maxbaines

Books?

Less and less now actually. Synthetic data is now being used on most new models as the public data runs out.

This is the first thing I do when using LLM's to create code. So we can and I do have LLMs writing documentation.

In my domain IIoT, I even created a yaml structure to make this easy. https://unsframework.com (wip)

Dyson 2.0

pretty fun, I did surpassingly well, apart from a McDonalds museum was within 100miles and 10years.

Fireworks AI is a great option - https://fireworks.ai/models/fireworks/deepseek-r1

This is notable as Game Tests are now available in the game.

I built my own client (llmpad.com) to originally solve this problem, as well as using other LLM's and features. A little surprised others have not done this too? You can try llmpad soon, feel free to message me.

I use cline and cursor, not sure they can be though of as GPT wrappers

I am building llmpad.com yeah you can think of it as a GPT wrapper I think differently. I started building it 18 months ago, actually because I wanted to customise the system prompt and keep my data. Then I wanted to use other AI's Anthropic, Groq etc. Then I wanted to generate images. Then I wanted to work with tools for search, or UI cards. Then I wanted to work with realtime audio api's. Then I wanted to create react components kinda like V0. this week I wanted to work on tasks, that run and chain scheduled prompts, its turning into a little agent. All the time using one UI that looks the same, is tweaked to my liking and has some nice power user features like Shift Arrow on input for history, switching models etc.

I am also sure tomorrow there will be another new thing.

I think of a little like Notepad++ for llm's instead of just Notepad.

I use it hourly, in the morning I have a summery of HN from overnight along with news, and reminders.

I host it on Digital Ocean, it uses about 15-20 API's from AI companies but also web search, document and kv stores etc.

Interested in trying it let me know.

I missed this feature on release, only just found it. Wondered if it's an equivalent to Inspect Element and Del or more protective of things like paywall overlays and ads?

28 Days Later comes to mind.

Knowing of and using Suno to begin I thought nothing of it, then the chorus kicked in. Pretty straight forward to rebalance the weights of the model though given the popularity of the song vs the input lyrics.

Was a similar hack, for sure another game though! I could never handle sports games, although I loved playing them about level 3 or shot put in olympics games I was done for.

Ouch!

Can't remember the game, but removing and inserting the cartridge, enabled some kind of glitch I could use to cheat as a kid.

Interesting article. Blowing into cartridges fixed them many times for me. Although in reality it was likely re-inserting them that maybe did it and also the paragraph about adding moisture being a bad thing is obviously correct, now I wonder if that added moisture actually improved the connectivity on already corroded contacts.

Thanks, this is going to be really useful for me.

It's easy to think of the models in this context, rather than the application built on top of the model, in this case ChatGPT.

ChatGPT is not simply the gpt-4o model, it's the model, a system prompt, tools and a virtual environment running python.

I built my own app, when using it i don't get links to partners because i don't mention such a thing in the system prompt.

Now lets make it work with Linux - I say this slightly tongue in cheek, but I run Linux on my intel surface laptop, always wanted on the surface pro x. This project is doing great work in the area https://github.com/linux-surface/linux-surface

This was previously the Pendant from Rewind, it's been released as a standalone app now and wearable in August.

This comment sums it up for me "Why would anyone think Quora would make something that wasn’t shady? ..."

https://old.reddit.com/r/ChatGPTPro/comments/1b6kzlo/comment...

Yes I found it an odd piece also.

Thanks, this is super useful. Another good resource from DO.

Walkers is Walkers in the UK.

Up until 10 minutes ago I would have agreed, but I was reminded of Elon Musk. https://nitter.net/DataChaz/status/1726987591457910944#m

Regardless I see the API's as remaining stable.

Why would they shut off public access?

Reading on X employees have ensured the uptime during this current situation, and from what I read no employee wants to affect users of what we're building and what they have achieved. Will the company be the same hmm who knows, regardless I see the API's remaining stable Microsoft will make sure of that.

Yes also Azure is a good backup anyway now or anytime, you can fall over to the Azure API with a key change and a url.

what Poe is doing seems odd for such a company. I am doing the same with baarilliant.ai in my spare time.

To train a bespoke LLM takes a lot of effort and compute, you are perhaps better off using Retrieval Augmented Generation (RAG). Here's some information from Langchain

https://js.langchain.com/docs/modules/data_connection/ https://python.langchain.com/docs/modules/data_connection/

Also OpenAi last week released Assistants which is an easy way to achieve RAG without needing new tools such as Vector Db's. Although 5000 docs is perhaps to large for assistants.

The first decision is whether you would use an Open Model such as Llama2 and host that yourself or a Model such as GPT 4 from openAi or Claude2 from Anthropic etc.