HN user
Data developer and generative tinkerer.
https://benrutter.github.io/
Miss sized models plan out each requirement, and smaller models implement a tightly detailed task? Isn't this the standard approach?
It is, but the article is suggesting something different. The central idea is to use /prewalk to load slightly edited context from the more expensive model into the one, so that the cheaper model doesn't burn tokens reading all the files again.
I can't tell if you're being serious, but if so, this is a bit silly. It feels like saying Covid 19 was no big thing because the Black Death of the 1300s was much worse.
But even more so than that, your own wikipedia article seems to fly in the face of your point:
Estimated annual temperatures in 1540 were by far the highest between 1500 and 2000, while estimated precipitation was by far the lowest registered until 2025
The weather events we're seeing now are even more extreme that those of 1540, but the impact hasn't been as severe because our food and water supply chains are more resilient.
I loved this article! Regardless of how you feel about AI as an industry or tool, the economics of AI is fascinating. It's awesome to see something like this that gets into the business side a bit more.
I don't know if I agreed totally with the assessment of the risk Chinese labs pose to US labs though, in particular I think the main part I wasn't sure about was this:
I highly doubt that Chinese models are cheaper to serve on a marginal cost basis, they just seem cheaper because Anthropic and OpenAI are so supply constrained that they are charging far more than they would if there were sufficient supply to meet the demand for intelligence.
How true is this? My understanding from Deepseek's original paper was that they focused heavily on optimising training and inference costs, in particular so that they can operate on cheaper (and more accessible to China) hardware.
It's possible I'm just not in the loop, but nobody seems to talk about US models innovating in this way (I'm just talking about cost-to-serve/train, not saying US AI companies don't innovate in other ways).
It seems to me at least, like there's a fair bit of evidence that AI shifting to a price based commodity market (vs a "best-model takes all" type market) would put China at a significant advantage? And even more significantly, require a pretty hefty correction of company valuations in the US?
historically we didn’t have LLMs which don’t mind “unpleasant[ness]” and “heavy rework”.
I'm not quite as optimistic. "Unpleasantness" might sound like subjective disgust, but it often relates to objective criteria like complexity, that increases the surface area open to bugs in a real way. Similarly, if reworking has a non-0 chance of bugs, its gonna have some form of risk, and a major one will of course be higher.
LLMs may be capable of a lot, and they miggt change what projects people can take on, but they definitely don't write bug free code.
The bun rewrite/transpile did use some neat ideas like lots of testing and per-file translations, so hopefully they have more in their toolbox than "LLMs hopefully won't be affected by complexity or write bugs"
If you're looking for a non-ghostty recommendation (not that I have anything against ghostty) I use alacritty + zellij and it works fantastic. It doesn't seem to impact battery, I certainly get good battery usage on both my work and home laptops.
Staging Ground is available to two different types of users on Stack Overflow: > - Users who are asking their first questions - Users who want to review and offer guidance on newcomers’ posts
It looks a little different, like a "learn to use stack overflow correctly" type spot. I think what newbies want (I would have loved when I started out) is a "why is my code broken" type spot.
I remember some moderator having a go at me for answering someone's question (apparently the question wasn't high quality, and me helping them has encouraging "the wrong type of question" for the site).
AI is a part if SO's downfall, but I've also seen a big shift of asking for help to places like discord.
SO has always had this thing that it's a wiki, not a Q&A site for people who are stuck - I feel like people have always wanted the second though.
- tangled federates: https://blog.tangled.org/federation
People have been talking about federation across forges for a couple of years and seems like its finally at least close to being a real thing!? That's absolutely amazing!!
Of course!
I was thinking about the combination of:
- Human impact on the environment, which e-waste is a big part of, is causing damage to wildlife, and serious climate change.
- There's huge awareness of and push back against bad hardware practices (often around non-replacable batteries) so they must have at least considered this.
Yeah, that's really not great at all! From their website:
Wait, it's single use? Yes. We know this sounds a bit odd, but in this particular circumstance we believe it's the best solution to the given set of constraints
I don't want to be too harsh, since it seems like the pebble team are working hard at producing some exciting tech. But intentionally making a single use device is phenomenally irresponsible in today's climate.
I know they say they'll recycle them, but it'd be naive to expect anything other than a tonne of these becoming e-waste.
Tangent, but anyone who's a fan of Typer, I'd massively recommend Cyclopts! It's essentially the same design, but takes advantage of some of the type features python introduced post Typer being written, to make things a little more consise.
I like the idea of trying to make email more about human-to-human connection with less focus on automated-stuff-to-human.
The bit I don't understand, is what about where you need the automated-stuff-to-human, like, authenticating a new account? Would this just block those types of emails? Is the expectation if you used this, that you'd have seperate emails for contacting people vs accessing online services?
I get about 10 AI generated spam emails a day that get through my spam filter.
Yikes, that's sounds unworkable for a lot of people! Most people in here are techies and use email all the time, but some use cases (say, contractors who are mostly out in jobs) people might only sit down in front of emails properly once a week. 70 emails is a mammoth waste.of resources to have flooding your inbox in that kind of context.
Done isn’t when your work is over. Done is when value is created in the world.
I don't get what the context behind this is? What's the intended goal of gatekeeping the word "done"?
For me, I normally come up into "is it done?" as a question when working on tickets in something like jira/trello/etc and there's some "done" column to move things to. Is "value in the world" useful for that context?
Like, "Oh yeah, Craig's finished implementing the API endpoints as per the spec, but we're leaving it in active until the front end team integrate it and it starts delivering real world value".
