Wouldn't that be a sign that those models were trained to optimise that test case specifically, as in, by exhibiting equal or higher performance while using less power?
At this point, even if the labs wouldn’t actively pelican-max their models, isn’t the training corpus full of pelicans already? Funny to think how one models choice of adding a red scarf in one of Simon’s earlier experiments might have led to an archetype of pelicans riding bicycles…
Not just pelicans, but pelicans with commentary about how good they are. simonw accidentally created the world's least efficient hill climbing algorithm.
I bet if he started critiquing pelicans with red scarves, they'd go away.
How's that different? You are still shipping a web browser and rendering through the inefficiency of the DOM event/rendering model. Sure it helps that the business logic is written in a native language, but modern, JIT-ed JS isn't a sloth either, and I don't think that's what people have gripes with and what causes abnormally large runtimes.
> Essentially. you do not ship a browser in Tauri.
I mean, essentially you are? It just so happens that this web browser that your application very much runs on top of may link (to a more or less large extent) with the host OS "WebView"/built-in browser (assuming there's one to begin with). My point still stands: if you are lucky, you may save some on the binary size download, once, but the run time is a web app with all the quirks and inefficiencies of the web as a platform.
You run on a browser, but that's very different from shipping a browser.
And you save more than merely the download size: there's the equivalent disk size savings but also not having to load n browser engines into RAM at the same time.
There's definitely stuff I don't like about Tauri — the way the UI is quite so separate from the native logic, for example — but there are still definite advantages over Electron.
> but also not having to load n browser engines into RAM at the same time.
have you noticed the amount of sandboxing required by (the lack of a) modern web's security model? Each app/tab needs its own JS/WASM interpreter in a dedicated process (for memory isolation), common assets (images, JS, CSS, …) must be duplicated in memory as to defuse timing attacks (since spectre/meltdown), rendering/WebGL contexts are a similar story. You may be over-estimating the savings there.
More to the point, I never said that the above is pointless (the argument was never about the pros and cons of tauri's using the system WebView), the argument was that it is mostly irrelevant: tauri apps, just like electron apps, are inherently web-based. They have more in common than e.g. a tauri app and a native C++/Qt app. The gripes people have with electron apps (lacking integration, wasting resources, …) apply almost equally to both to the point it barely matter from the end-user's perspective.
Is that a whole UI re-write in a native/non-web framework? Because in case that's not, this is still running in a browser, i.e. it stands at 80% or so of the bloat factor of electron, and what makes it faster is possibly (mainly?) that they haven't gotten yet to adding as much stuff as seen on the regular client.
Discord is a server-side protocol (the client isn't doing much besides displaying the server-prescribed state), there's very little logic/computation going on on there that should explain a world of difference.
> Instead Europe is flooded with underpriced Chinese cars that they have no control over.
Please someone fact check that, as far as I know there are import taxes on Chinese EVs, and protectionist measures. The only "loophole" I know is for vehicles built in the EU, which is fine as it employs local workforce.
> it was so refreshing using a system that does not purposely work against the user.
I don't share this optimism, Apple has not been a hardware-first/computers company in a very long time and nowadays makes more bucks profiling your consumption habits to sell you "content" on their ever more "content-consumption" oriented devices.
Edit: just to add to that the same ads (for cloud storage and premium services) that people in the Microsoft world get pissy about very much exists on that side of the fence. It's a race to the bottom, Apple isn't there to lead by example.
on the surface, it's a boring hierarchical note taking app, and you can use it just as one, keeping things simple. But then, you can jump into the rabbit hole and figure out how attributes, templates and properties inheritance can give your notes a structure, be organized as collections, and become a very dependable PKMS.
I personally settled on Trilium after a long search for a PKMS to fit my needs, including Obsidian, LogSeq and many others. My top criteria were:
- I need typed notes: if I'm set to maintain long references of People, Places, Movies, … I want the system to keep track of types, attributes, and relations, just like a RDBMS would. In practice, many systems fail at that and give you a ton of maintenance work just to keep your knowledge base consistent, well-structured and usable
- I need something that syncs with my server at home, and lets me work offline when needed. Trilium client app does that without fuss nor requires subscription/premium/membership.
- I need r/w access to my notes on devices which I don't own (e.g. a work computer), so a fully-featured and secure web version with MFA must exist as an alternative to a client application, and again, Trilium delivers there.
I've been daily-driving Trilium for almost 3 years now, and I haven't stumbled upon an alternative with envy. OTOH, Trilium got (much) refined and gained lots of clever features.
For movies, nothing very special, I generally just use the note body to record impressions. The extent of the "typing" happening here is limited to making sure (by the use of a Template⁰) that all my "Movie" notes share the same icon, and possibly colour, so that they are easy to identify on e.g. a "Note Map"¹.
What I'm after is the ability to retrieve "when" I watched that movie, "where" and "with whom", and this context is inferred by the journal: I will have a "day note"² containing a block with a sentence such as "10pm - Went to the theatre at @ to see @ with @ and @". This will create backlinks (just like in LogSeq, Obsidian & al.) in a manner that opening either of @, @, or @ will lead me straight to the context if I ever need it.
Now, say I want to build a dynamic view/report on something, say for instance, all the times I went to the movies at @, and render that on a calendar, "Search notes"³ enable that: I can search for all the journal references to @. More than one theater? Have you by chance added a `"Place Type":="Theater"` attribute to your "Place" notes? you are covered. Do you need fancy search? You can use search scripts, or just tap into the SQLite database directly.
In other contexts, Trilium's properties/attributes management is a godsend. For instance I have "work projects", with "customers", "project codes", "work locations", etc, and having the ability to represent those projects on a timeline with the calendar collection, or as a semantic graph (of who does what/where) at a glance is really nice. If I want to extend the "data model" by having a "Project Manager" field, I just need to add it to the template and all the dependent notes are reflected away, and then I can bulk edit them in a grid collection (which is not too dissimilar to a Notion database for instance).
You can probably see how the searchability from above, combined with basic attributes management, can make dealing with certain daily tasks very easy (e.g. your boss asking "how many times this quarter did an Azure outage cost us uptime on application X", whether to renew a gym membership "how many times did I go to this gym last quarter", …)
Or more probably, it's imputable to biology and the natural replacement of the "old guard" by adults who didn't grow up during nukes schizophrenia. Or in other words, I doubt that those folks reasoned themselves out of the irrational fear for nuclear energy that they did not reason themselves into.
I'm not sure it's irrational. It's perhaps oversensitive. There are dozens of nuclear power incidents rated at levels of 5+, affecting the wider region.
In the UK, Sellafield was a nightmare. Lots of less serious slip ups and malfunctions punctuated by a few critical incidents, between 1954 and 2003. You might understand why yesterday's Green Party might think it an issue worth pursuing.
I haven't read TFA yet, but having studied academically and then spent years implementing ML techniques for industrial applications (in the areas of machine vision and signals processing), I think that the premise of "AI exemption" is just nonsensical, by nature of being ill-defined. AI isn't a singular product or technology, it's an abstract family of computational techniques, some more complex or mathematically involved than others. Finding function parameters to fit data points using a least squares method is ML, therefore AI, and you would find that in every thermostat, ABS system or electric bike. Too simple? Here is a gradient of increasingly sophisticated ML techniques, with LLMs not even being on the top of the list. Where do you place the cursors of "what's too much AI"?
reply