Hacker Newsnew | past | comments | ask | show | jobs | submit | ByteAtATime's commentslogin

To quote the guidelines[1]: > Please don't post insinuations about astroturfing, shilling, brigading, foreign agents, and the like. It degrades discussion and is usually mistaken. If you're worried about abuse, email hn@ycombinator.com and we'll look at the data.

Personally, I thought this was a decently entertaining idea and was a nice break from AI hype and other slightly depressing news.

Also, HN is generally very random - some things succeed and others don't. While I understand the sentiment, imo your message came off a bit abrasive haha.

[1] https://news.ycombinator.com/newsguidelines.html


Interestingly, AA[0] reports a 59% hallucination rate on AA-Omniscience, one of the highest among all models and quite a bit higher than even Opus

[0] https://artificialanalysis.ai/evaluations/omniscience


The price numbers look weirdly specific, I wonder where they came from.

Personally, the thing I'm most excited about is the FrontierCode 2.0 score, which is a benchmark I trust a bit more - 31 seems to be roughly on par with 6.1 Sol and Opus 5.5 medium


I have essentially no knowledge in this field, but another concern may be that if the model has minimal cybersecurity knowledge, it may also implement insecure systems as a side effect.


My initial thoughts too, it's going to write insecure software. I recall early LLMs having problems with trivial stuff like SQL injection.

Definitely going to give this one a skip.


> LLMs need an insane amount of text. Dessa learned language from hundreds of times fewer words (in fairness, LLMs also have to learn C++ and whatever the hell Angular is).

Now that a larger proportion of text is AI, I'm appreciating human-written text more and more; this got a good chuckle out of me.


>and whatever the hell Angular is

I've definitely seen LLMs compose jokes with this structure a lot where "whatever the hell ___ is" is in parenthesis. Not saying this is AI generated tho


I wouldn’t be so sure that that’s human text

I can reliably get my custom models to output a 95% perfect replication of my tone and voice


As somebody who attempted that recently, I seriously doubt that percentage :D


Maybe you just have a voice that's, shall we say, in sparse latent space? ;)


New insult dropped!


I have an insane amount of writing output as training data going back a loooong time so it all comes down to how much training data do you have


Many of us can do it on just our hacker news comments (been there, done that).


What do you mean by your custom models? Do you mean custom prompting?


Couple things:

I have a “style guide” markdown that is pretty expansive

I run an RLHF style loop with me in the middle for all my writing now where I give a voice rant that is over expressive (aka redundant samples of concepts that I want to express clearly) in order to manage lossy compression steps

Then I will manually clean up that last 5% after a half dozen rounds

That sequence goes into the style guide to be referenced as a reasoning chain of thought to map rant > output

It’s been three years of tuning I’d suspect at this point


How did you train your style guide, though?

I have a somewhat expansive style guide, and it still reeks of Claude.

My cleanup is much closer to 40%, which still makes it worth the effort, but sometimes when speed matters, I'll apologize for the slop.


I've recently been using gemini argon as a voice judge. so it goes claude ( with style guide) -> voice judge with argon -> claude adjusts. I do that in a few loops before it hard stops if it can't reconcile, and pushes to me for review.


Guessing - finetuned? Or LoRA?

EDIT: Nevermind, see below. I would definitely go with a LoRA though.


That’s if I was going to build my own model on myself, which I’m not planning on doing at this point


Inconceivable!


Super interesting read; I especially found this letter from "P791" fascinating:

> Hi, Wonyoung, I map North Korea to help North Koreans.I select areas that look like they might be good points to escape. In particular, the detailed mapping around Phungso-ri (풍서리) is almost entirely my work.I mostly map along the Tumen River, both on the North Korean and Chinese sides. It is important to map the fences, roads, and towns on the Chinese side.


Once in China, however, aren't escapees routinely exploited by organized crime?


They chose to cross anyway knowing the risks, and it's still preferable to death by firing squad.

"Beyond Utopia" is a harrowing documentary about a South Korean pastor who helps people escape NK, filmed on the trail side by by side with along some actual refugees. It illustrates how complex, dangerous and utterly corrupt the process of getting out is, and also mentions the role of local gangs: they detain refugees and then do a bidding war between the authorities and relatives to extract every penny.


