I live near Dublin, CA and almost 99% of the residents of Tri-valley (Dublin, Pleasanton, San Ramon) or in the larger bay area are hardly aware of this. It is family friendly neighborhood. I am surprised that mainstream media or the local media did not bring the attention to this. I knew about this only through local wikipedia articles.
If that were true Obsidian wouldn't let you sync for free with iCloud, or provide guides to alternatives on the official Help site [1], or have a syncing category on the Community site with hundreds of plugins to choose from for every alternative service you can think of [2]. The truth is that Obsidian is a really small team, and there many many different services and ways to sync, so we prefer to focus our energy on making the core app.
Non-intentional sabotage would be Obsidian providing the server and supporting it as a first-class option where you just paste in your server URL into the app without the need for a convoluted third-party plugin.
I don’t think that expecting end users to finish the software themselves with an LLM is a very good user experience.
Check my other comment in this thread, see the setup steps for yourself. I think they’re way harder than they should be for something as simple as a notes app, and it shouldn’t require a third-party plugin in the first place. They should just provide a server docker container and then the app should let you connect to an arbitrary URL without this very long and technical process.
Yes you can just use a cloud folder for storage (iCloud only on iOS) but that’s not the full feature set including live sync.
Obsidian is purely file based, a you can use anything that syncs files (w.g. Google Drive) , but you need a shared filesystem across multiple apps for that.
What parts of Obsidian are slow? I am genuinely curious because I have been relentlessly working for years to make every interaction under 0.1s (and often much faster). To me every millisecond counts:
I did a simple human benchmark with currently latest appimage (extracted, to be closer to native experience).
first startup: ~1.5s - faster than I expected, but still slow
then I start spamming "new note" button - it takes around 40 new notes when it starts lagging, throwing "file already exists" errors and creating like 1 note per second.
was I clicking faster than a file system can respond - no. so what is taking it so long to process one of a simplest operation in the whole app?
okay, let's check ram consumption - 400MB - it is completely unreasonable for a note-taking app to take 10% of precious ram on my PC
And this is out-the-box clean, no cruft, no plugins experience. I can't imagine how bad it gets further on.
I understand that the problem is not implementation, but the core concept of using a web-browser interface for a note-taking app written in javascript. No, that doesn't excuse it being bad.
The work that Kepano and his team have put into making Obsidian perform well has been incredible, even writing their own open-source replacement for Mermaid, Pele.
The performance of Obsidian is generally good, but be aware of what plugins you add. This critique reminds me of people who hate on games using a specific game engine when that never is the deciding matter.
> people who hate on games using a specific game engine
Unity has evolved in time (or perhaps the other engines have devolved) but back when it was new you knew an Unity game is going to kill your laptop battery 2-3x faster than a handwritten game.
Case in point, I could play the first x-com "reinvention" 2x longer on my laptop than most unity games.
Can't say anything about Air, but it works fine on 2019s Pro with 16Gb of ram.
Same for VSCode (for example)
To be honest I don't really understand what can even be slow in an app like this? Do you experience some sort of lag during text input? Or switching between files/notes?
The core of Obsidian is CodeMirror 6, which is hard to replace for its live preview feature. No rust editor and GUI toolkit even comes close to CM6/browser. But electron can be replaced.
OpenCode comes with the cost too. The cost of the hosted version is on par with cloud hosted proprietary models. And we cannot simply run the LLMs on every laptop. Given this, how do we us the browse efficiently?
Yea it was a pity when opencode limited the token usage and told us to subscribe to their opencode go service. Like openai, it's clearly angling for commercial viability status under the guise of being 'open'.
In the age of LLMs and Multi-modal models, do you think students across the world will benefit from this? If so, how do you think they will benefit.
For some context, when others released courses in youtube/coursera/udacity decades ago, many students, across the world, did not have any real access good teachers and qualified lecturers who could explain the concepts well. Those releases were boon to them.
Now, we learn by asking AI against anything, include youtube content and AI has been best teacher so far.
On a long enough timeline, we are all dead. In the meantime, come check out Graham Neubig and Daniel Fried teach AI agents, and you will see that we have value still.
> I tried testing each frontier llm on questions w.r.t deferred tax and it made up stuff
This would be a stronger example if it were related to Comp Sci. It’s entirely possible that an LLM would perform better on CS fundamentals than details about taxes.
You're going to get flagged for asking - it does seem to be in bad taste IMO - but I do think it is the elephant in the room.
"AI" will teach you anything at a undergrad level - customized to your style.
The real reason you'd go to CMU or MIT is for credentials, or, more honestly, for networking opportunities. That doesn't exist for watching YouTube videos any more than prompting an AI.