So you have Android Automotive? (stupid name)
Yes I do! And yes it is!
It's a fairly recent Volvo, and it has a sim card (or something along those lines) that connects to the internet. It also can use a phones connection via bluetooth, so internet wise I think would be fine, and uses that connection to update the cars software.
I can't say how long the software will work for - I guess that depends on how long Volvo support their software for. I'm optimistic but I guess I'll find out in quite a few years time!
I haven't tried it (I thought my comment made that clear, but sorry if it didn't). I definitely did look at the page and watch some of the videos though. Maybe I'm missing something but to me:
- It describes itself early in the page as an "Agent multiplexer" (not a terminal multiplexer)
- It describes itself in the videos as "tmux for agents" (again, suggesting that it's designed for use specifically for agents, rather than as a general purpose multiplexer)
- The "compare" page of the website box is focused extremely heavily on integration with agents
Sorry if I'm coming across as snarky, but your reply seems to be making out that I didn't read any of the content and just made a drive by "this seems to be designed for integration with agents, rather than as a general purpose mutliplexer" comment without any research - but that's literally the entire sell of this project as far as I can see? I'm not arguing you wouldn't be able to use it as a general purpose multiplexer, but that clearly isn't their intention, and the web page and associated videos and content make that really clear.
Are you seeing anything on the website at all that does suggest they have the use case of "general purpose multiplexer" in mind?
I think you make some good points, but as a UK citizen, there probably some extra context 'd want people to know.
Firstly, the Farage/Binface byelection is a result of Farage being investigated for breaking laws designed to stop political corruption. He's called an early local election to "let the people decide". Even if voted in, if he's found accountable for those charges, he'll have to call another election. So all major parties are staying out of this one, but would likely stand in a second. Essentially what I'm saying is, this isn't a normal local election by any means so I wouldn't use it to talk generally about political parties offering to the UK.
I would also urge against using HS2/electricity-prices/Hinkley to make a " therefore the UK is in dire trou me". There all quite different: Hinkely and HS2 are certainly managed and unpopular, but they're individual projects - a lot of the UK issues are caused by much more complex, larger economic problems than government projects being mismanaged.
And then on the energy price, the UK has really expensive energy. One reason for that is because there's a huge amount of invesent in renewables right now, and the investment is paid for by the consumer. I'm not arguing this is necessarily good, but it's a political choice to prioritise more expensive electricity today, with the goal of cheaper / more secure energy tomorrow. It's a lot more complex than just "the UK is cooked because it has expensive energy".
Sorry for the minor essay! I think the UK's economic situation is serious, but also very interesting (it's possibly the most extreme global example of financialisation backfiring), complex and worth understanding!
From what I can figure it is actively calling itself a multiplexer, so pretty closely linked to something like tmux (one of the review videos on the page calls it a "tmux rewrite").
But they seem to be targeting very different scopes use wise.
I think if you use this to do stuff like, edit files, run htop, drop into an interactive shell, compile some code, my guess is you're gonna be fighting uphill and have a bad time.
On the other hand, if you're only using tmux so that you can have a bunch of terminal windows for agents. Then this is probably going to make your workflow a fair bit smoother.
No, I meant literally the car has google maps built in - it's an up to date android app.
There's so many comments on the lines of "I would never buy a car without android auto/car play support". I 100% used to be on this camp, "A car without android auto, what am I, a caveman!?".
Anyway, separately to anything car related, I got fed up of some of google's treatment of android and switched to Murena's /e/os which is a de-googled spin of android. It doesn't support android auto, and that held me back from switching for ages.
Long story short, I don't miss android auto at all. Bluetooth is absolutely fine for playing media from my phone etc.
(my car does already have functionality like google maps built in though, so I might feel differently if it didn't)
For images, there are some clear styles AI leans heavily on if not actively steered away[0].
It can definitely be prompted pretty successfully though, a bird spotting app was up her on HN recently with some really nice looking woodblock prints that were AI generated (I always feel disappointed/tricked when art turns out to be made by AI, I'm not sure why, it seems to pull the joy out of it for me)
the front page is full of LLM smells
I'm curious, What kind of details are you thinking of? I'm not sure I really have much of a radar for LLM websites in the way I do for LLM pictures or music.
That might be true for social media, but there are 100s of American brands that make a large point of being thought of as an "American company" or "American made" goods.
Wtf. I've always wanted this, and it was right there.
Haha, same! I was literally looking for this feature last week without realising.
What could be more git- the problem is rarely that it can't do something, but the ergonomics of discovering how to make it work correctly.
Another great, if not so relevant quote from JMK:
My only regret in life, is that I did not drink more champagne
That's an awesome idea and one that's already being worked on![0]
I really hope it works out, would bring so many benefits.
there are disadvantages: ergonomics, space, cost, environment.
I read a lot, was really into ebooks and now mostly buy paper books. The inverse of the cost point iis a big reason, cost of ebooks is much more than paper books in a lot of cases because there is no second hand ebook market.
Environmentally I think it's complicated- an ebook is certainly better, but an ebook reader itself is much worse for the environment than many books. I can't claim any moral high ground since I have both.
I don't agree with you, but I think the interesting thing is that we'll both get to find out.
In two or so years time, we can find out if heavily AI produced projects become more maintainable, ship more features and dominate the open source landscape. Or if human written and maintained code has a long term advantage. (Or more likely somewhere in between)
Either way, whatever anyone claims, none of us know, but we'll find out son enough.
I'm almost certainly missing something, but I'd love to be able to call await within a sync function. If I'm waiting on the async function, it feels like there's no need for it to "colour" my caller function like it does.