🅒🅾 Chase Olivieri

@chaseolivieri

Frequently on planes. Building ✈️ – Adventure traveler & surfer. Prev @DukeU + @teachforamerica

35,000 ft

Joined July 2008

261 Following

243 Followers

833 Posts

chaseolivieri retweeted

Dillon Loomis

@DillonLoomis

6 months ago

Some uncomfortable, no hype realizations on day 3 of OpenClaw: • If you don't run models locally, you will indeed burn through tokens. Just for normal conversation, some light tasks and a few simple cron jobs (automations), I've spent $70 on Claude credits in ~24 hours. Certainly not sustainable for my use cases right now. Most of my back and forth wasn't even using Opus, it was Sonnet/Haiku. I'm starting to think a Mac Studio with extra RAM is the superior long term play so you can avoid API tokens as much as possible. But just know, all the posts about these overnight automations and morning briefings and project creation are likely costing hundreds of dollars a week if the models aren't local • The coolest thing so far has been going through my OpenClaw bookmarks folder of X posts with features and add ons and just feeding each post to my Bot. So whether it's memory optimization or checking for security vulnerabilities or creating small projects and pushing to GitHub, whatever I've told it to do, it just does it on its own. It really is wild. The capabilities are real, but if you don't have specific use cases for your life, you should think about exactly what you'd want automated before jumping in • I created all new accounts for my Bot, and watching it create Google Docs in Google Drive on its own is really cool. My primary focus right now is on health optimization and taking advantage of the permanent memory to have one database with all of my health history that can quickly be exported and to have an expert researcher to diagnose, detect and stay proactive. This alone was always going to be worth the financial and time investment for me and so far, it's exactly what I was hoping for and more • The firewall and separate accounts I've created is part of the problem. I've been overly concerned about security issues so I haven't given it any access to my emails or files or anything personal. I might in the future but the prompt injection risk and unkown attack vectors seem like too much of a risk for me, I have too much to lose. For now, this is going to limit a lot of the potential benefits and automation, but I'm just not ready to take on the risk and there is plenty for me to learn/explore and build before crossing that chasm • For me this entire endeavor was always going to be a multi-year project. I was never expecting to be retired on a beach in a few weeks. I just wanted to start learning this world more intimately and to be ready as things advance from here. But I know a lot of people will dive into this wanting big automations in the short term and I think many in that group will end up disappointed. This is very clearly the future, but if you're not willing to give it access to your actual life and accounts, the early benefits may be limited relative to your expectations going on • It really is addicting. The self-correction is impressive, anytime there's a limitation or problem, it just figures it out with minimal input from me. There is so much potential with this and it's only the beginning, but the token burn and accessibility concerns should be emphasized more than they are right now

DillonLoomis's tweet photo. Some uncomfortable, no hype realizations on day 3 of OpenClaw:

• If you don't run models locally, you will indeed burn through tokens. Just for normal conversation, some light tasks and a few simple cron jobs (automations), I've spent $70 on Claude credits in ~24 hours. Certainly not sustainable for my use cases right now. Most of my back and forth wasn't even using Opus, it was Sonnet/Haiku. I'm starting to think a Mac Studio with extra RAM is the superior long term play so you can avoid API tokens as much as possible. But just know, all the posts about these overnight automations and morning briefings and project creation are likely costing hundreds of dollars a week if the models aren't local

• The coolest thing so far has been going through my OpenClaw bookmarks folder of X posts with features and add ons and just feeding each post to my Bot. So whether it's memory optimization or checking for security vulnerabilities or creating small projects and pushing to GitHub, whatever I've told it to do, it just does it on its own. It really is wild. The capabilities are real, but if you don't have specific use cases for your life, you should think about exactly what you'd want automated before jumping in

• I created all new accounts for my Bot, and watching it create Google Docs in Google Drive on its own is really cool. My primary focus right now is on health optimization and taking advantage of the permanent memory to have one database with all of my health history that can quickly be exported and to have an expert researcher to diagnose, detect and stay proactive. This alone was always going to be worth the financial and time investment for me and so far, it's exactly what I was hoping for and more

• The firewall and separate accounts I've created is part of the problem. I've been overly concerned about security issues so I haven't given it any access to my emails or files or anything personal. I might in the future but the prompt injection risk and unkown attack vectors seem like too much of a risk for me, I have too much to lose. For now, this is going to limit a lot of the potential benefits and automation, but I'm just not ready to take on the risk and there is plenty for me to learn/explore and build before crossing that chasm

• For me this entire endeavor was always going to be a multi-year project. I was never expecting to be retired on a beach in a few weeks. I just wanted to start learning this world more intimately and to be ready as things advance from here. But I know a lot of people will dive into this wanting big automations in the short term and I think many in that group will end up disappointed. This is very clearly the future, but if you're not willing to give it access to your actual life and accounts, the early benefits may be limited relative to your expectations going on

• It really is addicting. The self-correction is impressive, anytime there's a limitation or problem, it just figures it out with minimal input from me. There is so much potential with this and it's only the beginning, but the token burn and accessibility concerns should be emphasized more than they are right now

115

929

969

90K

🅒🅾 Chase Olivieri

@chaseolivieri

Who to follow

Last Seen Users on Sotwe

Trends for you

Most Popular Users