Hard disagree on the reason to attend a school like CMU or MIT (disclaimer, I’m a CMU MatSci PhD)
Networking is a big value proposition, but the biggest benefit IMO is learning what questions are important and learning how to think independently about them. LLMs are great for learning but I don’t think they fully replace the skills you need to ask important questions. (This perspective is probably slanted towards research degrees but I think it’s at least somewhat transferable to engineering)
Partially true. It's the rigor of the education, the thinking process taught, the caliber of people you work with / meet and especially the office hours with top faculties at CMU is really what at mold you. Not information.
"AI" cannot teach a person what they don't know they don't know.
Human curation will always have a place in passing on knowledge for people who desire deep understanding of a topic, particularly young people and novices in a field.
You can ask ChatGPT for a detailed listing of all the areas within a discipline and it will give a totally decent answer, better than you'll likely find anywhere else. You can ask it to list disciplines that might be useful, that are taught at university, or whatever.
You can even just ask it "help me know what areas/subjects/topics are useful that I wouldn't even know exist"
Me: Tell me about a data structure I may not know? Suppose I am a software engineer with 10 years experience
Claude: Here’s one that a lot of experienced engineers haven’t run into: the XOR filter, along with its newer variant, the binary fuse filter.
What it does. It answers the same question a Bloom filter does: “is this key possibly in the set, or definitely not?” The difference is that it uses noticeably less memory and answers queries faster. The trade-off is that it’s static. You build it once from a known set of keys, and you can’t add keys afterward.
...
Does that sound like something a developer with 10 years experience may not know?
I'm not a professional developer, only a scripter. I wouldn't know.
Suppose you're a college student with barely any concept of CS, let alone the capability to know what kinds of questions to kick-start your own education.
Nothing is new in my argument. AI is particularly useful for professionals that already know how to learn, and know how to sniff out BS and second guess the tool.
You have an amazing dataset, handed out for free by a top rated institution like CMU that can be used as the baseline for whatever learning flow you want to do with your AI.
You do understand that data like this is precisely the goldmine that makes AI good at answering general knowledge questions?
Why would I personally be invested in what information someone else's AI is ingesting? No normal person is developing their own AI that will benefit from this, and lack of content is not what's holding back llms. Your list is the opposite of interesting. Humans should care about the knowledge opportunity and not the ais.
LLMs are so sterile, I would never use them for primary learning. It would be joyless and require too much effort on part of student. They're no substitute for lectures and a good teacher. They're great for supplementing the learning material.
Common sense around a lot of topics can be misleading. Especially when human bias is taken into account. It's not at all clear to me that a properly trained LLM would not out perform a professor. Even we are talking about known great teaching professors, not average teachers, the LLM has several advantages. It can tailor curriculum to each individual. It can learn and adapt curriculum on the fly which a prof is unlikely to do. It can adjust testing and grading so that the student is maximally engaged. It never misses a day or has a bad day. It can answer questions thoroughly and with as much detail as the student wishes, taking as much time as the student desires. What are the disadvantages?
> It can adjust testing and grading so that the student is maximally engaged
This part is not reality.
In order to do this, the AI would need a way of measuring student engagement.
As soon as you have a measure like that, it is going to be gamed.
Somebody needs to set the standards. You can’t rely on an LLM for that because the LLM doesn’t make up a goal.
Even in a very basic non-LLM example like learning vocabulary, the acceptable rate of correct answers (a day or a week after studying) needs to be set externally.
My Stats I and II professor (don’t ask me how) spend the first 5 weeks explaining his take on best-practice US train travel, and sharing his experiences with us.
General common sense tells everyone that not all teachers teach well, and that not all students can learn well from all teachers who may teach well because style preferences may collide. One may easily calibrate an AI to teach them in their preferred style; not so much a college professor.
It's more that you can continue prodding chatgpt/whatever. If you are a naturally skeptical person and try to poke holes in ideas by nature, you don't really get to do to your hearts content with a professor.
But, some people aren't skeptical. A professor says X and they just accept it and feel like they learned from a trusted source or authority or whatever.
No....just like that doesn't work on humans. But with prodding you can work out when it's wrong or at least when you should be skeptical. People usually call this "critical thinking".
If you go around assuming authorities on subjects are just right, you will be in for a difficult time. If you endlessly prod and show skepticism towards human experts/authorities, you'll piss a lot of people off. Most of the time they won't even engage.
If you are genuinely interested in learning things and have critical thinking skills, sota LLMs are as good as it gets.
