Well I’m guessing they actually did testing on local AI using a 4GB and 8GB RAM laptop and realized it would be an awful user experience. It’s just too slow.

I wish they rolled it in as an option though.

I think it makes sense. I like ChatGPT and I appreciate having easy access to it. What I really wish is the option to use local models instead. I realize most people don’t have machines that can tokenize quickly enough but for those that do…

Why didn’t you like Hashicorps Vault? I want to know for my own edification.

Obviously don’t pet service dogs. Just to be clear. This photo was intentionally photoshopped to make it appear as though the airport was saying “Travel Advisory: when traveling don’t pet dogs”. As in, when in the UK never touch a dog. I thought it was funny/cutesy. I didn’t intend on sparking a big debate about the ethics of petting dogs or the rules about service dogs.

Don’t touch service dogs.

Pet dogs if you know them or are introduced to them.

Basic dog rules people. Teach your children

How do they prove your age? Non-technical savvy people probably just give their kids a phone and don’t do much to lock it down.

Would be kind of cool to allow people to choose an install method. As someone who has experienced low bandwidth in rural homes, it would be nice to avoid the waste at the cost of possibly managing chromium versions myself.

That’s the point. The storage is a bad metric. While it might indicate poor performance, it’s not a direct indication of poor performance. The bloat and optimization comes from the usage of Electron. And people use Electron because it’s far easier to make cross-platform deployments for Web and desktop using a framework like Electron. Show me the QT/JavaFX app that mimics Signal and we can compare the cost to develop it. Electron isn’t the best choice for memory usage and reducing bloat, but it’s the best choice for quick development (in my opinion but also proven out by the market share it has)

Nah it’s literally a waste of physical resources. Crypto currency is a waste of fossil fuels. AI has its functions at least.

For me it’s a pattern of “Ctrl+t” to open a new tab and then I search “my interesting query”. After that, I use “shift+tab” or “Ctrl+shift+tab” to navigate between tabs. Rinse and repeat until I get tired.

I don’t like searching in my current tab because I don’t want to lose the info I have.

deleted by creator

deleted by creator

I didn’t until apps started breaking. The snap version of steam, Firefox, and Unity (I think?) all started to have issues. When I googled around people would often ask “deb or snap”? I uninstalled the snap packages and installed the deb packages and most of my issues went away.

I ultimately switched to Linux Mint because I kept having stability issues and I was just desperate for a solution. But snap was not a great experience for me.

Yeah the majority do it and I think it’s bad.

Thermostats are easy to change out. So this isn’t a huge deal. But I don’t love the idea that tech isn’t built to be self-hosted or maintained in any meaningful way. If you’re not shipping an open source version of your software when you close up, you’re an asshole.

Yeah, self hosting isn’t for most lay people if it’s just a GitHub repo. But GitHub repos quickly become adopted by nerds like me who build tooling around it that eventually let lay people self host software with the click of a button.

I see. Well without a command line, I wouldn’t call it a terminal. I think you just want tooling to be available on an Android? It would probably look like a button or series of buttons on an app. Maybe you could connect the dots between them to insinuate a pipe? E.g., you have a “mv” button and a “file” button. When you drag from mv -> file you could maybe kick off a process that moves the file. Maybe it would prompt you for other arguments like destination? I suppose this theoretical app could allow people to install additional tooling and make their own custom commands.

But I just feel like a button UI for these kinds of things will always be awkward. If you don’t have a keyboard/terminal interface, it’s hard to implement anything that would even behave like terminals in terms of functionality.

Great examples are already in the thread, but generally speaking the answer should be “no”. Smart phones are just slow at typing. In the case of a smart phone, hitting a button is far faster than typing a command. Not to mention our devices aren’t really being used for file management, tooling, complex work, etc. So it doesn’t even make sense to have a command line unless you’re a huge unix fan or if you are doing something quite niche. And in that case, I recommend just connecting via adb.

We’ll never know.

I think this article does a good job of asking the question “what are we really measuring when we talk about LLM accuracy?” If you judge an LLM by its: hallucinations, ability analyze images, ability to critically analyze text, etc. you’re going to see low scores for all LLMs.

The only metric an LLM should excel at is “did it generate human readable and contextually relevant text?” I think we’ve all forgotten the humble origins of “AI” chat bots. They often struggled to generate anything more than a few sentences of relevant text. They often made syntactical errors. Modern LLMs solved these issues quite well. They can produce long form content which is coherent and syntactically error free.

However the content makes no guarantees to be accurate or critically meaningful. Whilst it is often critically meaningful, it is certainly capable of half-assed answers that dodge difficult questions. LLMs are approaching 95% “accuracy” if you think of them as good human text fakers. They are pretty impressive at that. But people keep expecting them to do their math homework, analyze contracts, and generate perfectly valid content. They just aren’t even built to do that. We work really hard just to keep them from hallucinating as much as they do.

I think the desperation to see these things essentially become indistinguishable from humans is causing us to lose sight of the real progress that’s been made. We’re probably going to hit a wall with this method. But this breakthrough has made AI a viable technology for a lot of jobs. So it’s definitely a breakthrough. I just think either I finitely larger models (of which we can’t seem to generate the data for) or new models will be required to leap to the next level.