Yeah, I've had (and seen) interns and new-hires do fantastically during COVID.
You get what you pay for / put the work in for. If you're just hiring them, saying "read these docs and ship some PRs", and ignoring them, it's not too hard to predict what'll happen.
Engage at least as much as you would in person (more likely more, because you don't have passive hints about struggles), and it works out fine.
I've done that with mine. Worked great, and now I get around 30 hours of battery life with a lean linux distro, as long as I'm only like reading websites or writing on it.
>Which windows program are you looking for, specifically?
All of them, specifically.
I don't want to think about which windows program can or can't run with Wine.
This includes:
* Microsoft software, from MSTeams to Windows itself
* Audio production software (DAWs and VST plug-ins)
* Games
* Device-specific software (like drivers/software for portable thermal printers)
* CAD (nTop, only supports Windows, for example, and don't tell me I don't need it; same for many Autodesk products. NX and Rhino don't have Linux support)
The last one is the most fun, as I'm a CAD developer who worked on nTop in particular.
Aside from Microsoft Office, the rest is workstation stuff, and Microsoft Office is pushing "web first" (at least if their pricing is to be believed, the lowest O365 subscriptions do not offer access to the native apps).
We'll have to see how the AI softwarepocalypse goes. If I only need 10% of the features of Photoshop, I really don't need to be spending money on the full software suite.
>This attack occurs when any untrusted data source (e.g., from an imported sheet or ChatGPT connector) manipulates ChatGPT to run an attacker-controlled external script, which executes leveraging permissions the user has granted to the ChatGPT for Google Sheets extension.
So... does this imply "requires permission to run scripts without approval"? Or is that something that it can always do?
>Note: ChatGPT for Google Sheets has a setting called ‘Apply edits automatically’ that determines when human approvals are required before an agentic action completes. However, this attack succeeds even when the user has explicitly disabled automatic edits.
Yeah, that makes sense, it's not editing the sheet. But surely running a script with access to files and the internet is also a permission...?
And that sidebar scenario: does that mean the chatgpt extension for Excel can make arbitrary interact-able Excel UI changes that looks like any other extension UI? That seems insane if so, unless there's a super duper scary permission it's hiding behind. And it's still insane after that.
I mean, this is all par for the course for "AI" "security", but what
Libraries purchase books. One for every book they loan out at a time, digitally or physically.
The most recent noteworthy counter-example is archive.org breaching their "one purchase = one concurrent loan" limit during COVID, and they lost that court battle.
If you're equating libraries to LLMs, then every leading-model company would have purchased ~every book, newspaper, movie, and song in existence at least once. They have not.
Because libraries predate copyright and publishers and all the industry behind it. If libraries were to be invented nowadays, they won't let them purchase a single physical book to be enjoyed by several different physical persons over the course of time. What the publishing industry would like to have is 1 physical person = 1 or more physical copies, not the other way round.
So how do you explain LLM companies paying roughly none of it? If you're saying the protection is stronger now... how have they not been sued out of existence?
Honestly at this point a library might be an easy sell, in some ways. Copyright holders would be getting something rather than nothing (or the nearly-nothing they get from streaming), they might leap at that.
This can't be overstated! Libraries would never be allowed to come into existence today. I think we should all think long and hard about the society we have collectively created. It is not too late to make an effort to fight to reclaim the rights and norms we've ceded...
I have some hope that this'll all lead to a revival of semantic web / microformats / etc. Why write an API when you can just add some markup to your existing API, which already looks like stuff that it was trained on, and won't fall out of sync (because you use it too)?
> I have some hope that this'll all lead to a revival of semantic web / microformats / etc
Why would it? Do you see any agents or models use that? No, instead vibe coders at Anthropic vibe-designed a bespoke protocol that sidesteps and ignores the last 60 years of API development and integrations.
SW is about data interchange and standardizing semantic representations… it relies on formal, structured taxonomies of All The Nouns and then All The Verbs For Those Nouns.
MCP is structured at the protocol level, but its whole value is the bridge from structured to unstructured: natural language agent intents to specific APIs, domain-specific data to fuzzy natural language.
Semantic Web only works if someone has clearly defined the domain you want to operate in. MCP can work without that kind of definition (an API or prompt needs to exist, but it doesn’t need to be modeled in any way similar to RDF)
> What are you talking about?!
For instance, MCP-to-SPARQL bridge[2] that lets agents query SW data.
That was exactly my thought when I saw MCP: like we know so much about creating protocols but get a bunch of people together with no experience and that’s what you get.
Reminds me a lot of Microsoft’s WS-disaster of the early 2000s except the latter was thought through a little better.
To be fair a while back I did design an API for a general purpose model trainer which was absolutely atrocious for a few reasons, my own ignorance was a factor but the problem of accommodating everything from “model that can be trained in 30 seconds on a small machine” to “model that takes 30 days of training on a cluster” problematized it.
