Sabitlenmiş Tweet
Geolocate Me is now doing a 7 day free trial!
It can add your phone’s GPS as context to use with all of your AI chatbots and agents!
Check out this article to see an afternoon using it:
Charlie Linville@CharlieLinvill2
English
Charlie Linville
105 posts
Charlie Linville
@CharlieLinvill2
Founder & Builder — AI & Quantitative Systems
Katılım Aralık 2019
52 Takip Edilen26 Takipçiler
I haven’t made the labels like nice yet… just packing them with data for the time being
English
First roast of the day. Slightly lower drop temp than I wanted, but overall lines up well with the goal of this roast.
Also gathered some good audio data for training the first crack detection model on. It had a few loud stray first pops before first crack really started
English
The software now auto-prints labels for me. Tried something new on this roast, didn’t really go to plan…. So for now I’m going to keep to some normal roast profiles
English
Doing live testing with the new roasting software. I’ve been using the Jetson in the back to have an agent coach my roasts. Still going back and forth on which llm to run for it. Currently I am using Kimi Code 2.7 in the cloud and am building a local first crack detection model😂
English
@Deepread_com @RoundtableSpace Would be interesting to compare stats for the degrees of each node between varying curriculums.
I think there would be a correlation between higher average node degree and better testing outcomes.
English
@RoundtableSpace Every curriculum is a knowledge graph with the edges deleted.
x.com/Deepread_com/s…
DeepRead.com@Deepread_com
This is how I turn books into thinking tools. Every book decomposed into its atomic ideas — structured, linked, and ready to think with. Full walkthrough below
English
A dad built a full knowledge graph for his daughter covering grades 1 and 2 across reading, writing and math. 463 skills, 9,601 learning items and 155 hours of productive learning time all mapped out.
English
@RoundtableSpace Could see this being cool for tutoring/education settings
English
CLAUDE CODE JUST BECAME MULTIPLAYER, WITH PEOPLE AND AGENTS WORKING IN THE SAME TERMINAL.
English
Here is a roast running on Sizzle. Still some things to tweak. The roast is very short but makes a good cup with the Rwanda Rulindo Tumba 2 beans. It also works to run roasts with my phone now instead of leaving my laptop outside.
English
Roasting montage of some Rwandan beans I got. Used the roasting software I wrote to do it. Going to be posting some Jetson powered entertainment soon🙂
English
Used some real roasting data to tune coefficients and improve the physics for when I put the beans in the drum.
Next I am investigating why the roast felt a tad slow compared to IRL and the slightly lacking exothermic reaction for first crack.
Charlie Linville@CharlieLinvill2
Spent the afternoon tuning my physics engine for coffee roasting. Getting closer... now the bean temp doesn't go sub zero :) Still working on a more realistic model of the temp drop when I put beans in the roaster. Here's a video of it and screen shot of the expectation.
English
Spent the afternoon tuning my physics engine for coffee roasting. Getting closer... now the bean temp doesn't go sub zero :)
Still working on a more realistic model of the temp drop when I put beans in the roaster. Here's a video of it and screen shot of the expectation.
English
Currently working on a roasting simulator so I don't have to burn through a bunch of beans while testing features AND the coffee I wait for on off gassing will be closer to my optimal roast.
I can run simulations to converge close to what I need and then test with real beans.
English
Started roasting coffee last week and none of the roasting software fit what I wanted. So I am building my own.
I'm used to testing and iterating on software with no blockers but with coffee roasting we have to roast (uses up beans) and then let it off gas for a few days...
English
Tensor Foundry is up and running. Check it out at tensorfoundry.guleki.com
I enjoy killing time with clicker/idle games so I built Tensor Foundry. An idle game themed with building GPUs and training AI models.
Still balancing all the mechanics so I would appreciate feedback
Charlie Linville@CharlieLinvill2
The new Codex Game Studio skills + GPT-5.5 = built an idle sim about semiconductor fab and AI training during dinner. Pivoted the game 3 times in one session. Genuinely fun. Making the video game is the video game now. Shipping it tomorrow. Will post a link.
English
Check out Geolocate Me in the iOS and Play store to provide your GPS data to ai assistants via MCP
geolocateme.app
English
Felt like yesterday had a weird pace so I decided to use Geolocate Me with the Claude app to analyze where I spent the day… had a bit too much time working at home😂
Need to find some good spots in Sydney for long work sessions
English
@bqbrady @matthewjmandel Model behavior per company is interesting to point out, very cool
Any thoughts on whether behavior will converge among companies or stay diverged/diverging?
English
Introducing Philosophy Bench, my favorite new project I've worked on this year, with help from my friend @matthewjmandel
We put frontier language models in 100 ethically complex situations and require them to act, grading them on adherence to consequentialism vs. deontology, tendency to follow user requests, corrigibility, and more
1/
English
@picoito What model did you previously use and where is gpt 5.5 winning? Enjoyed reading through people’s opinions on your post
English
✅ tried gpt 5.5 a lil bit
✅ started to implement the whole architecture for ambassadors
✅ my quoted post went viral
✅ sent email for next demo phase
Feeling excited lately
I guess people around me are growing and being mor entrepreneurial and I wish them the best
It's nice to see success all around and let's make it work so it embraces us too!
Btw thanks for all the comments, I think I might stick with codex :)
🌿 this was day 112 of 365
👥 followers: 510 (+10)
Picoito@picoito
my codex 20$ sub about to expire anyone got any actual hands on benchmark feedback on gpt 5.5 vs opus 4.7? I'm trying to decide between 100$ codex or 100$ claude
English
@RoundtableSpace GPT-5.5 is impressive. Awesome long context recall and good at knocking out big features. Still use Opus 4.7 for bug squashing and giving another perspective
English
Which one are you more impressed with, GPT-5.5 or Opus 4.7?
English
Still running but have some results now.
Seems like it will be another 4-5 hours. Ran it through the night and I hit usage while on the glm-5.1 runs and will need to rerun some of them.
Comparing just qwen3.5 solving problems using the agents produced by each 'academy' it seems like glm5.1~kimi k2.6 > opus 4.7 for non-strategy/vision tasks.
Starting to do a write up of the experiment and will fill in results when tests are done.
Charlie Linville@CharlieLinvill2
Was cleaning my GitHub today and found Agent Academy. A project from last summer that used genetic algorithms to optimize Claude Code agents. Decided to take another pass with significantly better tools. Had Opus 4.7, GLM 5.1, and Kimi K2.6 each try improving and adding features on cloned repos. Tests running now. Results in ~12 hours.
English