I genuinely think it's part of a psyop. If we bloat all codebases and eventually start printing the models on chips to reduce inference costs by 50-100x they'll take in massive profits from 5M line codebases instead of 350k
Proof they don't nerf it only after testing that the benchmarks there stay the same? So overall performance degrades but they isolate those benchmarks?
Ok so in a situation like regular orchestration you would essentially layout all possible steps the LLM can take in your code in a big orchestration layer, and if it hits the sensitive endpoint the orchestration that can occur past that will block off web search. In the design that is. But for something like a manus style agent where you're outsourcing all the work but allowing it to hit your MCP it just becomes a regular API the LLM can call
I feel like I don't fully understand mcp. I've done research on it but I definitely couldn't explain it. I get lost on the fact that to my knowledge it's a server with API endpoints that are well defined into a json schema then sent the to LLM and the LLM parses that and decides which endpoints to hit (I'm aware some llms use smart calling now so they load the tool name and description but nothing else until it's called). How exactly are you doing the process of stopping the LLM from using web search after it hits a certain endpoint in your MCP server? Or is this referring strictly to when you own the whole workflow where you can then deny websearch capabilities on the next LLM step?
Are there any good docs youve liked to learn about it, or good open source projects you used to get familiar? I would like to learn more
i was at the cvs right next to the extra storage when the helicopters showed up and all the police it was kinda nuts to be so close to an event like this.
i have not written about it due to me being a silent partner in the company and dont want my name publicly attatched to it, but the code and prompts is all more like talking to a buddy is how i use it, i ask it to build specific things then i look through and make changes. For instance a few examples i can give is there is a lot of graph traversal in my data i built, I'm not an expert on graph traversal, so I researched what would be a good algo for my type of data, and then utilized claude to implement the papers algorithm into my code and data structures. I dont have the llm in any steps that the customer interact with (there is some fuzzy stuff but nothing consistently run) but i would say an llm has touched over 90% of the code i wrote. its just an upgraded rubber ducky to me.
If i wasn't experienced in computer science this would all fall apart however i do have to fix almost all the code, but spending 10 mins fixing something is better than 3 days figuring it out in the first place (again this might be more unique to my coding and learning style)
i mean i can state that i built a company wihtin the last year where id say 95% of my code involved using an LLM. I am an experienced dev so yes it makes mistakes and it requires my expertise to be sure the code works and to fix subtle bugs; however, i built this company me and 2 others in about 7 months for what wouldve easily taken me 3 years without the aid of LLMs. Is that an indictment of my ability? maybe, but we are doing quite well for ourselves at 3M arr already on only 200k expense.
to me this is the only important one. Not only can they subtely influence the entire us culture, if they were to get in trouble for it, then what? the US doesnt have any influence over them we would just ban them and at that point its too late. realistically it already is too late. a huge point imo aswell is we ARENT at war right now, but if we are at war the amount of information china can both push and obtain through tiktok would be large enough to change the tides of a war
the original chunk is most likely stored with it in referential format such as an id in the metadata to pull from a DB or something along those lines. I do exactly what he does aswell and i have an Id metadata value that does exactly that pointing to an id in a DB which holds the text chunks and their respective metadata
it is absurdly easy to get a 5090 on launch. ive gotten their flagship from their website FE every single launch without fail. from 2080 to 3090 to 4090
the irony in you blasting all over this thread is that you dont know how it even works. You have 0 idea if their claims of scaling linearly are causing bottlenecks in other places as you state, if you read actual docs on this its clear that the actaul "compiler" part of the compiler was put on the backburner while the parallellization was figured out and as that is now done a bunch of optimizations will come in the next year
people like you are the problem. the people who join a website cause it to be shitty, then leave and start the process at a new website. Reddit didnt become shit because of Reddit it became shit because of people going on there commenting as if they themselves are an LLM repeating enshittification over and over and trying to say the big buzzword first so they get to the top denying any real conversation.
in this case we had 1.5 millioon ground truths for our testing purposes. we now have run it over 10 million, but i didnt want to claim it had 0 hallucinations on those as technically we cant say for sure, but considering the hallucination rate was 0% for 1.5 million when compared to ground truths im fairly confident.
At my job we are scraping using LLMs. For a 10M sector of the company. GPT4 turbo has never not once out of 1.5 million API requests hallucinated. We however use it to parse data and interpret it from webpages, this is something you wouldn't be able to do with a regular scraper. Not well atleast.
The model.pth is a custom LSTM for detecting phonetic similarity, as long as you're running it from the pythons folder ( I didn't manage file location very well) it should work.
would you prefer if the timestamp was hidden since it takes up a bigass portion of the screen or that being an option to hide it in the extension settings?
you are correct, originally youtube didnt have this when i made it in 2019 with deepspeech, now they do but i just always preferred the idea of it being on the timebar to just click and go right to it. tbh i should just make a simple addon to take the youtube timestamps and slap it onto the timebar. also for the split chunks this would have no problem there as the transcript is actually stored in a json file, so any concurrent words will always be matchable for phrases. ofc downside being you need to run the model lol
yeah I remember the whole transcript youtube coming out a yearish after i made the first version of this in 2019, but i still perfer the timebar highlighting, but thats just a preference thing
ahh never really looked cause i built my original one in 2019 off of Deepspeech haha just updated it for fun mostly. I know youtube captions themselves are good, but one thing on his code would be that not all videos have captions. Since mine actually downloads the audio and runs it, it would still have values on those older videos that never got captions
yeah a common misconception people think because the input is the same they forget that their is a pre attention linear transofrmation for q k and v (using the decoder only version obv v is diff with encoder decoder bert style)
eventually we will reach the limits of what can be discovered with physics. the same applies here. eventually the limit of what a human can improve on through a job will be done. is this that point? idk but there will be one time that "new jobs" arent made
that is true and isnt. GPT4 has shown itself to halfway through a answer say "wait thats not correct im sorry let me fix that" and then correct itself. For example it stated a number was prime and why, and when showing the steps found it was divisible by 3 and said "oh i made a mistake it actually isnt prime"