Whats your thoughts on Ai in your terminal?

Steamymoomilk@sh.itjust.works to Linux@lemmy.ml – 80 points – Warp: Your terminal, reimagined
warp.dev

Today i was doing the daily ritual of looking at distrowatch. Todays reveiw section was about a termal called warp, it has built in AI for recomendations and correction for commands (like zhs and nushell). You can also as a chatbot for help. I think its a neat conscept however the security is what makes me a bit skittish. They say the dont collect data and you can check it aswell as opt out. But the idea of a terminal being read by an Ai makes me hesitant aswell as a account needed to use warp. What do you guys think?

95

Warp lost me at the account requirement. You're telling me I need to sign in to a terminal? Seriously? Like with an internet connection? Nope. What if I'm opening my terminal to configure my network? Warp seems to be fixing a problem that doesn't exist. I don't think anyone has looked at a terminal emulator and gone "Yeah, this could use AI and a cloud account".

"Alright, now that I'm logged in to my cloud terminal account, let me enter my root password for sudo."

You are not in the sudoers file. This incident has been reported and your account suspended.

Not just that, they want an email just to get a download link. Call me when someone forks it with local AI.

Wait. An email just to get a download link AND a cloud account. Fuck that.

I would definitely like an AI to remember some complex commands for me. But something small and specifically trained that runs locally

I use fuck, it's not ai but gets the job done.

You can define a bunch of aliases in any shell environment for that. Or use a history manager (a database client essentially) that groups commands you've entered so far based on frequency, return value, working dir. when they were issued etc.

fzf does the job

Yeah; & by the way, warp is funding fzf, as there's a big thank you banner on fzf & fzf-vim's github pages nowadays. I'm glad fzf is getting support, of course; though it feels odd somehow.

My thoughts are you can fuck the fuck off.

Totally agree. People using cli are probably more skilled and their knowledge has been fed into these ai models.

So we will all end up with some mediocre level of knowledge, because the next input for the LLM 's will be more of the some old stuff. Flattening the curve and less innovation and smart ideas.

These kind of "solutions" are for a non existing problem. Looking at the investors, this is only about making money.

People using cli are probably more skilled and their knowledge has been fed into these ai models.

I am not skilled at all. I only use it because I hate Microsoft and Apple more.

AI that can auto generate all those command line arguments I keep forgetting? Sure.

Closed source terminal that requires account? No way.

And also, like... Data privacy... My terminal commands and command outputs contain sensitive data. Even company sensitive data. I don't want to be liable.

Absolutely not. And they can fuck right off with that whole needing an account to use a terminal thing.

To help make skittish people feel at ease with the concept, why not give it a friendly on-screen avatar? Perhaps something like a cute little animated paperclip.

I dunno. Maybe an orange dog? Give it big brown ears.

Also animate it at ~10fps, making it visibly sad when it can't retrieve the files you ask for.

1 more...

I have zero interest in having AI in my terminal. And needing an account to even use warp is a non starter for me.

So,

This is a proprietary and therefore untrustworthy terminal in a space where virtually all the competition is libre/open source.

It's connected to the cloud, therefore insecure and privacy-invasive as there is no reason for something as basic as a terminal to be connected to the cloud. Who wants their SSH keys leaked? Anyone?

They require an account but don't collect data? Sketchy to say the least, a unique account is the perfect tool to collect data and there is no reason a terminal, the most basic interface to the underlying OS should require an online account. It should be tied to the system. (After further reading, apparently they do collect data by default).

It has a built-in AI autocomplete, because apparently normal auto complete isn't good enough (just wait until it tells you to rm -rf /*).

Yeah, no matter how nice it is, I will never accept this terminal.

EDIT: They also forked Alacritty to create a "demo", they took advantage of a libre/open source project for their proprietary terminal and never did so much as thank the authors of Alacritty. That's scummy.

AI can be neat, but this is a solution looking for a problem, like most AI things.

IANAL but it looks like they are violating Apache 2, as they are supposed to retain the license and mark any changes.

I'm neutral towards AI, what I can't wrap my head around is forcing users to sign in / sign up to use offline apps. Fuck you too, Postman.

