Curious what people have been building with LLMs.
I worked on a chrome extension a few weeks ago that skips sponsorship sections in YouTube videos by reading through the transcript. Also was trying to experiment with an LLM to explain a function call chain across languages (in this case MakeFile, ...
I worked on a chrome extension a few weeks ago that skips sponsorship sections in YouTube videos by reading through the transcript. Also was trying to experiment with an LLM to explain a function call chain across languages (in this case MakeFile, Python, Bash). I've tried running a few telegram bots that are PRE prompted to do certain things like help you with taxes.
What are you building?
What does the stack look like? How do you deploy it?
Comments
I don't like selling. I wanted a way to practice cold calling in a realistic way. I set up a phone number you can call and talk to an AI that simulates sales calls.
I ended up using it for more general purpose things because being able to have a hands-free phone call with an AI turned out to be pretty useful.
It's offline now, but here's the code with all the stack and deployment info: https://github.com/kevingduck/ChatGPT-phone/
Edit: forgot to mention this was all running off a $35 raspberry pi.
So the AI tries to sell to you, or you try to sell to the AI? This sounds very intriguing but I can tell by your README that you're an engineer and not a sales guy - there are no distinct value propositions.
But it sounds damn creative as a project.
The AI answers the call and acts as a potential customer. They take on personas to simulate behaviors like difficult or reluctant customers. You then do your pitch, handle objections, etc. At the end you get a transcript that's 'graded' to show you where you could improve your sales approach.
And you're right, I'm not a sales guy. This project is for people like me who want a risk-free place to learn the basics of sales so that when I do talk to an actual human, I won't panic and freeze up like I always do.
By cabinguy 2024-02-0519:59 I absolutely love this idea.
Most high-level sales people rely on role play partners but that requires a pretty a big commitment. This would make a great product, imo.
Also (tip): Study, memorize and internalize a sales script for your product/service...along with the objection handlers and closing questions. Practice every single day. You'll gain massive confidence because you know exactly what you are going to say, every time.
By tkgally 2024-02-0523:01 > a risk-free place to learn
That's turning out to be a valuable feature of LLMs in many areas. You can practice complex interactions with them without worrying about boring or annoying them. Even the most patient human teacher gets tired eventually. LLMs don't.
By BenoitP 2024-02-0611:45 I'd buy that. I'd buy that for interview preparation as well. Maybe 5$ per hour, up to 15$. I wouldn't buy a subscription, only actual consumption of the service.
Please consider putting it in online.
By teleforce 2024-02-074:30 Love the idea of AI grading the answers, hopefully this can be extended to marking/evaluating/grading subjective manuscripts.
For me there's none is more boring than marking/evaluating/grading manuscripts. I prefer hard labor like gardening or farming than doing that activities although I'm quite good at evaluating stuffs I think.
Can you please elaborate how you do this and based on what metrics/scheme/etc the answers are being evaluated?
By 3abiton 2024-02-069:49 This, for some reason, reminds me of Nathan Fielder rehearsal skits.
By nico 2024-02-0719:13 This would be amazing to do practice code/tech interviews for software engineering roles
It could work both for practice and for automating interviews as well
By spywaregorilla 2024-02-0523:33 do you have any reason to believe the phone calls are realistic?
By anxman 2024-02-066:14 This could be a product. AI sales training.
By satvikpendem 2024-02-0521:201 reply Now you can turn this into an AI sales cold caller based on the data you could collect from how the AI reacts to your selling. That is to say, the entire system becomes a generative adversarial network.
By harvo 2024-02-098:57 [flagged]
Like exposure therapy for people afraid of sales. Very nice idea.
By alentred 2024-02-0521:48 I like the idea very much! Using an LLM as a "sparring partner" for training in various areas. LLMs tend to hallucinate, so I find it harder to use them reliably in the context of decision making. Training however is a nice idea indeed: mistakes are not as critical, just as in real life any peer can make a mistake.
By VoodooJuJu 2024-02-0613:26 Very cool, sounds like a saleable product. I feel like there's already half a dozen landing pages with people trying to sell what you just made in the 18 hours since you've shared it here. That should however be a red flag to those same people, a demonstration in just how easily commoditized LLM products are.
Are you finding response time to be an issue? I can imagine some very long pauses might kill the flow of conversation.
It's not perfect, but it's tolerable, and not unlike some real-world calls where there's a slight delay. There are some "Hmm ..." and "well ..." scripted in as well to make it feels natural if there is a long response.
By splatzone 2024-02-0521:18 I love the scripted filler words, that’s smart
By elicash 2024-02-0520:49 To that point, I would love to hear an audio file of it in action since I see from GitHub the phone number is down.
That's cool. Thanks for sharing the source. What else has it been good at for you?
By duckkg5 2024-02-0519:45 The cold call sales part can be replaced to suit any need. I had another version that was just a generic AI (no sales stuff). I found myself on walks frequently ringing up the chatbot ("Hey siri, call ChatGPT") and just asking it whatever is on my mind. "Tell me about Ghengis Khan" or "where's a good place to catch trout in north Georgia" or "how do I make baked ziti". Makes the walks go by super quickly.
By craigdalton 2024-02-065:07 Would you be willing provide a live demo (via web interface) - as a preludebto providing a similar training bot as a consultant?
By DrNosferatu 2024-02-0523:26 Now do it for dating practice - great for nerds ;)
I helped "writing" a cookbook from my grandmother's recipes. For her 100th birthday, my dad rescued more than 250+ pages of recipes that my Grandma had collected over the years. Some were written in typing machine, others written by hand by her. So, my dad scanned (pictured) all the typed recipes, and "dictated" all the handwritten.
