sottol

It's also devoid of minerals so you might want to supplement, if you mean to consume distilled water.

I think the Fusion posts are still available in the free version - afaik certain things like overriding the feed speed no longer work though.

And I only briefly looked into it, but most Fusion post-processors are just Javascript! (or at least based on it)

The bottom panel is hdpe currently and "trayed", but you wouldn't run this machine with flood coolant anyway. mql might work but the idea was to cut dry, suck away chips and rely on coated end mills.

I've seen people experiment with diesel heater pumps for mql, might try that some time.

coolant always ends up very messy. even mql lubricated chips can be a relative mess compared to dry cutting.

I designed a little machine that's smaller than the Carvera with a 300W spindle at a lower price-point but requiring DIY (my build is probably around $550-600 now), I kept it smaller for a couple of reasons but stiffness was one of them.

The Carvera has an excellent work-volume to machine size ratio. Afaict it's not the stiffest machine but quite decent. The spindle power looks about right for it's level of stiffness judging by the YT videos I've seen.

Good point - there is a provision for laser-cut steel or alu plates that sit inside pockets inside the enclosure panels as I figured many would be turned off by a full plywood design.

The metal parts are symmetric so you can cut 2x of each - that is way cheaper on send-cut-send (afair only 25% extra for 2nd part).

Documentation is definitely lacking - I just got the design to a point where I'm pretty happy with it. So far only 1 machine exists and the build has a few bits that are... less than ideal.

The plan was to build a second machine and optimize some of the assembly, take notes and document the build.

I uploaded a quick video of it cutting some alu recently if anyone's interested:

https://youtube.com/shorts/XUsj06iMbb0

I tried [1] - replaced the sides with 2040s, blind-jointed every extrusion, replaced X with MGN12H + 4 carriages, replaced the whole Z with 4080U and even bolted it into an MDF box to stiffen it up. The 3018 could cut alu, but not well. Maybe it was the 4080U but it just didn't work for me, it could cut alu but had lot of chatter.

For $600-800 it is fully enclosed, includes a decent spindle, wifi-enabled +offline RRF-based controller with folding/rotating LCD screen, 2 power supplies (24V + 48V for the spindle), a chip vacuum and probably some more I forget.

This machine was designed [3] specifically to cut alu rather well for < $1000, can run adaptive clearing toolpaths at 0.5mm optimal load and 3mm doc at 1800 mm/min with a 6mm end mill and produce decent chips. It can probably do more, my standard settings are 1200mm/min, 0.5 woc, 3mm doc. Mind you, this is all still hobby-level though you could still push the feeds and speeds a bunch I reckon.

[1] https://youtube.com/shorts/C0ngUJrWrB4

[2] https://youtube.com/shorts/XUsj06iMbb0

It is definitely more expensive than 3d printing in terms of consumables but china/aliexpress has really good hobby-quality equipment now - if you don't mind the country of origin. Many of their coated end-mills are decent and < $5 depending on diameter.

Maybe not stuff you would want to run professionally but it works quite well in a hobbyist setting and a mistake won't cost you $50.

I also looked into the Millenium Milo, only downside for me and kinda why I decided to design my own Voron V0-ish sized mini CNC machine was the really large footprint for about 3x the work volume (work _area_ is ~2x afair). An enclosed Milo would take up about the same volume as two 350mm Voron's side-by-side. So footprint of a small desk.

Imo besides the price, the other big factor is just how much less forgiving CNC machining is than 3d printing - so many mistakes you can make, zeroing the WCS, wrong WCS, mounting the work different than you had it in CAM, ... bam, at least the work is ruined. That's kind of another longer-term goal with my CNC machine, reducing some of these errors if possible with a web-based UI and maybe some computer vision. But that's far off, I'm currently playing with using a camera for work-probing/WCS-zeroing and it's sloooow progress :')

Shameless plug if anyone is interested - I'm working on a $600-ish open-source, reasonably capable, but small and somewhat "tidy" hobby CNC machine with BOM cost around $600 that requires some DIYing.