For me to even consider using AI in my terminal, it'd have to meet a couple of requirements:

  • needs to be open source
  • needs to be run without network access
  • needs to be an extensible utility to any terminal program.

(And that's off the top of my head.)

The terminal seems like the very last place I’d want A.I. I’m usually using it specifically to be precise and don’t just run commands I don’t understand. If you forget some long command, just use history |grep whatever and see what it was. (And then turn it into an alias or function.)

Exactly. I generally like typing out my commands because I’m learning and it helps me remember what I’m doing and what the commands mean/how they work. And if it’s a particularly long one I’ll make an alias for it.

What the terminal needs is better discoverability. Maybe command recommendation if it isn't going to hallucinate flags and paths that don't exist. All this bullshit is just some company trying to capitalize on that desire.

What's my thoughts? My thoughts are FUCK relying on the internet for basic things. So no "AI terminal" for me. This is yet another way to mine data cloaked in futurism.

Sounds like a major security risk. All it takes is one "hallucination" (and an overly trusting engineer) from the latest and greatest bullshit generator to compromise an entire network

Yeah. Sometimes a "barrier to entry" on running commands serves as an important forced pause to help prevent people from charging headfirst into dangerous options they don't understand.

It's something I often have to consider at work. It's not too hard to script out ways to make it easier to do certain things, but is the trade off of making it easier to do accidentally or without understanding the full effects worse than the hassle of doing it the "hard way"?

Yes, let's get a list of all machines in this network segment, then loop through sending shutdown commands so everything is ready for the hardware move!

What do you mean that the switch itself is in the list of machines? And that I just shut it off prematurely, so now we need to shut down everything locally... shit.

(Details fudged to protect the guilty)

Maybe if you can use it with a locally running LLM server like ollama, but otherwise fuck no

Nice idea for fun and diversity (you can't prohibit people to make such apps after all) but in daily usage? No, no, no and no

Here's my opinion. This terminal app is inefficient as fuck. I feel like it's too much bloat for what a terminal should be. She'll Completions have existed since forever, I don't get what's bringing new with that. And all these AI's that just resell chatgpt are getting expensive. "Please pay me 10$ a month to have OpenAI in your " . If I were to activate all the AI subscriptions in all the apps I use it would go over 100$. If I need ChatGPT I will just go on their website and get it from there. It's even cheaper that way, 20$ for unlimited use.

And also sharing info with your team THROUGH THE TERMINAL? WHAT KINDA SHIT IS THAT. That should be documentation in THE REPOSITORY, IN THE PROJECT. You're just fragmenting information, and it's going to make it harder for you to keep it up to date and for people to find it.

And... I don't want to force on my team "hey you have to use this terminal otherwise you won't have the info". I feel like with this I would be encroaching on their personal space and way of using their computer.

Kitty terminal with my zsh plugins are far better than asking me for an account on your terminal.

I feel like all these apps with OpenAI ask me to pay just so I don't have to manually copy and paste to the ChatGPT website. Oh please...

And also like "workflows" just get yourself a makefile or a task file and now you can reuse your "workflows" in any terminal, shell or environment.

I don't like generative AI in my tools. The little prompt that explains a command and arguments that can be passed as you type is nice, I will give it that, but AI should not be any part of it. Fuck right off with it.

Local AI.

To complete complex commands.

Make ne feel like taking to a Star Trek ship computer.

So I took some time to look around and as far as my perspective as a non dev regular user. While this does seem like a useful tool that could be useful for someone who interacts with the command line on a infrequent basis, the drawbacks on it seem pretty big.

  1. Everywhere on their website seems clear that they don't store your data, but I have trouble believing that? Why on earth they would need for you to create a account that you must log in to use the terminal if they don't have a need to monitor your data?
  2. While they claim that they are intending to monetize this by charging enterprise users and letting small teams use it for free, they limit free requests to 20 per dday which seems less than useless.
  3. Maybe this is just some confusion since I don't have any experience as an enterprise but it seems like it would be an unacceptable security risk having a program that it telling you that it sends telemetry back home that users are interacting with using sudo and elevated privileges. Especially when it is a closed box.