Interesting - I wonder if it's because coding doesn't use the specific token probabilities, while decision models do


Yes this is the reason. Transformations like quantisation preserve the rank of outcomes much better than probability mass.


FWIW, my tests used only ranking. I used Kev to drive E2E test runs using agent-device, where Kev had to decide the next action - like pressing a button, or scrolling down etc, to reach a certain goal (UI state). It went surprisingly well, but 4B BF16 was significantly better than 8B Q8. The latter made some wrong decisions in every run while the former was almost at Luna precision.


Fascinating insight. I'm curious, how do you know this?


I think it also helps that speed and latency aren't as vital for coding and we can afford to let the model think for longer.


I think, considering the size of this model, it's closer to a Pro than a Flash on everything other than speed


Fair point! I've actually thought about that before; I've tried to be extremely clear in the project's README with a disclaimer that this is a non-commercial hobby project and is not affiliated with the official Raycast team in any way.

The name is just for identification, as the project's goal is to be a compatible, open-source alternative for the Linux community, a platform they don't currently serve.

That being said, I'll definitely keep it in mind. Thanks for bringing it up!


That shouldn't matter. They might even have a Linux version in the works, given they are doing Windows now. They are within their rights to ask you to get rid of it. You also seem to be using their brand assets like the logo. I get your intention and love what you were able to accomplish but if you intend for this to grow, you are better off doing this now.


> They are within their rights to ask you to get rid of it.

Only the logo and name. The functionality side should be perfectly fine, there are oodles of precedent for reversing a workable program from someone's commercial API: https://en.wikipedia.org/wiki/Google_LLC_v._Oracle_America,_....


True, but a cease & desist will likely include the demand that they cease all operations, and they might be willing to fight that in court, possibly seek damages, etc however likely or unlikely it is that it would hold up. They can pay for a few hours of a lawyer's time to intimidate the OP.


But prolly can still go after if you resemble it closely enough in terms of looks? I get the API point though, thanks.


Eh, probably not. There are any number of things that look like it. The appearance is the natural outcome of its functionality.


It should matter that it's a non-commercial hobby project. It's nice to see the light pushback.


They use react-native, no? One of the main issues is that all react-native implementations are half baked and dead.


The launcher UI is AppKit while the extension APIs allow third-party developers to write them in TS and a React subset. They have a custom renderer for React that turns that into AppKit primatives. So similar in concept but custom and specific to their usecase.


That makes sense -- I'll go figure something out, thanks for the explanation!


“Just for identification” is the point — the name Raycast identifies their product, and it’s their trademark. You can’t use it without permission, even with a disclaimer in the readme.


> The name is just for identification

That’s precisely what they’d dislike about it. You’d be creating “brand confusion” by using their trademark in your own name. You couldn’t make “My Cola (Recreating Coke)” without getting an expensive and inconvenient letter from their legal team.


It's worse than that -- "Recreating Raycast" is just the blog post title, the project is called "raycast-linux".


It's also worse than that. It uses the Raycast logo directly in the launcher itself. Which is odd because just above this, OP says:

"I've actually thought about that before; I've tried to be extremely clear in the project's README with a disclaimer that this is a non-commercial hobby project and is not affiliated with the official Raycast team in any way."

Clearly a bright kid, but that's quite a fumble. Among my ideas for being extremely clear about not being affiliated with Raycast I would have to say using their name and using their logo together would be the worst way to communicate that.


Oh dear. Yeah, that’s guaranteed to end badly.

OP: please do change this ASAP so that the Raycast gang doesn’t protest the really neat project you’ve made!


I've had to look into for my own side projects, you don't want to include brand names, trademarks, etc in your own projects' names and you should be very wary about how you use them in your marketing if you choose to do so at all.


I propose cast-a-ray (sounds like castaway and goes along with FOSS wink+tonge-in-cheek style). Or maybe waycast?


I'm sorry, but in my accent (speech impairment?) "cast-a-ray" sounds almost like "castrate", so I can't back your naming idea. :P


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: