Aaaaah. I really, really wanted to complain about the excessive amount of keys.
(My comment above is partially a joke - don’t take it too seriously. Even if a new key was added it would be a bit more clutter, but not that big of a deal.)
I have nothing against the people that are working on AI and appreciate the work they do. However every time I see an article about a company using AI like this I just get the vibe that it’s a bunch of middle aged men trying desperately to make things like the “future” they saw when they were a kid. I’ve seen amazing implementations of AI in a lot of different ways but I’m so sick of dumb ideas like this because some guy that used to watch Star Trek as a kid wants to feel like they live in the future while piggybacking on someone else’s work. It’s like the painted tunnel in cartoons where it looks like a real tunnel but in reality it’s just a very convincing lie. And that’s all that it is. Complexity does not mean sophistication when it comes to AI and never has and to treat it as such is just a forceful way to make your ideas come true without putting in the real effort.
Sorry, I had to get that out. Also I have nothing against Star Trek and I used to watch it as a kid because my parents watched it all the time.
some guy that used to watch Star Trek as a kid wants to feel like they live in the future while piggybacking on someone else’s work.
I don't think they care about their own nostalgia. I think they ant to use other people's dreams to make a lot of money. I'm also sure some of them genuinely just ant to push the technological envelope just cause they can, ethics be damned. But ultimately, it's just money.
I would love nothing more than the utopian future Trek promised but greed is killing it.
Complexity does not mean sophistication when it comes to AI and never has and to treat it as such is just a forceful way to make your ideas come true without putting in the real effort.
It’s a bit off-topic, but what I really want is a language model that assigns semantic values to the tokens, and handles those values instead of directly working with the tokens themselves. That would be probably far less complex than current state-of-art LLMs, but way more sophisticated, and require far less data for “training”.
I’m not sure I understand. Do you mean hearing codewords triggering actions as opposed to trying to understand the users intent through language? Or is are there a few more layers to this whole thing than my moderate nerd cred will allow me to understand?
Not quite. I’m focusing on chatbots like Bard, ChatGPT and the likes, and their technology (LLM, or large language model).
At the core those LLMs work like this: they pick words, split them into “tokens”, and then perform a few operations on those tokens, across multiple layers. But at the end of the day they still work with the words themselves, not with the meaning being encoded by those words.
What I want is an LLM that assigns multiple meanings for those words, and performs the operations above on the meaning itself. In other words the LLM would actually understand you, not just chain words.
Semantic embeddings are a thing. LLMs “work with tokens” but they associate them with semantic models internally. You can externalize it via semantic embeddings so that the same semantic models can be shared between LLMs.
The source that I’ve linked mentions semantic embedding; so does further literature on the internet. However, the operations are still being performed with the vectors resulting from the tokens themselves, with said embedding playing a secondary role.
This is evident for example through excerpts like
The token embeddings map a token ID to a fixed-size vector with some semantic meaning of the tokens. These brings some interesting properties: similar tokens will have a similar embedding (in other words, calculating the cosine similarity between two embeddings will give us a good idea of how similar the tokens are).
Emphasis mine. A similar conclusion (that the LLM is still handling the tokens, not their meaning) can be reached by analysing the hallucinations that your typical LLM bot outputs, and asking why that hallu is there.
What I’m proposing is deeper than that. It’s to use the input tokens (i.e. morphemes) only to retrieve the sememes (units of meaning; further info here) that they’re conveying, then discard the tokens themselves, and perform the operations solely on the sememes. Then for the output you translate the sememes obtained by the transformer into morphemes=tokens again.
I believe that this would have two big benefits:
The amount of data necessary to “train” the LLM will decrease. Perhaps by orders of magnitude.
A major type of hallucination will go away: self-contradiction (for example: states that A exists, then that A doesn’t exist).
And it might be an additional layer, but the whole approach is considerably simpler than what’s being done currently - pretending that the tokens themselves have some intrinsic value, then playing whack-a-mole with situations where the token and the contextually assigned value (by the human using the LLM) differ.
[This could even go deeper, handling a pragmatic layer beyond the tokens/morphemes and the units of meaning/sememes. It would be closer to what @njordomir understood from my other comment, as it would then deal with the intent of the utterance.]
