Last modified date: 2024-06-20 09:24 Commit: 16 --- please drop anonymous feedback, very helpful: https://www.admonymous.co/louis030195 i try to follow this: [[Paul Graham - How to Get Startup Ideas]] e.g. focus on problems not solutions, but might do a few mistakes, happens ### problems **fundamental b2c problems**: - **health** (people want to feel better, or become immortal lol) - **wealth** (people want more time, money, wealth, and freedom) - **relationships** (people want to spend quality time with their friend, family, loved ones, have better relationships) **fundamental b2b/e problems**: - **computation** (the universe is a computer): turn energy into computation most efficiently - **energy** (computation require energy): collect energy and distribute it most efficiently **narrow problems** just brain dumping everything coming to my mind (careful, could be not real problem or becoming a solution) - maintaining CRM - #edtech AI education (how to use ChatGPT to its full potential) -> especially for people older than 40+ and outside tech hubs - atm what ppl do: slack channel "share your ai usage". mostly experts in it and learner (ppl that are bad at ai) dont care - using LLM in disconnected environments -> especially in defense, maybe other idk - chatgpt data privacy -> not a strong need though. people leak their data if the value is high (just as ppl will leak to apple intelligence soon) - search is always a problem but in new segment of some market (google, algolia, jina.ai, etc.) - matchmaking -> founders/or intellectuals struggle to find relevant people - preserving companies knowledge (e.g. 10 founding members know how the company went from 0 to 1 and when they leave it creates a huge gap) - (based on some Andreessen video) - dev tools - i dont want to use closed source devtools in general (prefer to use something worse that is open than something closed, better) or would require a huge value difference (too much lock in) - i dont want to use trendy tools (in the past stars meant quality, now it means that it's going to die in 3 months) - clearly need for some tools that allow people to write code that do more in less code "AI agents" (dont like the term, basically glue between LLM and the web/traditional software) (autogpt, langchain, etc etc, devin, all fails) - do more with less: new interfaces, new iphone (humane, etc.) all fails - loneliness: pattern with ai girlfriend c.ai etc. covid. tend to be tar pit - plagiarism, copyright stuff related to gen ai - homeless / poor ppl -> [[060124 2028 idea dev agency powered by homeless or poor ppl]] - strong desire for open-source and self-hostable alternatives to mainstream, often costly, software solutions. (seen from producthunt, github trends) - supabase, posthog, etc. - Information Overload and Disorganization - pkm Integration of Visual and Textual Data - turning chaotic ideas put on a canvas into a clean presentation (my problem - done that for like 100 excalidraw canvas) - breaking dunbar number - dealing with 150+ humans on a daily basis etc - getting feedback on social media posts before posting (quickly) (my problem) - clicks in audio recording for newbie content creators (nothing online to remove it in less than 1 min) - using LLM without leaking data to openai, etc. - collecting analytics across multiple social channels effortlessly (insta, x, linkedin, substack, tiktok, etc. my problem) - building "fake" content fast (example i do: say go to chatgpt and edit the html to take a screenshot and share on social, or use my own website edit html manually etc. current use case: fake a Slack message from a bot by editing html manually, i wanted to showcase a slack bot interaction that i didnt code yet) (solution: LLM that go to website and edit as you want, maybe chrome ext. and faciliate turning into pretty screenshot like pika.art ...) - cafe in EU often dont allow laptops on weekends or forbidden between 12-2pm - my reasoning is that laptop ppl basically "buy" a 6 hours+ coworking space for the price of a coffee and basically coffee places get more value from non-laptop customer who spend more ### ideas - #coss something COSS for simplifying usage of LLM at the edge for: - oil drilling - spacecrafts / deep space - defense - reduce latency / cost / etc. - #observability no code/low code software observability routing platform. basically a product my team built when i worked in cloud industry. we were collecting billion of data points from international data centers and ops teams could use this to decide which alert, in which condition they wanted. a bit like sentry.io and highlight.io but what was interesting what that it was in a kind of python language, so allowing