Ignoring all the reasons to be cautious and skeptical about AI in general I struggle to see the use case for this particular tool.

And now I am imagining some sophisticated hack that breaches their AI generator and starts slipping command arguments that might expose your system. Probably too much of an effort but still plausible.

I think AI exposes how little trust people have in Capitalist organizations.

Shouldn't be too hard to make that run locally. Although I'm not sure what I'd use it for at the moment.

Although I’m not sure what I’d use it for at the moment.

How do I find all instances of "blarg" in the second column of this CSV file?

I could see it being useful - but I wouldn't want it integrated to my terminal. I'm fine with it being a separate thing I can use.

Executing commands of powerful command line tools without losing time finding them.

There are lots of fish shell extensions, zsh stuff and loads of things that make suggestions, autocomplete, remember your shell history and remember frequently executed commands and visited directories. All of that works WAY better than the AI suff. (And sometimes also has nice pop-up menus.)

So compared to plain bash without autocomplete and Ctrl+R it may be useful. It is probably a step back for everyone else. Especially if they roughly know what they're doing.

But I didn't try this specific software. Maybe I would if it were free software and connected to a local LLM.

So compared to plain bash without autocomplete and Ctrl+R it may be useful. It is probably a step back for everyone else.

I think it could be much worse than even a plain shell with ^R, as the llm will be slower than the normal history search and probably has less context than the $HISTFILE.

I think so, too. I mean the traditional history search and command option suggestions are instant and come at no additional cost. I don't know how fast ChatGPT is, I only ever play around with local LLMs. And roughly exploring what Github Copilot is about, just made my laptop fans spin on max and started to drain the battery really fast. Would be the same for an 'AI' terminal. And when asking the LLMs for shell commands I got mixed results. It can do easy stuff. So I guess for someone who wonders how to find the IP address... It'll do the trick. But all the things I tried asking some chatbots that would have been really useful to me, failed. It hallucinated parameters or did something else. And I needed to google it anyways or open the man page.

I'm not sure, I currently don't see me using such tools. I like talking to chatbots and have them draft stuff and provide me with ideas. But I also like computers in the other way, that they are machines that just follow my orders and don't talk back. And when working in the terminal or coding, it seems to distract me if suggestions pop up and I need to read them and decide what to do, or occasionally laugh... For me it seems to work better if I think about something, have an idea in my head and type it down without discussing it with the machine... I mean not 100% of the time, sometimes a suggestion helps... But I think I rather have the chatbot in a separate window and only loosely tied into my workflow if at all. And I don't like proprietary and cloud-based products for something like this.

It hallucinated parameters

Sound like LLMs to me. This is not going to stop being a problem. This is the fundamental problem with LLMs - they are text prediction algorithms and have no comprehension of their output.

I'm not sure. Afaik the research is happening. And AI related stuff always happens faster than I can imagine. Ultimately I want the LLMs to hallucinate. They should be able to combine ideas and come up with new and creative answers and be more than just autocomplete. I think what we need is the LLM knowing what it knows and what is made up, and a setscrew. I can see this happening with a higher level of intelligence and/or a clever architecture. I'm not an expert on machine learning myself, however that is what I took from news, companies struggling with their chatbots and everyone wanting their AI assistant to provide factual information. And I don't see anything ruling that out completely. I mean we humans also sometimes get things wrong or mis-adjust our level of creativity. But I think the concept of facts can be taught to LLMs to some degree, they already seem to grasp it. And concepts have been proposed and things like AI agents that come up with ideas and other agents that check for factuality are in active use. Along with the big tech companies making their AIs cite the sources. In my eyes, progess is being made.

But this is why I currently don't use LLMs for important and unsupervised stuff, and i try to avoid them when I need correctness. However... I really like to tinker with them, do AI assisted storywriting, or have them come up with 5 creative ideas for a birthday party for my wife. That works well, and with a bit of trickery you can make them output more than the most obvious ideas. And I'm impressed by their ability to code, but as I said it's still far away from being useful to me. I currently don't fear for my job. And I additionally struggle with the size of models I can do inference with and their respective intelligence... We're in the Linux community here, so I think I can be open... I don't like big tech companies doing my compute and providing me with closed and proprietary services. I don't use ChatGPT, only open-weight models I can run myself. They aren't as smart, but I don't want the future of humankind to be shaped by services and good will of big tech companies.