Like, I know the megacorps that control our lives do (since it’s a cheap way of adding value to their products), but what about actual users? I think many see it as a novelty and a toy rather than a productivity tool. Especially when public awareness of “hallucinations” and the plight faced by artists rises.
Kinda feels like the whole “voice controlled assistants” bubble that happened a while ago. Sure they are relatively commonplace nowadays, but nowhere near as universal as people thought they would be.
Not a single soul wants this. They just want to use every foul trick to get you to use copilot (by accident even) just like they do with bing and their other garbage.
I think it's those stupid hard coded buttons on my remote that I accidentally press every so often then have to repeatedly try and back/exit out of the stupid thing it launched that I cannot remove/uninstall from my tv.
If you can figure out how to get the remote open, you’ll probably find that the buttons are all part of the same flexible rubbery insert (unless it’s 10+ years old). Put a little tape on the bottoms of the ones causing you problems. The insulation should keep them from working, and it’s 100% reversible if you ever do find a use for them.
If it’s one of the older, more expensive remotes with individual switches, then, yeah, pliers and superglue. 😅
Maybe I'm a pessimist but this is going to really resonate with the people who are "looking forward to AI" because they read headlines, but haven't actually used any LLMs yet because nobody has told them how.
I want a voice controlled assistant that runs locally and is fully FOSS and I can just run on my bog standard linux PC, hardware minimum requirements nonwithstanding
All I want is a real life iteration of J.A.R.V.I.S. and several billion dollars so I can blurt out cool ideas and have them rendered and built in a couple hours.
Current LLMs are manifestly different from Cortana (🤢) because they are actually somewhat intelligent. Microsoft’s copilot can do web search and perform basic tasks on the computer, and because of their exclusive contract with OpenAI they’re gonna have access to more advanced versions of GPT which will be able to do more high level control and automation on the desktop. It will 100% be useful for users to have this available, and I expect even Linux desktops will eventually add local LLM support (once consumer compute and the tech matures). It is not just glorified auto complete, it is actually fairly correlated with outputs of real human language cognition.
The main issue for me is that they get all the data you input and mine it for better models without your explicit consent. This isn’t an area where open source can catch up without significant capital in favor of it, so we have to hope Meta, Mistral and government funded projects give us what we need to have a competitor.
Sure, all that may be true but it doesn’t answer my original concern: Is this something that people want as a core feature of their OS? My comments weren’t that “oh, this is only as technically sophisticated as voice assistants”, it was more “voice assistants never really took off as much as people thought they would”. I may be cynical and grumpy, but to me it feels like these companies are failing to read the market.
I’m reminded of a presentation that I saw where they were showing off fancy AI technology. Basically, if you were in a call 1 to 1 call with someone and had to leave to answer the doorbell or something, the other person could keep speaking and an AI would summarise what they said when they got back.
It felt so out of touch with what people would actually want to do in that situation.
I suppose having worked with LLMs a whole bunch over the past year I have a better sense of what I meant by “automate high level tasks”.
I’m talking about an assistant where, let’s say you need to edit a podcast video to add graphics and cut out dead space or mistakes that you corrected in the recording. You could tell the assistant to do that and it would open the video in Adobe Premiere pro, do the necessary tasks, then ask you to review it to check if it made mistakes.
Or if you had an issue with a particular device, e.g. your display, the assistant would research the issue and perform the necessary steps to troubleshoot and fix the issue.
These are currently hypothetical scenarios, but current GPT4 can already perform some of these tasks, and specifically training it to be a desktop assistant and to do more agentic tasks will make this a reality in a few years.
It’s additionally already useful for reading and editing long documents and will only get better on this end. You can already use an LLM to query your documents and give you summaries or use them as instructions/research to aid in performing a task.
A year ago local LLM was just not there, but the stuff you can run now with 8gb vram is pretty amazing, if not quite as good yet as GPT 4. Honestly even if it stops right where it is, it’s still powerful enough to be a foundation for a more accessible and efficient way to interface with computers.
