ach9l · HackerTrans

That's not the same thing at all. Stadia was never about streaming from your own dedicated pc, it was about cloud computing. Something Valve does not have. In order for Steam Link to become Stadia, Valve requires the same data center capacity to serve games.

ach9l

4 miesiące temudiscuss

have you heard about rlhf?

ach9l

w zeszłym rokudiscuss

oh, i meant humans as teachers. if i had to choose between a human teaching me quantum physics equations or a collection of richard feynman’s lessons turned into agentic lessons, i’d pick the agent. i mean, we can compile the whole collection of lessons in agentic form way faster than any human could deliver the same number of lessons.

the product is exclusively for humans. the teacher, i'm not so sure.

ach9l

w zeszłym rokudiscuss

we should be able to use this as a vscode extension to solve this issue. is there an sdk to integrate this into electron apps?

ach9l

w zeszłym rokudiscuss

yep, this is not for humans, agents with dia-1.6B or anything similar, they will outclass humans at this, really quick. i'd like to work on a poc if you are interested, i train and deploy models for a living.

ach9l

w zeszłym rokudiscuss

instead of ranting, maybe explain how to make a qat q4 work with images in vllm, afaik it is not yet possible

ach9l

w zeszłym rokudiscuss

so the actual implementation of the code to log people off was also hallucination? the enforcement too? all the way to a production environment? is this safe, or just a virtual scape goat?

ach9l

w zeszłym rokudiscuss

i've been testing all models that fit the mac studio 512 gb ever since i got it. previously i was mostly focused on getting tool use and chain of thought fine-tuning for coding, around the size of llama 3.2 11b. but even some distill r1s on llama 3 70b run well on macbooks, although quite slow compared to a regular api call to the closed models.

for mac studios i've found the sweet spot to be the largest gemma, up until llama scout was released, which fits the mac studio best. scout, although faster to generate, takes a while longer to fill in the long context, basically getting the same usability speeds as with the qwq or gemma 27b.

the refactoring is a test driven task that i've programmed to run by itself, think deep research, until it passes the tests or exhausts imposed trial limits. i've wrote it by instructing gemini, r1 and claude. in short, i've made gemini read and document proposals for refactoring, based on the way i code and strict architectural patterns that i find optimal for projects that handle both an engine and some views such as the react.js views that are present in these vscode extensions.

gemini pro gets it really well and has enough context capacity to maintain several different branches of the same codebase with these crazy long files without losing context. once this task is completed, training a smaller model based on the executed actions, (by that i mean all the tool use: diff, insert, replace and most importantly, testing) to perform the refactoring instructions is fairly easy.

ach9l

w zeszłym rokudiscuss

yep, this is the way, i guess. as somebody who has taken this very same exercise of cloning cline using cline for my own cline, a cline that compiles itself, i’ve also learned to steal* things over the years. i’ve seen your extension, but i was reluctant to give it a try just because it looked just like any other clone, but i guess i’ll do the same thing again. i’ve started to see the value when i decided to fork and declutter again, this time roo code. actually i’ve perfected forking cline and derivatives with my own framework. when you know what you’re doing, these tools don’t put you in the flow. vibe coding done right is another level of progress. i’ve got a cs major though, so i’m a bit biased, also helps that i’ve done masters in theoretical computing, theoretical linguistics and machine learning, so i’ve always been attracted to these toys and frameworks, not so much to javascript or web development however. this whole exercise, or should i say automation? now, takes me back to the days i wrote compilers. this is just as fun as code that can compile itself in the end. same shit all over again.