It would have made so much more sense to build a standard for documenting ordinary API endpoints and CLIs.
I definitely see it going that way from a marketing perspective if you want what you send/produce to be machine readable and actually used in intermediated surfaces like email and web search.
I loved having one of these until the non-stick coating inside started flaking off badly after a couple years.
Nowadays I'm only touching steel and/or glass on these things. An all-steel-interior Zwilling kettle has lasted me twice as long already, and looks brand new (on the inside) if I give it a brief acid cleanse to remove deposits. It's not quite as convenient or as stable temperature, but I'm more than happy to have something that lasts many times longer.
What brand? I couldn't find any last time I looked, and I do generally prefer the dispenser style (slower to hit a temp, but rock solid once there, and basically always ready to go. surprisingly useful / habit-altering).
As much as I truly love replaceable cords, I'd be more than happy enough if I could just unscrew the shell of the device, unscrew the power wires from where they attach, and put in any cord. Where I care about waterproofing, some silicone caulk is an extremely cheap (and removable) fix.
Unfortunately practically everything you can buy now is welded plastic halves, or easily-broken clips. So even when you might be able to do this internally, you have a high chance of breaking the shell even if you know what you're doing.
Right: There's a spectrum of choice here, it's not as binary as "you must throw away the whole device if the cable breaks" versus "conveniently detachable with two fingertips or a light breeze."
You mentioned waterproofing, and I think another major factor is how often the device (and indirectly the cable) is going to be moved around, and which stresses and strains will be placed on the connection. What's best for my printer is not what's best for an immersion-blender.
>To be perfectly upfront with you, this post will be glossing over many Meshtastic and MeshCore features, because I feel they are both non-serious solutions compared to Reticulum for reasons I will explain later on in this post.
Yeah, that's the general feel I get every time I poke into Mesh*. Neat radio tech, fun toy to find other nearby nerds, instantly-obvious problems that are fatal to growing beyond being that toy (or small specialized personal nets, where it's totally fine). They feel more like a tech demo than anything actually intended to survive.
Which is fine, you kinda need that to start out, and they do work today. Just... hard to get excited about.
Everyone you meet on a mesh is a real breathing nerd, who due to proximity has a lot in common with you. They are not trying to influence you or sell you anything
> Everyone you meet on a mesh is a real breathing nerd, who due to proximity has a lot in common with you. They are not trying to influence you or sell you anything
I wish... the Hamburg Meshcore mesh has some dumbass spammer spamming far-right youtube videos in the public channel for example. And from what I hear, Meshtastic also has issues with this kind of idiots.
Start spamming far-left YouTube videos to the public channel at the same time, according to the general theory of nutball political physics the two should cancel each other out
In MeshCore a concept of regions has been recently introduced - you can scope channels and messages to a specific geographic region (which is set on a repeater) and it will not propagate outside of the given region.
That way local channels don't need to flood the whole mesh, same as with trying to send a message to someone or reach the management interface of a repeater you know is in a given area.
The problem is, all repeater administrators in a region need to cooperate - and a lot of repeaters are abandoned. And it doesn't help at all if the troll is aware of regions.
In practice at this quite early stage of MeshCore development I would say that repeater need some regular maintenance (like the recent firmware update needed 1b -> 2b) so if they are abandoned, they will eventually fall out of the network.
As for trolls and regions - based on how you setup the regions a troll might need to physically travel to cause wider issues, as their traffic might no longer propagate through the whole mesh.
I run a few Meshcore nodes in Toronto - mostly as a nerd hobby. In some ways it has the feel of the 1990s internet and in some ways it's the same feel as ham radio . It has smart nerds, but also some unhinged people who are desperate to force people to hear them. Then there are the trolls...
Boating seems like a pretty good fit here, yeah - adhoc and semi local, and everyone already relies on radio for stuff. And no issues with line of sight.
Is there no longer-range p2p digital tech here though? It tooks to me like this would always be island-local, if even that, and probably useless while further out. Though for inland stuff like smaller lakes that's not a problem.
The traditional longer-range P2P is HF radio and maybe Winlink email on that. APRS over HF is also possible, and Reticulum would be too if hams were allowed to use encryption.
On the passage from Panama to Polynesia we had daily HF radio contact with Hawaii and California. Sometimes radio weather was pretty bad, but generally it worked out.
In practice nowadays passage communications for 99% of boats is WhatsApp over Starlink, with maybe Garmin InReach as backup.
The inherent limitations of free spectrum mesh technologies will never lend itself to a replacement for the Internet so will always largely exist in niches. Niches like personal nets, local nerd networks, or emergency response (tho actual first responders are not the most eager to try this stuff based on my experiences in the community.) All of this can be a feature or a bug depending on whom you ask.