more advanced usage than sentry and highlight. i could see this with LLM like "plz when router xZ63 temperature is above 65 degree send a message to bob the ops manager nb 42" https://www.ycombinator.com/rfs#developer-tools-inspired-by-existing-internal-tools - [[060124 94026 idea screen to action]] - #neurotech #bci ouraring for the brain - #neurotech #bci eeg/fnirs brain computer interfaces for new use cases than neurosity, muse, prophetic.ai, etc. - #spacetech #defensetech space/defense (interested but cannot GTM due to limited to US citizen) https://www.ycombinator.com/rfs#new-defense-technology (talked to 100+ ppl working in defense in the past few weeks) - LLM for situational awareness & decision making at the edge - LLM for Wargaming (e.g. Starcraft UI for defense haha) - LLM for PKM - private chatgpt for defense back office https://www.ycombinator.com/rfs#llms-for-manual-back-office-processes-in-legacy-enterprises -> LP https://enterprise-private-chatgpt.vercel.app/ - explainable AI for defense https://www.ycombinator.com/rfs#explainable-ai - Berkshire Hathaway for tech - #oss #devtool #llm OSS LLM devtools https://www.ycombinator.com/rfs#commercial-open-source-companies (talked to 100+ LLM devs in past few years) - problem theories: - 90% of world's software still does not use LLM despite the gain in development time saved - LLM cannot use traditional software autonomously easily (lack of tooling) - lack of tooling for code generation + execution - lack of tooling for autonomous api calls - lack of tooling for autonomous UI headless interactions - lack of tooling for rag - lack of generative UI tooling (esp. on iOS, Android, etc.) (parallel to vercel/ai lib) - LLM not fast enough - hard to use for edge computing - LLM hard to use at the edge (lack of infra, etc.) - lack of education on LLM in europe & other third world countries - #content-creation i am lazy content creator, id pay for a tool that allow me brain dumping few liens of text daily (<5 min effort) and generate a video meme tiktok style (whatever that attract attention/optimised for algorithms with low effort for me) - new interfaces / new framework for the web - e.g. new nextjs for canvas based apps. the world is moving towards everything generative ui + canvas based (excalidraw framework thing) - #sales something for sales check [[screen to crm]] - #ai-hardware new AI hardware (0 insight) - physics based digital twin / the matrix https://www.ycombinator.com/rfs#using-machine-learning-to-simulate-the-physical-world - i love this code https://github.com/dimforge/rapier. imagine using this w LLM for simulation in space/defense, other industries - simple POC -> vercel/ai-chatbot + rapier = "yo chatgpt make a simulation of aircraft flying at 865 km/h above africa" idk :D - #llm #devtool Llm lib to build integration autonomously automatically. As npm package  - #canvas Excalidraw personal website  - #website-builder Give ur name and prompt and gen website - Share idea quickly  - #search Perplexity but in excalidraw - #ai-organization chaotic ideas to presentation -> https://framer-cursor-excalidraw.vercel.app/ - #human-crm human crm in infinite canvas -> shipped v0 https://relforce.vercel.app/ - #edtech a platform for -> shipped https://look-how-i-use-ai.vercel.app/ 1. people that lack AI education "how do i best use chatgpt?" and can pay to see video screen recording of AI nerds (say SF AI engineers) 2. AI nerds that record their screen 24/7 while working and sell on the platform - more local AI apps. basically LLMs unlock a new kind of apps that leverage very sensitive data #local-ai - software is becoming more and more personalised #generative-ui - infra to build private cloud AI like Apple Intelligence but for any use case, on promise - proxy to jailbreak woke LLMs like chatgpt (Thales built this internally to be able to ask chatgpt "how do i build a bomb?") - internet time machine for you social branding: bot that take screenshots of your social profiles daily and add to excalidraw, organised, so you can see in infinite canvas how you social branding evolve over time (ideally w some analytics too) - statistics shows people spend most of their time looking at their own profile - screen pipe https://github.com/louis030195/screen-pipe - ### newly funded startups just a list of startups that raised recently that come to my mind (can be hiding underlying problem) $9m withmartian.com -> llm cost $17m patronus.ai -> ? langfuse -> llmops $25m langchain / langsmith -> llmops $9m unify.ai -> llm cost $30m extropic.ai -> computation -forgot name- $7m - aggregator of LLM inference cloud or somthing -forgot name- -> pushing gpus in orbit, orbital data centers #idea #ideas