I don't know what AI could bring to the table in this case that you can't do without it already. Command completions or fixing typos works without using AI. If there was an actual benefit, I'd be open to try it out but only by using an open source LLM running locally. I'm definitely not creating an account and paying a monthly subscription while not even being able to use it offline.

I'm not the biggest fan of the forced account thing, but I do like a lot of Warp's features. The command suggestions especially make dealing with tools that have like 1000 switches so much easier (like docker for example). Other than that... It's easy to customize, fast and looks good.

Tl;Dr: I like Warp, cry about it.

Command suggestions can be provided by the shell too for what it’s worth. fish ships with autosuggestion and autocompletion. For zsh, you need a separate plugin (but it’s well worth it)

I use fish and have used it for a long time and it works very well with warp, actually. You get both it's autosuggestions and warp's autocomplete. 's nice

Its not that i hate the idea of having an AI in the terminal its just the idea of having a account to use it. I played around with it last night and tried diffrent ideas which it sometimes is useful. I did #what is my graphics card "Lspci -k "VGA"" Ok that was helpful Then i tried #what driver is my graphics card using? "Lspci -k "VGA"" Which does not list your driver. Its hit or miss.

Hell no! I absolutely do not want AI in anything if it's not running entirely locally.

tldr, fzf-tab suffices for me. For anything else you may give shellgpt a try. But I love my Alacritty with a zsh and p10k.

Helping with complex Terminal commands/shell scripts is basically my #1 practical use-case for AI right now... especially if you use tools like JQ a lot. Saving keystrokes is a lifestyle, after all.

I am also a really big fan of Warp, and was even before they added the AI feature (the editor-style functionality is wonderful). For the record, the AI isn't always running in Warp, to use it you start a prompt with hash (#) and then ask for what you want and it presents options.

Terminal with GUI drop down menus every time you try and type something seems like the kind of Terminal a microsoft executive would dream up

This is faily easy to build using offline models. Only problem is GPU whirring away running typically light terminal commands.

optional autocomplete is a nice-to-have, eager autocomplete is a pain in the ass. as long as it only completes when I ask it to, I don't mind.

To me this is complete nonsense --- but they (warp) seem to be funding fzf ... which is good on their part, I guess.

If I have to use a cloud service or create an account to use the terminal, it’s a no for me dawg.

Did warp ever follow through with allowing folks to use it without signing into your GitHub account?

The AI can have access to a terminal in a docker container on my raspberry pi. If I’m convinced it’s trustworthy it might move up to a docker container on my desktop.

I really enjoy Warp. It’s sleek and modern, plus it saves me a lot of time with its advanced autofill features. It also gives me helpful suggestions for minor edits if I’m making small errors that keep a command from running.

I haven’t used the chatbot, but I have found the user experience of the program to be better than most other terminals I’ve used before.

I'm in the same boat. Best terminal app I've used in a long while. Not using AI features

As long as AI is not being forced into the existing terminal standards, it's good, as you can just choose to not use a terminal with AI.

I feel like every use case they showcase is useless if you remember the commands. And if you don't know a command, the classic googling until you find something that works usually does the trick.

I'm likely going to try out Wave Terminal with a self hosted LLM. I think it may well be quite useful, just don't want to upload my entire command history to OpenAI.

the non AI features of warp, such as the modern editing and easy function creation, are more interesting to me than the AI features. I wish there was an open source terminal that felt this modern.

I have used AI to generate commands many times, but not often enough that I need it built into my terminal. I prefer my default terminal experience be more minimal.

Fish's autocomplete is enough for me. I do like having Copilot in my editor but I can't really think of a reason I'd need it in my terminal. Most of my time in the terminal is just installing things, git or moving things around and I have all those commands down as muscle memory.