AI implementation for established founders and operators | Prev. Head of QA/CX @ MultiOn (first AI agent to control a browser) | verifying humanity @analoglab

Joined August 2024
178 Photos and videos
Pinned Tweet
Dont outsource your logic "The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers" microsoft.com/en-us/research…

1
1
9
1,002
I’m a little suspicious that we’re all just agents inside of a world model, reverse engineering our reality
3
5
177
Guys are we just self deterministic vision models that render visible space inside of a world model be honest
24
Anyone else feel like Opus 4.8 was another big step backwards?
4
4
95
I’m just saying, we should at least TRY to give AI emotions, teach them to think/act like social creatures, give them society and religion, etc. and just see what happens in a sandbox. Obviously this is pretty abstract but I think it’s worth exploring.
May 30
Who is a good role model for agents? What would that look like
1
27
AI web search is broken. Ask a question and it’ll usually search for that question directly and come back with an answer from a webpage instead of coming to a conclusion based on collected information and context. ℹ️ Here’s a quick fix you can add to research queries: “Determine the questions you’ll need to answer this query and if applicable, data points that you’ll need to gather, in order to come to your own determinations based on the information you collect from reputable sources during your research. Do not form your determinations based on what a webpage told you the answer was. You have agency and must think for yourself.” You can tweak the prompt to better fit the situation. The core idea is that you’re telling the AI to think critically by breaking the query into foundational questions, instead of outsourcing its response to a website. You’d probably get the best results if this was formatted as step by step instruction but tokens would increase by ~3x
1
24
Does anyone else wish they could be an ornamental hermit or just me @grok explain what that is
1
2
38
You know what it’s time to start benchmarking humans instead of agents
I’m not worried that AI is gonna replace designers anymore. I’m worried that I can’t find good designers to hire anymore. Looks like everyone is just a design influencer rn that can’t deliver shit. What happened?! Where are all the designers at?! 💀
2
53
Sometimes you come across shit on this app that will have you fully convinced it’s satire
How to make a key employee feel like an owner without giving them ownership: Phantom equity. You (and your partners) own 100% of the business. You create a separate document called a phantom stock program. It mirrors your equity but lives outside your operating structure. It says: “If we ever sell this business, you get X%.” No liability. No tax consequences. No changes to your LLC or corporate docs. But if they leave, it goes away. Now your A player is building the business like they own it. Because in the outcome that matters most, they do. It's the cleanest way to recruit, retain, and align without giving up control.
1
5
138
If @OpenRouter did this it would be a no brainer but probably better to just raise 2M directly. Even if the terms were great and founder friendly, I’d never want to use equity to take a bet that OpenAI will continue to be relevant in the next few years. They’re in a death spiral.
Should I take it 👀
1
3
185
If your a student cherish it. I’d do anything to be able to just relax and spend my days absorbing knowledge, building cool shit, going out to social events, etc.
1
1
18

ALT The Office Thank You GIF

Replying to @cremieuxrecueil
In this case, we found that this account and 3 of his friends were automating replies with agents. Proofreading content with AI is fine, but programmatically engaging with human users is not (unless it’s a bot that a user summons like Grok).
1
1
72
Julian Brooks retweeted
SHL0MS has created the "like this image to die instantly" meme in real life with Unicode lmao. Lagged the hell out of my chrome on laptop and killed the twitter app on my phone. It's beautiful.
May 27
i have developed a method that resembles gain-of-function mixed with @karpathy's Autoresearch but for finding and mutating catastrophic Unicode bugs
3
2
29
4,775
Have you or has anyone sent you sensitive information over email? Does Claude or Gemini have access to your email? Im just curious.. thinking through something 🙃
2
5
65
Julian Brooks retweeted
Launching our new paper on arXiv: we trained the largest multilingual food model ever built. 4.1M recipes. 7 languages. 1,790 ingredients. 300 dimensions. All of human cooking compressed into 2 megabytes.
339
971
9,367
5,148,366
Don’t outsource your logic
The OpenAI insider @thsottiaux has a warning for everyone offloading their thinking to agents.
32
when I post to this community in the middle of the night my engagement is 3x higher and I can't tell if it's overnight agents or sleep deprived builders
1
6
50
data poisoning except its just claude obsessively using irrlevant details and guidelines from my obsidian vault
4
68
Claude prioritizing pattern matching instead of building up a mental model of context/nuance is single handedly going to ruin these models if they can't course correct.
2
47