It's meant to be an alternative to the Desktop CNCs like Nomad, Carvera, Bantam, ... moreso than a PCNC or other proper entry-level CNC.

The ultimate goal is to make it hobbyist-friendly, capable of easily cutting alumin(i)um and not taking up a lot space, not being messy or loud enough to require a dedicated workshop. Unfortunately, cutting metal is inherently loud so you probably would not be able to run it in an apartment as I'd hoped.

I've made a couple decisions around being friendly for people coming from the 3DP space around probing, using roborock CPAP as chipvac, running it mostly dry, fully enclosed. I'm also starting to work on computer-vision-based probing and the idea is to later enable a host of more user-friendly and safety-focused features and maybe integration with Kiri:Moto's CNC mode for "guided" CAM and so on - basically a beginner-friendly CNC that guides newbies using an integrated web-interface.

More info on Github: https://github.com/thingsapart/mini_nc

GH is a little outdated but I've been using the little machine to cut alu for a while (mostly parts for itself) and it's working quite well. There's more videos and such on the Discord linked in the GH readme - feel free to ask questions on the Discord, I try to respond as quickly as I can. The full model with all its components is completely open in Onshape (I know it's not ideal but how I learned CAD - link also on GH).

Too little, too late?

I haven't rented in SF in a while but I imagine these would have done well 5-6 years ago. Afair, I've heard of the company before so that's probably how long they were caught up regulatory purgatory.

I was thinking of an objective measure of the impact of GenAI on society - labor productivity seemed like possibly one decent measure.

Gen AI and esp. ChatGPT has been mainstream for around a year and very popular for at least 6 months (reaching almost 2 billion visits in March '24, business leaders having rolled out GenAi-mandates), probably hundreds of billions USD invested, and reports of 2x-10x individual productivity gains, when would you expect to see an impact on some broad economic measures? Which ones?

Any better datasets? The general non-farm labor productivity is clearly not ideal as it encompasses so many jobs that would not be impacted, but manufacturing and lodging+dining (hospitality) seem to account for only 20% of US GDP. More sector-specific datasets are more out of date or more sporadically collected.

I saw as more of a "look, this can be done without impacting speed - Apple, just do it" kind of video. Which is why I thought they didn't speak to the how.

That seems to be often the pitfall with socialism-within-capitalism as often done in California imo - many of these socialist-ish policies only work if there's no easy "escape hatch" or it's at least uncommon. But not having access to private alternatives also often goes against free-market ideas.

Afaict, the wealthy and/or well-connected often "drive" the quality of public services - they're not going to insist on improvement if they can easily check out and pay for a private alternative.

I'm not saying to stop private alternatives, just that policies that might work other places probably don't work 1:1 in the US.

The claim was about _full_ driving being anything but, ie not _fully_ self-driving, not being completely fake. Disengaging every 10-110 miles is just not "full", it's partial.

And then the gp went into details in which specific situations fsd is especially problematic.

It just needs an "app" to run a JS-based X86 Emulator for the infinity mirror effect.

Not taking sides on the attack, but, wow, you ommited half the text - important in this context:

"unless either:

(a) they are placed on or in the close vicinity of a military objective; or

(b) measures are taken to protect civilians from their effects, for example, the posting of warning sentries, the issuing of warnings or the provision of fences."

> Many people still claim that "self driving cars don't exist", in so many words, even though they are deployed in multiple cities.

But just look at the predictions of that time - cities will change, ... and so on. Sure, we have self-driving cars but the reality looks very different (and a lot more like the past!) than the pundits and futurists imagined! I'm not sure anyone will make their billions of dollars investmented back within even 20 years.

Just two random examples from ~10 years ago (2013-2016), you can google many more of that time.

* "Ford Targets Fully Autonomous Vehicle for Ride Sharing in 2021; Invests in New Tech Companies, Doubles Silicon Valley Team" [1]

* "Disruptions: How Driverless Cars Could Reshape Cities" [2]

[1] https://media.ford.com/content/fordmedia/fna/us/en/news/2016...

[2] https://archive.nytimes.com/bits.blogs.nytimes.com/2013/07/0...