You seem the kind of guy who distrusts professors and experts and used to trust YouTube idiots, and now turned to LLMs.
Professors will occasionally teach something wrong, but you'll also be studying from books, and if you really want you're free to eventually also use LLMs.
Beginning from them, though, is not too smart, and relying exclusively on them is just plain stupid.
LLMs are going to make stuff out of thin air, and can be more and more confident of it the more you "prod" them.
You seem to be struggling to read the thread. Nobody (except you) said trust, just use. Critical thinking is always required. And no one said rely exclusively on LLMs. And "prod" doesn't mean what you seem to think it means. You prod to find holes.
I haven't watched the movie yet. I've read the book too. I want to read the book and then watch the movie.
The fact that it is documentary, and is being marketed along the lines of a block-buster on X with memes should have raised eye-brows. There are lot of genuinely good documentaries out there, they all are left to be discovered, and people discover them. But this one, was/is being marketed as must-watch documentary, and with no reference to the book bad blood?
Why would the documentary reference a book that it has nothing to do with? It's not an adaptation of prior works on Holmes or Theranos. You Can See Everything is something else all together.
Wont there be any overlaps? I tend to think, that great documentaries follow from greatly researched books. The later usually have more space to explore and can in depth, the documentaries to try to condense.
As much we credit Elon, can we take a moment and credit the amazing engineering feat of the Telsa and SpaceX engineers who have pulled off this feat. Incredible. Thanks for advancing the human race forward.
Yeah, only yesterday I let claude build and deploy me a end to end project that will flicker my lifx light-bulb when commits to any open source project lands. In the morning I saw it stopped around 11 pm PST yesterday. I was wondering why and I was wondering, how easy it is do a complete working project like this in minutes and what is stopping people to do 100s of such project per day. Github wasn't built for this scenario.
Free Github Actions weren't built for this scenario.
Make Github Actions a paid-tier exclusive and this problem goes away.
GitHub are too scared to do that after the backlash around charging people for their own runners, but this would be a very different thing, simply stop giving free actions to people who have never paid them a penny.
> GitHub are too scared to do that after the backlash around charging people for their own runners, but this would be a very different thing, simply stop giving free actions to people who have never paid them a penny.
I had no idea they charged you to run your own runner. I am immediately of the opinion that is wrong.
Conversely, I would have no issues if GHA become a premium product. Or maybe "your actions only run when we have spare compute". I've personally given Github money (back when private repos cost $$) and wouldn't be concerned about doing it again.
They don't charge you for your own runner but they wanted to. The downtime hasn't really impacted me and not really a reason for me to leave github, but making me pay for using my own infrastructure would quickly encourage me to leave github.
> Conversely, I would have no issues if GHA become a premium product.
This is a real problem for acquiring customers. CI/CD / Actions in general has a low initial perceived value and a high perceived value after use while being primarily a commodity product (serverless compute). To acquire customers they need to offer a free tier, to make money they need to charge their heavy users more, but because it is a commodity their heavy users will self-host runners or jump ship.
Action minutes is a really good proxy for value gain from cloud git services, where repo count, user count, file count/size, etc isn't.
I think a reasonable amount is fine for them to serve as the orchestrator, queue, and control plane for self hosted runners, has to be reasonable though. Think monthly utility service cost to maintain service vs usage.
It isn't just that they wanted to charge for using your own action runners. It's that the price was almost as much as using GH's hosted action runnners. Which didn't even include the cost of hosting your own runners.
I think removing the free tier altogether would be a bad move, especially given that some important open source projects rely on it. But they could possibly lower some of the limits, especially for new accounts, with a way for open source projects to apply for more, and maybe have an automatic scale up for projects that are more popular, by some metric. And maybe limit the free tier to public repos, and maybe even only repos that use an OSI approved license or something.
Likely easier said then done, but why not separate infra out from shared vs dedicated hosts? Something like, actions will still run but at some degraded capacity, where actions and PRs are still at least functional.
My company has heavy handily moved from self-hosted version control system to github and now we’re suffering more outages. Though the self hosted system had less features, and we needed to maintain our own jenkins infrastructure for CI/CD, at least we really only suffered from very major cloud-provider level outages
worse than this is that if you fork a repo that has actions, the actions will run on your account as well - which is completely unnecessary. I have all sorts of stale forks that are just churning away with actions that I dont care about at all because I am not getting charged.
The solution, of course, is not to simply start charging for all actions, because then everyone would wake up to all sorts of mystery charges. They'd need to notify that forked actions would be turned off and then opt-in for paid actions.
Is this really the case? Because GitHub availability has been going downhill since way before models hit that critical mass of quality end of last year.
reply