On the mesh in Toronto with meshcore we have regular communication that reach all the way to Buffalo. We are past the « toy » stage, it’s truly impressive.
I think meshes do work extremely well in practice, and are quite resilient with regards to errors, and load balancing, and they get better as you add more nodes.
I think it's perfectly feasible for a small neighborhood of regular people to have internet shared over a wireless mesh network, yielding experience comparable to standard approaches.
They don't. Resilience isn't additive. The more nodes you add, the more announcements and traffic you add, which congests the network further. Regular internet topologies work because of high throughput backbones.
my impression was that there are algorithms that work well (dont remember the name sorry), and they don't really keep track of the entire network per node, each node only has a heuristic idea of how to route packages both to local and faraway neighbors.
This is very much like pathfinding in a video game - you know how to get to a next the next grid square, and also your region you are in knows how to get to its neighboring ones - recursively to the top, but said info is distributed heuristically through the network.
Also, meshes need not be composed of constantly moving and changing nodes. An example of a neighborhood of houses or radio towers, where each node is semi-reliable and doesn't really move is an absolutely valid and real world use case.
> my impression was that there are algorithms that work well (dont remember the name sorry), and they don't really keep track of the entire network per node, each node only has a heuristic idea of how to route packages both to local and faraway neighbors.
That doesn't change the fact that announces and extra nodes can still degrade performance. For one, it's hard to know whether a new link is better or worse than other existing links. Link conditions can also change which is very common in the kind of mesh networks folks are putting together using LoRA on antennas.
> Also, meshes need not be composed of constantly moving and changing nodes. An example of a neighborhood of houses or radio towers, where each node is semi-reliable and doesn't really move is an absolutely valid and real world use case.
Yes these are called repeater nodes in Meshcore and are already well known. Repeaters still don't have the same QoS that fiber backbones of the usual Internet have. The reason the Internet works well is because your packets are routed from your house eventually to a backbone network that helps the packet get close to its destination, and the backbone networks themselves are very top-down statically routed affairs with high capacity and QoS.
Once all of your repeaters are connected via high QoS fiber links you end up with the Internet topology we already have today.
It's pretty hard to imagine anyone "getting serious" with these tiny radios designed for reading gas meters and weather stations and using that to build some kind of off-grid alternate Internet.
I'd almost have more faith on dragging out all the old acoustic coupler modems and building a city-wide string-and-tin-can telephone network.
Of course they are not useless. My hiking/camping friends put together a fun orienteering game which used Meshtastic to do live GPS tracking. Worked great for that. But a country spanning meshternet it is not.
I don't think it needs to be though. There are plenty of things to explore using these things at the local 1-10 km scale.
To be fair it is already a miracle that there is enough Metastatic in my area that I can (sometimes/rarely) send a message between my home and work!
*disclaimer that I am coming off a recent disillusionment with Meshtastic. I thought it would be fun to make a single raspberry pi image with all the dev tools on it to do some off-grid dev/maintenance work if you were to treat this seriously and pretend the main internet is down. That moment came when I tried to compile something and the Pi ran out of memory. Really? I need more than 1GB RAM to compile something used to send short text messages?! That's nuts!
Yea, I abandoned it before then too - it was clearly over-filled with old answers with no context as to if they're still valid.
Everyone has had the experience where the top response was right for 3 months, but is now impossible to fix, so you have to just somehow know that you need the 19th answer with only 6 upvotes (versus 387). That was a problem early on, and afaict has exclusively worsened over its whole lifetime. They seem to be proud of it as if it was a major feature, and have afaict never built anything at all to try to address it.
(Yes, editing answers is "a solution", but you can't do that as a newcomer who knows better. You've got to commit to the ecosystem for months... when it had just failed you. Of course most visitors don't do that!)
Bitcoin is pretty much explicitly designed to use as much electricity as the market will allow, without becoming any more useful. If you removed 99% of the miners from the current system, Bitcoin will still be exactly the same - it won't be any faster or slower, and the same number of transactions will flow through. The cost of electricity serves only as a lower bound on the expected value of a coin.
Neural nets on the other hand generally show more capability as you add more compute power. There's a point where it's less valuable than the cost increase, so people don't do more than that, but it isn't constant value like Bitcoin.
It wouldn’t be exactly the same, because if you had all that mining capacity and 99% magically took a holiday, there is now enough mining power to take over the network anytime. It’s not secure.
Same with AI. Now that the Mythos and other models are finding exploits in every code base and anyone can run them, you can’t afford anymore not to keep burning credits securing your code base. It’s like proof of work red queen theory. You have to run faster and faster just to stay in place. Great business model.
You get what you pay for / put the work in for. If you're just hiring them, saying "read these docs and ship some PRs", and ignoring them, it's not too hard to predict what'll happen.
Engage at least as much as you would in person (more likely more, because you don't have passive hints about struggles), and it works out fine.
reply