[3] https://www.gensler.com/dialogue/30/the-game-changer-for-cit...

> If trained properly, LLMs can develop advanced representations of logical systems that will surely outpace what humans can do in terms of raw reasoning.

We have already trained the LLMs on most of the human knowledge base (so like 4-5000 years?) - imo training data will become a problem and will soon be more expensive than compute. Sure, you can work around some of this using synthetic training data but I personally would not count on general-purpose LLMs (especially LLMs aka transformer models) developing super-human representations of logical systems anytime soon.

I think this is valid though. Transformer models don't explicitly do logic but implicitly "vibe" out the answer from the input sequence (using the attention mechanism) and learnt knowledge - they're predicting text sequences after all. So adding more irrelevant context to the input would quite likely influence the the output.

I could see attention possibly being able to overcome this, but if not that would be a pretty big gotcha for real-world applications and reliability in real-world scenarios where, as others have said, it's not immediately clear what is relevant info. These models would be a lot less useful if a human had to decide which information to feed them and the output would be dependent on human judgement. I understand it's where we're at right now and that they are quite useful already but the valuations hint at investors expecting more imo.

The general opinion here on HN seems to be that the most talented that can easily land a new job are the first to leave - might be a good time to skim the cream of the crop, even if for 12-18 months. Who knows, many might even stay when every other major company moves to RTO.

If AI were to plateau, OpenAI would be one of many providers without a clear edge, they'd lose market share. Companies might even start competing on price. Imo, it's not clear any of the software providers would really do that well in an "AI is commodity" scenario, HW companies might though.

I'm running Phi3, Llama 3.2 and Mistral Nemo locally and they're decent enough for many things.

Part of the earlier appeal of YC was that it was a batch of approval that, at least imo, was a strong signal to VCs to take a good look at the company and more-likely-than-not invest in.

With larger batches and more lax selection that signal is so weak that YC comapnies are mostly like other companies with some traction.

Or the police could do that but somehow they can't seem to.

By capturing off-shored manufacturing of Western companies, Chinese companies have acquired some decent "basic" manufacturing know-how/quality at unbeatable prices, often resulting in finished components that are lower-cost than locally sourced raw materials here.

The second-order effect is that other companies can then build on those cheap basic components and integrate into something higher-tech. Initially usually some sort of cheaper knock-off of some idea but constantly improving it while keeping cost cheaper than the Western competitors, reinvesting most of the proceeds into continued development. Eventually the product stands out in its own right and the manufacturer becomes world-class in its niche.

I've seen this a couple times - DJI or Bambu Labs to some degree are imo examples of this, Unitree seems to be another example.

Probably because their well-being is already being oh-so seriously considered in areas of online safety and communication/surveillance - you _have_ to choose your battles, you know. There's only so much lawmakers can do to "think of the children".

</s>

"Shocking" indeed?

Yea, it's mainly for fun/experimentation. I got a $15 Heltec board and a decent antenna and could reach some nodes in SF down to San Bruno from hills in the East Bay.

But I was surprised that there's only 5-10 active nodes in the North Bay and fewer people than that - Bay Area being what it is (was), I was hoping there'd be a a little underground community.

Most of the (federal) tax money goes to Social Securiy + Medicare/aid, servicing interest and funding the military, probably 60-70% afair.

It's hardly "pay[ing] their elite friends for things you’ll never see"?

Except, the money is generally used to fund the basic functioning of society and benefits all citizens to a large degree somewhat equally - yea some % is lost to inefficiencies but I doubt it's > 50%.

But also, most individuals ultimately making use of these tax havens' competitive services do not choose to restrict themselves or their businesses to said tax havens only. They tend to want to live and do business in the higher tax societies with a broad middleclass to hire from and sell to, educate their children and tend to their, say, medical needs.

I understand that one can just be(come) rich and "roll one's own" replacement for government services and eventually it's hard to see why you would even need gov't services in the first place but there will be sore losers. Otoh, I also understand that most voters would rather skimp a little on universal health insurance here, education and infra there to have a little more money at the end of the year to lease a brand-new SUV every 3 years.