I don’t use Windows, but given that their office key just sends ctrl+shift+alt+meta, I’m afraid that this could send something like meta+alt that windows users don’t use, but it would be useless for some Linux users that already use that key combo.
I’m stuck on Windows at work, and I have a terminal open on my work PC 100% of the time for tasks, questions and formatting duties (format list, reorder etc) so maybe this could be OK.
Wording is confusing however, as Github is what I think when i hear “copilot”, but I do know both are owned by M$.
I’m not for the extra key, I use a custom ortho split anyway, but I’m guessing it’s just the “windows key” again with hype…
So you can pressed accidentally activating the fucking AI and make the numbers go up so Microsoft can then go and say to investors look millions are using my AI. So annoying.
Tbh I was kinda sad they killed her off instead of trying to make her an actually useful AI assistant. Seemed like a missed opportunity since her Halo counterpart is an AI as well, and it would’ve been cool to maybe have an AR partner app that would have shown pre-Halo 4 version of her.
This is the dumbest fucking thing I’ve ever heard of. I’m not buying any keyboard or laptop that has this key. There’s enough Linux-first vendors these days that it’s easy to avoid (Framework, System76, Tuxedo, etc). It’s time to be done with Lenovo and Dell.
I fully agree with you, but Framework is definitely not Linux-first. The only OS they offer preloaded on their laptops is Windows. You have to install Linux yourself if you want it.
I think they’re referring to Framework’s support for full Linux compatibility for at least Ubuntu, and making sure that the parts they use have first class Linux support and drivers and kernel integration.
This is the dumbest fucking thing I’ve ever heard of. I’m not buying any keyboard or laptop that has this key.
Which is exactly what people said about the Windows key.
Now it's all but impossible to buy a keyboard that doesn't have it. Worse, most of us use it without thinking.
Sure you can call it Super if you like, and even have a Tux key-cap on it, but there used to be a literal gap between the Alt keys and their Ctrl brethren in the lateral directions away from the space bar, and those days are long gone.
There'll be the niche users who stick with old keyboards without this new key, just like there are the die-hards who have stuck resolutely to the old IBM keyboards and the like from pre-1995, but if you want a new keyboard?
Gonna have to shell out a small fortune for a custom build or make do with that dumb new key.
(Shoutout to the Context Menu key which went as unmentioned in the above as it goes unused in day to day use, despite having been included with its Super cousin since day one.)
We have so many unused potential binds already, though. Knowing the way tech goes these days, they’ll find a way to hard-code the key to one macro and that’s it lol
Pure hyperbole “late stage capitalism”: they’ll have it wired directly into the board. At best it will cover one key chord.
Even later stage, it’ll send some proprietary data that only windows 11 can interpret. Linux users will figure it out and make use of it, then will be promptly sued out of existence for copyright infringement or something lol.
What, fuck licenses, we’re doing subscriptions here. With multiple tiers, first one just reduces the charge per activation, and the ones after that give you X “free” uses per 12 hours.
The article actually says the Copilot key will mostly be replacing Menu or Right Control on existing layouts. So if you’re already not using those (or are already re-binding them), it’s just a new keycap.
As you said, there used to be a gap there. Replacing a gap makes not that much harm and people find it useful even in Linux for keybindings. In more of an Alt kind of guy, but Super is also there for more combinations available.
The Copilot key appears to be going were the right Control or right Alt key are right now, so that’s going to be a bother for a lot of people.
It depends on how and what you’re measuring. A lot of Linux first, like system 76 and purism, do so e serious work on the firmware and boot systems of their systems. Which for some is a huge value add compared.
Same, I think I might give the System76 Darter a try when I eventually have to replace my Xps 9370. It’s bad enough that my computer comes with a windows logo on the super-key and often windows preinstalled. Shipping with a non-ANSI/ISO layout is a no-buy for me.
I don’t care as long as the placement is ok and I can map it to something useful. I’m a GNOME user so the Windows/Super key gets a lot of use. It’s nice to have. A new key that I use for all my custom shortcuts would actually be kind of nice. Who cares that the default key caps are a Windows icon and this Copilot thing? Change the key caps and they are just keys.
Add comment