Taayjus (@taayjus) profile picture

Sometimes I post

San Francisco
Joined March 2023
Photo by Taayjus (@taayjus)Photo by Taayjus (@taayjus)Photo by Taayjus (@taayjus)Photo by Taayjus (@taayjus)Photo by Taayjus (@taayjus)Photo by Taayjus (@taayjus)Photo by Taayjus (@taayjus)Photo by Taayjus (@taayjus)Photo by Taayjus (@taayjus)Photo by Taayjus (@taayjus)
To celebrate @cluely's $15M round by @a16z, we're launching the same thing... for free! Oh, actually, we're completely open-sourcing it too. Good for them on the marketing front. In fact, despite the criticisms, they're pushing boundaries for one thing. I had always thought of overlays but never implemented them, so it definitely stirred me to act. So now's the time for us to push further too :) The core logic is the same as Cluely's: - Take a screenshot via the Apple SDK - Transcribe audio in parallel (not added yet) - Feed it to GPT - Stream it back We went even a step beyond and figured out how to do instant OCR in 200ms to generate even better responses. Screen parsing + image. I haven't added audio yet, as I don't attend meetings and have been busy building something else. However, I'm sure that if someone can Venmo me $300, we can create an open-source bounty to add it in. I'll add the links in the comment below, check them out, and I'm pumped to be alive at this time in history.
11
8
2
35
58,414
I lost the password on this acc and then tried resetting and then viola, X, gave me this I guess in the meantime pls follow @tejasthesharma, I have written the pass down on my butt this time .-.
1
4
427
That or someone is out to get me... I wonder who
1
315
Can open source really win? Not sure if we should go all in on building a great product and keeping it free VS maintaining open source
4
376
Woa 429 upvotes on @Reddit , the powaaar of nerds!
4
352
while cluely spend that $15m on marketing vids, this random guy on reddit built and open sourced the exact same product
114
286
62
7,283
571,268
The random guy is me!
while cluely spend that $15m on marketing vids, this random guy on reddit built and open sourced the exact same product
6
348
Replying to @taayjus @cluely @a16z
I should've worded the post better, the first line was satire .-.
2
199
Replying to @taayjus @cluely @a16z
And as mentioned, here's the open source: github.com/Tej-Sharma/horizo… or the direct link to download: onhorizon.ai/
2
294
We're only halfway through June, and Constella is doing $10k per month now with 0 marketing. No UGC. No Viral X Posts. No nothing. Here's what I did as well as my mistakes: Going well: - Honing in on my personal pain and building a great product around it - Spending a lot of time on design and crafting a great experience - Talking to a lot of users via automated Loops.So emails Mistakes: - Bit off more than I can chew, should have started with a simple Web App instead of trying to build a full OS from the get-go - Rushing features to leave beta & start marketing because because I feel FOMO from other companies doing millions. Notion took 2 years to build, Figma took 4, and so on. - But, Apple took 6 months. While Constella is a very long term project, the AI hype is real... so it's tricky working on something meaningful - Not having clear user persona before hand
1
6
319
It turns out Cerebras is waaay better than OpenAi’s fastest model
We figured out how to do blazing fast "Jarvis Mode"🤯 It reads your research paper, connects the ideas, and then presents you with created notes in seconds! Thanks to the novel tech of @cerebras #ai #research #llm
1
4
352
Light themes, mobile full editor, and genius mode updates... A new update will be dropping soon!
4
9
605
day in life of non-b2b founder
14
182
Forget files and folders. Not your thoughts. 🦾 taaft.co/constella Constella is an AI-powered second brain that turns scattered information into a visual, connected canvas. No more digging through notes, links, and files. Constella captures everything, connects the dots, and surfaces insights right when you need them most. ⁠ Why you'll love it: 📌 Instant capture: Quickly save notes, web pages, files, podcasts, and more. 📥 Magic recall: Relevant notes and insights appear exactly when you need them. ⚡ Visual connections: Automatically see relationships between ideas in an intuitive infinite graph. ⁠ Bonus perks: 📲 Mobile-friendly capture 📖 Web clipping and highlight integration 🌍 Offline-first, seamless exports, total control of your data Clear your mind. Think better. Create effortlessly. Keep reading for more detail…👇 ⁠
6
4
22
32,681
Introducing Genius Mode — Thinking on Another Level Spotting patterns and finding related thoughts used to take hours of searching through documents. Now, AI search will automatically show you related thoughts as you go. This basically grants you a high level of intelligence for all your thinking — whether personal, work, or research. Try it out in the new 2.23 update!
1
3
8
4,892
I accidentally discovered a path to AGI. It's overly simplified, but I think it should work. Here's my step-by-step explanation, and curious to hear people's thoughts @ilyasut @karpathy @AndrewYNg in case you ever see this, is this something you guys have thought of? Rather than token LLMs, I think graph-based ones would surpass average human intelligence #agi #AI
1
9
401
We could take all the principles in the world — evidence, learning, knowledge claims, research papers — and construct a giant knowledge graph? GPT scraped the web but it just did a numerization of the values and then does text competition, a simple transformer. But what if instead of naively scraping the web, it went along, built up a knowledge graph, corrected ideas as it went, and resulted in one giant web? We'd have a AGI who could precisely answer anything in a logical fashion. And as we humans discover new things, we'd simply feed this AGI new data and it would recalculate principles & end of the graph claims (i.e. mindfulness actually releases GABA -> better brain structure -> further implications OR something random like new research -> implies humans think bottoms up -> top-down goal setting is bad.) Basically any questions we have, AI could actually logically answer it instead of just giving us a summary of what the web says. That too, what only the websites the latest GPT has scraped says. Obviously, each step along the way is difficult and would require models of its own, but this is similar to @AIatMeta's "Large Concept Model" LPC paper which goes away from the mainstream LLMs we have now
1
3
197
I'm mainly thinking about the best way to build "this principle relates to other" — anyone got any ideas
3
110