Wouldn't it be easier (and cheaper) to just move to a different room?
759 karma · joined September 3, 2025
I'm not sure if it works much better than STT -> LLM -> TTS pipeline would. Also, this is of course just a first impression and not serious testing.
I also had a an idea for a solution to this problem long time ago.
I wanted to make a thing that would allow you to record a meeting (in the company I where I worked back then such things where mostly discussed in person), transcribe it and link parts of the conversation to relevant tickets, pull requests and git commits.
Back then the tech wasn't ready yet, but now it actually looks relatively easy to do.
For now, I try to leave such breadcrumbs manually, whenever I can. For example, if the reason why a part of the code exists seems non-obvious to me, I will write an explanation in a comment/docstring and leave a link to a ticket or a ticket comment that provides additional context.
Is it because this kind of phones are a very niche product so they can't benefit from the economy of scale?
Maybe android phone manufacturers can get better deals from chip manufacturers because they buy chips in large quantities?
I'm not sure what is the advantage. Each user will have to waste time and tokens for the same task, instead of doing it once and and shipping to everyone.
How is it "simple"? There are like a ton of different downscaling algorithms and each of them might produce a different result.
Cool article otherwise.
What was the point of first asking an LLM to expand prompt/"specs" into a book and then asking it to compress it back to a summary? Well, I'm glad you asked! To promote this Z*n tool, of course.
Every time you send what appears as a "chat message" in any of the programs that let you "chat" with an "AI", what you really do is sending the whole conversation history (all previous messages, tool calls and responses) as an input and asking model to generate an output.
There is no conceivable scenario when sending "<tons of tokens> + no" makes any sense.
Best case scenario is:
"<tons of tokens> + no" -> "Okay, I won't do it."
In this case you've just waisted a lot of input tokens, that someone (hopefully, not you) has to pay for, to generate an absolutely pointless message that says "Okay, I won't do it.". There is no value in this message. There is bo reason to waste time and computational resources to generate this message.
Worst case scenario is what happened on the screenshot.
There is no good scenario when this input produces a valuable output.
If you want your "agent" or "model" or whatever to do nothing you just don't trigger it. It won't do anything on it's own, it doesn't wait for your response, it doesn't need your response.
I don't understand why, in this thread, every time I try to point out how nonsensical is the behavior that they want is from technical perspective (from the perspective of knowing how these tools actually work) people just cling to there anthropomorphized mind model of the LLM and insist on getting angry.
"It acts like a bad human being, therefore it's bad, useless and dangerous"
I don't even know what to say to this.
P. S. If you wind this message hard to read and understand, I'm sorry about it, I don't know how to word it better. HN disallows to use LLMs to edit comments, but I think that sending a link to an LLM-edited version of the comment should be ok:
Like, sure, if the model were "smarter" it would probably generate something like "Okay, I won't do it.". What is the value of the response "Okay, I won't do it."? Why did you just waisted time and compute to generate it?
> Do you never have a second thought after looking at your output, or realize you forgot something you wanted to include? Could it be that they saw "one new function", thought "boy, there should really be two... what happened?" and changed their mind?
Sure, all of those are totally valid. And in each of this cases it would be better to just don't make the request at all or make the request with the correction.
> like calling your intern
LLM is not human. It can't act on it's own without you triggering it to act. Unlike the intern in your example who will be wasting time and getting frustrated if they don't receive a response from you. With model you can just abandon this "conversation" (which is really just a growing context that you send again and again with every request) forever or until you are ready to continue it. There is no situation when just adding "no" to the conversation is useful.
I don't see anything odd about this question.
What kind of response did the user expect to get from LLM after spending this request and what was the point of sending it in the first place?
Unrolling a metaphor into its literal meaning is one of the most annoying features of the "AI voice", IMO
I agree that the boundaries may be blurred in practice, but I still think that there is distinction.
> visible, public interface
Visible to whom? A class can have public methods available to other classes, a module can have public members available to other modules, a service can have public API that other services can call through network etc
I think that the difference is the level of abstraction we operate on:
unit -> functional -> integration -> e2e
Unit is the lowest level of abstraction and e2e is the highest.
Taken to extreme this would mean getting rid of unit tests altogether in favor of functional and/or end-to-end testing. Which is... a strategy. I don't know if it is a good or bad strategy, but I can see it being viable for some projects.
It became very hard to understand what exactly is sent to LLM as input/context and how exactly is the output processed.
As a non-native English speaker leaving in a non English-speaking country I thought about this too, but in a more selfish and practical manner: what if my English in particular converges towards AI mannerisms?
You see, if you live in an English speaking country and your family speaks English you still have some amount of guaranteed-to-be-human language input from taking to people face-to-face in real life.
But for me, 99.9% of English input I receive is online. So I wonder, how much of it is already AI and how much has the non-artificial neural network inside my brain has retrained itself to mimick AI.
This is scary, because before I was absolutely sure that consuming content online improves my ability to understand and use English. Now I'm not so sure anymore.
eternal AItember
I think that if you actually try reading someone else's conversation with LLM, you'll find out that it's less exciting than it seems.
For the one who has the conversation the excitement comes mostly from the ability to steer it the way you want. Reader doesn't have this ability, so they are just forced to endure the excessive wordiness, that is so typical for most LLMs.
If you learned something interesting, then why not express this knowledge in a normal article/blogpost? What advantage does a conversation between you and LLM has over just a normal text or, perhaps, text with pictures, diagrams, maybe some interactive illustrations etc
Looks cool, but how exactly do you gather proven-to-be human comments?
I think it would be better if you used pre-ChatGPT (Nov 30 2022, I think?) stories.
delve into noteworthy realm
leverage tapestry