to use long context of Gemini models, the assistant had an arbitrary limit of only 32k I think.

However for those where it can work, I think Kagi Ultimate is a great deal.

Agreed. It is a major annoyance and the primary reason I was exploring other options. However, even though it is infuriating when google refuses to not respect my query, the number of such queries for me are limited and thus I can use Kagi(and lately exploring search in ChatGPT). The amount of such searches I have needed haven't exceeded 100 on Kagi so I haven't had the need to resubscribe yet but if they do I guess I will maintain the subscription. However with the response time issue I doubt I would ever want it to be my primary search provider.

My experience with Kagi was not as positive as everyone else's here. I didn't find the search results to be better and perhaps that's because I am used to google foo to extract decent results there. So I made Kagi my default engine everywhere and used it exclusively for more than a month before giving up. The response time for search results isn't too long but that difference from google's response time, which I had come to rely on subconsciously for all my queries through a day, was too jarring and even after a month I couldn't get used to it. Having had an adblocker and Youtube Premium I don't really ever see any advertisements anywhere anyway so I couldn't find the value there too.

I would love to pay for search again and not be the product but as of my last experiment(Nov 2024) Kagi wasn't that for me. Curious to know if anyone else had such an experience or perhaps something I need to re-evaluate.

Well that's why we use CSS variables and any change would apply everywhere and also give us semantic information with naming like color-primary or color-surface.

Genuinely curious, how did Microsoft "shoved it down everybody's throats"? And weren't Jetbrains, Eclipse, Vim, Emacs dominant enough(especially Jetbrains) to have done so before Microsoft?