Best approach for automating human-like messaging:
const chatHistory = [
"hey everyone!",
"hihi :)",
"hey guys",
"when does it start?",
"Hey everyone, stream starts in 5!",
"9PM PST",
"it starts at 9",
"9",
"9 pacific",
"ok thx",
"hey chat",
"sup pearl",
"big! long time :)",
"indeed :)",
"sup chat",
"hey guys does anyone know when it starts?",
];
const prompt = `The chat history is ${chatHistory.join('\n')}. Predict the next message, don't include any other details in your response.`;
Works well with:
- live chat
- project updates
- invoice notes
- daily messages
Whatever your array of training messages looks like, your next automated message will resemble, it may even answer questions, etc.
It doesn't have to be chat, it could be a history of daily standup updates, predicting the next, etc.
Once you have your predicted next message/action/whatever, now you write normal non-AI code to do it.
Example:
1. here's my target user's past Tweets
2. predict the next one
3. use Twitter API to post it
The using of the API is external to LLMs, you are only using the LLM for the isolated purpose of automating language, or JSON, or some action, etc.
Hope this helps.
There are a million ways - I do this more than anything.
Same weirdness vibe and formating issue too. Ok this is a new account too but I am on another device and did not find my creds on first try. I bet on the advert intention too.
It’s used in order to prevent redundant rerender to child components.
When passing to a child component a function you’ve created inside functional component it’s get recreated because JS doesn’t keep references to a function created inside a function, thus making your child component rerender because a prop has changed.
This example actually really corresponds with this post comments, the need for useCallback have only emerged after the move from class components, because private methods you create in a class do not get recreated at each render
Anonymous functions are not primitive values, so JavaScript passes them as pointers. React uses strict equality when comparing props, so non-primitive values are compared with pointer equality. The useCallback hook allows one to cache anonymous functions in order to pass stable pointers as props and avoid re-renders from pointers changing. It's the exact same motivation as useMemo for passing props with "stable identity".
Prevents some re-renders when passing callbacks down the render chain, but the core flaw in modern day React is that this is added to the developer's list of things to worry about (same as memoization) and manage with the dependency array. It just feels like a lot of things that the framework should deal with and abstract away.
Yeah that's crazy that function pointers cause re-renders in React, I had no idea about this. My go-to reactivity library is SolidJS right now though, their fine-grained reactivity is really nice to work with.
Background tasks that involve human inputs that you could automate:
- Email inbound/outbound
- Social media
- Invoicing
- Code analysis/generation
- Video editor
For one project, trimming and editing down video is very time consuming. An agent works all day in the background - it takes hours but I don't have to think about it, I get an email when it's done. I only think about getting the overall footage now.
But my most used "agent" is a job that runs every hour in Node, consuming news from a list of APIs and then routing to a variety of other things based on what the news category is. Sometimes it posts to one of my LinkedIn pages, or it might email me some links, or upgrade my codebase and create a PR.
I built a personal Salesforce-like experience for my wife's business and pretty much every input she ever has to fill out for an invoice or status update is autocompleted with an LLM. She can regenerate/edit of course. Part of this involves agentic flows, like running CoT on previous invoices and combining live data to generate a new invoice.
With agents, you become an approver more than a doer.
It helps a lot that I run essentially an in-home data center with all on-prem LLM infra :)
PS I have no idea who OP is or what his grift is lmao just answering
Agree, but philosophy aside having essentially ChatGPT plugins that integrate with external services would still be useful for the ChatGPT user.
It will probably be like Slack integrations, but for your AI chat instead.
Example: An MCP for Gmail, so the LLM can check and send email in addition to just generating it.
That's what I think it will end up being. You are right that the developer is still narrowly using the LLM for generative language tasks while most of the other stuff is regular software dev.
Yes, people have brought up my contributions to Brave and LlamaIndex in interviews and it was a nice talking point to start from.
It doesn't guarantee you anything - you still have to describe what you did or how it's relevant.
In the case of LlamaIndex I made like a 2 character change to their TS repo then added some sentences to their README to reflect it. Where in the case of Brave I built entire features.
But, funnily, they only care about seeing that logo. It's as superficial as having a known logo on your LinkedIn IMO. And means about as much. Some jobs I've added a lot of value, others I felt like I did hardly anything, it was just meetings and process. Yet the logo is on everyone's profile just the same.
Idk, it helps. Go fix a typo xD
Oh and when you fix the typo, here's the hard part: Getting your PR merged. Hope you like brown nosing on Discord
1,185 comments
[ 0.94 ms ] story [ 590 ms ] thread- live chat
- project updates
- invoice notes
- daily messages
Whatever your array of training messages looks like, your next automated message will resemble, it may even answer questions, etc.
It doesn't have to be chat, it could be a history of daily standup updates, predicting the next, etc.
Once you have your predicted next message/action/whatever, now you write normal non-AI code to do it.
Example:
1. here's my target user's past Tweets
2. predict the next one
3. use Twitter API to post it
The using of the API is external to LLMs, you are only using the LLM for the isolated purpose of automating language, or JSON, or some action, etc.
Hope this helps.
There are a million ways - I do this more than anything.
And there are 2 new accounts commentating on it, just smells off.
Just left a code screen where I have never seen more useCallback in my life. I hope it's ok that I deleted it and did it the right way xD We'll see.
I pretty much useState/useEffect everything. Between fetch and those 2... you can make essentially any front-end.
Come across the occasional React.memo() I'm like BRO.
The FE data layers are:
• useState
• await response.json()
• const CONSTANTS_DEFINED_AT_THE_TOP_OF_THE_FILE
• process.env
Be a normal guy.
This example actually really corresponds with this post comments, the need for useCallback have only emerged after the move from class components, because private methods you create in a class do not get recreated at each render
- Email inbound/outbound
- Social media
- Invoicing
- Code analysis/generation
- Video editor
For one project, trimming and editing down video is very time consuming. An agent works all day in the background - it takes hours but I don't have to think about it, I get an email when it's done. I only think about getting the overall footage now.
But my most used "agent" is a job that runs every hour in Node, consuming news from a list of APIs and then routing to a variety of other things based on what the news category is. Sometimes it posts to one of my LinkedIn pages, or it might email me some links, or upgrade my codebase and create a PR.
I built a personal Salesforce-like experience for my wife's business and pretty much every input she ever has to fill out for an invoice or status update is autocompleted with an LLM. She can regenerate/edit of course. Part of this involves agentic flows, like running CoT on previous invoices and combining live data to generate a new invoice.
With agents, you become an approver more than a doer.
It helps a lot that I run essentially an in-home data center with all on-prem LLM infra :)
PS I have no idea who OP is or what his grift is lmao just answering
SoftBank x OpenAI have partnered correct?
It will probably be like Slack integrations, but for your AI chat instead.
Example: An MCP for Gmail, so the LLM can check and send email in addition to just generating it.
That's what I think it will end up being. You are right that the developer is still narrowly using the LLM for generative language tasks while most of the other stuff is regular software dev.
It doesn't guarantee you anything - you still have to describe what you did or how it's relevant.
In the case of LlamaIndex I made like a 2 character change to their TS repo then added some sentences to their README to reflect it. Where in the case of Brave I built entire features.
But, funnily, they only care about seeing that logo. It's as superficial as having a known logo on your LinkedIn IMO. And means about as much. Some jobs I've added a lot of value, others I felt like I did hardly anything, it was just meetings and process. Yet the logo is on everyone's profile just the same.
Idk, it helps. Go fix a typo xD
Oh and when you fix the typo, here's the hard part: Getting your PR merged. Hope you like brown nosing on Discord