so i gave roo code a try, set a few test cases, and proceeded to declutter, refactor, rewrite the whole thing. i’ve never really written long apps in javascript nor typescript for that matter, and man, i just think 3k lines of code in a single file is just bad code, and i’ve been proven right. 3k lines fucks your context really good. you can’t use cline to code cline because it will ruin you financially one way or another. jesus fuckin’ christ the old cline.ts file was like responsible for the whole damn extension, over 3k lines, the kind of code i would write 10 years ago as an intern. anyway, i’ve added (and learned in the process) react.js components to have an interface to easily collect the data for my own loras. honestly if you are looking to integrate large local models into kilo, i’d love to collaborate. my forks mostly provide data analysis for the fine-tuning of my own personal repositories, using years of commit history as training data, even bash history. i’ve benchmarked several tasks. i can basically fork roo code or cline, declutter it, refactor it, with a gemma or qwq running in a mac studio for a few watts. i’ve been logging everything that i do ever since we were granted api access to gpt3 at a lab i coordinated about 5 years ago. so i’ve mastered the filtering of the completions api, reconstruction of streams, all using airflow and python scripts. i added a couple buttons such as the download task you’ve also added, but more along the lines of “send this to the batch in the datacenter so we train a new gemma” filtering good solutions vs not so good, the old thumbs up thumbs down situation, helps a lot, adding a couple of mcp integrations for applying quick loras locally, plus the addition of test driven development, aiming for reinforcement learning based loras. i built myself a very nice toy, or should i say, i bootstrapped a very nice tool that creates itself? anyway, thanks for sharing this.

i think the next major thing that is gonna happen with these tools is that it gets free at home as new chips become cheaper. llama 4 running in mac studios or dgx stations is as fast as you can get today and it is already good enough (if prepared correctly) to build any yc startup codebase from before covid, or even from before chatgpt, in a weekend. it will definitely happen. i’m wrapping fixing llama4 scout, allow me to mention the fact that it has a tendency to fix bugs by commenting code and adding TODOs, fucking great architecture though, just what we needed, i mean for optimal local development. i’ll try to publish results soon enough, optimized for the top mac studio though, haven’t got a dgx yet. i’ll prepare macbook versions too. the world needs more of this, a cline that fixes itself just on battery power...

ach9l

2 lata temudiscuss

on ai being a niche/hobby at the moment... feels like something a unix greybeard would say about guis in the late 70s...

ach9l

2 lata temudiscuss

you could not say this better than this.

ach9l

2 lata temudiscuss

And you can say "Actually that is not the case with iPad, you can have your own applications there without rooting it." have you heard of Xcode? still... you are missing the point, you can't even use an iPad to properly code for iPad

ach9l

2 lata temudiscuss

why are people surprised they use playwright? how do they imagine a lam would work? in the end an "action" would always have to come as a computer command, be it an api call, or function call generation, an "action" is code to be executed. transformers generate tokens and only tokens, it is up to you to decide how you want the flow of tokens to be. i find the use of playwright quite clever, you could use puppeteer and make it write scripts for web browsing as well, but in the end an action will always come back to computer code, written as text as a human would do. the fact they use playwright does not imply it is not generative ai, on the contrary it is a clever way of showcasing how to configure a transformer to bypass an api lock-in such as the midjourney one.

ach9l

2 lata temudiscuss

you can't automate playwright without a decision making component in front of it, they are definitely using a transformer there. one could train a llama and make it perform triggers to playwright automations. you can even get deep into transformer tokenization and create action tokens and a formal grammar for your generation, build a parser on top of your predict function and have a "lam" working. the fact that they use playwright does not imply it is not generative ai. i'd say it is really hard to do those actions without a transformer involved

ach9l

3 lata temudiscuss

Black bar please! :wq

ach9l

3 lata temudiscuss

I'm on my 1888th day after I changed my way of eating, and I hate when this happens to me, I mean, losing the ketosis because something I ate had too much sugar. It happens in some asian restaurants where they cook salmon with sugar and they don't tell you, in some places they add sugar to most japanese salads / wakame. I get bloated and retain water, I get out of focus easily, and I'm tired for the most part of the day. One day a friend of mine thought he had vodka (which does not kick you out of ketosis) instead he gave me some transparent licor, very sweet. I had the most profound high on sugar of my life, with just one shot, I had a high rush of feelings and pleasure, increased heart rate, I wanted to go out and party. I had nightmares that night, in the dreams I needed to run or escape. It took me 72h to fully recover. That said, I've written the most creative pieces of code, mostly backends in functional programming while on ketosis

ach9l

4 lata temudiscuss

perhaps you could use it yourselves to have a reactive web ui of the landing page?

ach9l

8 lat temudiscuss

The formal grammar structure behind smart assistants is basically the same. Siri is as primitive as Cortana, Google Now and Alexa, the difference comes in the amount of data the interfaces have access too. Outside of the text to speech and speech to text algorithms, there is barely any machine learning when it comes to understanding what the user says. As of it right now comparing smart assistants is equivalent to comparing dictionaries, Siri having less entries and people writing them than the others.

comments