For the dictated recipes, I told him to dictate just "flat" the words and numbers. So that I had paragraphs of recipes.
For the scanned recipes, I used Google OCR (I found out it was the best one quality wise).
For both sets of recipes, I then used GPT4 to "format" the unformatted recipes into well formatted Markdown. It successfully fixed typos and bad OCR from Google.
We then pasted all that well formatted text into a big Google Docs, and added images. Using OpenAI image generation I generated images for each of the 250+ recipes. For some of them I had to manually curate it, given that some of the recipes are for typical Mexican food: For example there's a (delicious) recipe called "PibiPollo" that for the unitiated it may look like a stew, so I had to tell something like "large corn tamale with thick hard crust".
In the end, the book was pretty nice! We distributed digital copies within the family and everybody was amazed :) . I loved spending time doing that.
By syntaxing 2024-02-064:36 This is absolutely awesome. I really want to do the same for my mom’s recipe before it’s too late. Though I wonder what would have happened if you went for GPT-V or LLaVa and the like. I have a hunch you might have been able to skip over the OCR part and straight from picture to markdown? Would be awesome if you can try and compare!
By addandsubtract 2024-02-0610:351 reply Would you mind sharing the cookbook or excerpts from it? I'd love to see it.
I cannot share the full book because a) I don't own the copyright and b) My dad (who ultimately owns it) still has plans to put sell it. And I feel it still requires some editing. But I can share a couple of sample pages:
https://drive.google.com/file/d/1OGE-zfNHHDnALbhgmf3lykBjcSg...
It is in Spanish though.
By addandsubtract 2024-02-089:43 Very cool, thanks for sharing! The pictures also look a lot better than I imagined.
By wonderfuly 2024-02-0611:12 That's great!
My "stack" is just Apple Shortcuts making HTTP POST API calls to OpenAI, which does stuff in MacOS via BetterTouchTool. I trigger each by hotkey or typing a few letter into Spotlight (with Alfred). One transcribes and summarizes whatever youtube URL is highlighted. One does grammar and style correction of whatever is highlighted (and replaces it). One simply replaces the Dictate key with OpenAI Whisper but otherwise works exactly the same as voice typing. It's just way more accurate. One replaces the magnifying glass key to have a voice conversation with ChatGPT (using Microsoft voice synthesis). The built in prompt keeps it's answers short and conversational. It's like asking Siri something, but much better. One simply reduces the highlighted text by ~50% by rewriting it shorter, for when I have typed too much. One gives the key points of whatever article is in the foreground tab, so I know what I'm about to read. One outputs purely code, for example I use my voice to say "javascript alert saying blah" and alert("blah"); will appear at my cursor. Of course, it's usually more complex boilerplate stuff, but it helps speed up my coding. Every time I find myself using an LLM repeatedly for something, I make it into a little Apple Shortcut to streamline it into my workflow, as if it were a built in MacOS feature.
Could you please share the prompt for the grammar and style correction shortcut? I've just started using it for the same purpose, but I haven't been able to find a prompt that yields consistent results. Sometimes, ChatGPT completely changes the style of my text.
By geor9e 2024-02-071:55 I use role: system, temperature 0.7, prompt: Fix the spelling, grammar, punctuation, order, and sentence structure. It's does sometimes change the style too much, but not often enough to annoy me into fiddling with it.
Have you tried Raycast? It has all the AI scripts you mentioned and many more. And many things done better, like showing diff before inserting grammar-corrected text.
By rsanek 2024-02-1016:17 Looks expensive at $20/month if you want GPT-4
I have looking for a way to do "push to record audio" (instead of Mac's dictate) for ages, thanks for the push to look at Shortcuts!
Are you using the "Record Audio" action or something else? Ideally the shortcut would stop listening after a pause like the native Dictate feature does it. At a minimum Record Audio seems to require hitting spacebar to stop - not great but not terrible.
By geor9e 2024-02-070:36 Yes, "Record audio". BetterTouchTool launches the shortcut on keydown, then clicks the Stop button on keyup.
I really love that, super good ideas. I also generally love to create workflow optimizations. Will probably create some of your stuff for myself (I especially like the dictation replacement, could be super useful to me).
Wondering: How big is you monthly OpenAI bill when using all these tools? Only a few $$$, or is it higher?
By geor9e 2024-02-072:00 Only a few dollars a month
By tomcam 2024-02-061:39 You beast! They all sound awesome!
These sound amazing, if you don’t mind sharing somehow, I’d love to see how these work. I’ve never used shortcuts, but I think you’ve inspired me to try.
By geor9e 2024-02-072:06 I put couple screenshots here https://news.ycombinator.com/item?id=39283515 to show the API call part. The rest is just whatever you want it to feed into in Shortcuts. For launching a shortcut on keydown, and clicking out of it on keyup, I used BetterTouchTool like this https://i.imgur.com/sqJ7cOc.png
Would love to learn more details about your setup! I use BetterTouchTool, too and wonder how to make use of it + shortcuts + the API
By geor9e 2024-02-071:52 You might want BetterTouchTool too, since it adds things like cut and paste, to Apple Shortcuts All Actions list. I also use it as the initial trigger usually, to make a hotkey launch a Shortcut. Whisper looks like this https://i.imgur.com/ApAwf2E.png and ChatGPT looks like this https://i.imgur.com/g9f9ZDH.png .
By geor9e 2024-02-094:26 I have not. It looks like it just does some things Alfred was already doing 10 years earlier, and some ChatGPT. Seems like everything has ChatGPT integration these days. I don't think it's compelling enough for me to try tho personally, but I only skimmed the home page. The $8/mo subscription model turned me off a bit.
By nbbaier 2024-02-0719:47 What are the settings and prompt you use for the youtube one?
