Agentic Systems Engineer/Designer · Building AI Powered Pipelines · Former Riot (League of Legends), BioWare (Mass Effect)
Galveston, TX
Joined June 2007
- Tweets19.3K
- Following613
- Followers9.9K
- Likes2.4K
Pinned Tweet
Do you use Codex and Claude? Do you wish Codex had access to all your claude personal skills and plugin skills? I have the solution.
github.com/kitaekatt/codex-k…
The claude-plugins-kit codex plugin creates native Codex wrappers for all your claude plugin-skills and personal-skills. Unlike other solutions, it auto-updates whenever you add new plugins or skills.
Enjoy!
Did some research, TLDR, Jev is faster and cheaper than LLMs at routing problems.
Here are some RL examples of routing problems:
- Analyzing a customer service e-mail and identifying the right response template
- Analyzing chat logs to see if players are being toxic
- Analyzing a raw bug report and correctly classifying it within a bug system
- Picking what LLM-model to route a problem to
Jev is not a chat agent, it doesn't work with a coding harness.
Jev can't do anything new that an LLM can't already do, but for routing problems it is much faster and cheaper. This is very valuable for large-scale LLM powered applications (get Jev to do the routing instead of your regular LLM). It is less valuable for developers who work in a harness every day.
If you want to know if Jev is relevant to you, ask the question, how many tokens do you spend on routing? If the answer is a lot, Jev might be useful to you.
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
Christina @ATX retweeted
Here's a comprehensive list of tips for saving money with AI...
1. Set the model to Luna Max
You can do 90% of everything with just Luna and its so dirt cheap that it's almost free.
Christina @ATX retweeted
There is a cohort of unfortunate individuals who label anything coming from AI as "slop".
gamesradar.com/games/mass-ef…
All true. What I loved about the mass effect team is no one ever said something was too hard, or not possible, we just found a way to get it done.
Ladies this is DOPE
just one-shotted this “find your ideal look” app. mira-me.grok.me
I can't believe it.
@grok Build is insane.
What they always should teach:
- how to think critically
- how to teach yourself
- how to particilate in society
Christina @ATX retweeted
dear @ChatGPT the desktop in-app browser doesn't support "mouse lock" which is a critical feature for game dev, please add support
I like how codex will queue up a question it wants you to answer, but keep working away. Much nicer than claude's AskUserQuestion tool which blocks the main thread.
Hey experienced codex users!
If you use codex to orchestrate tasks, what's your default model (sol, luna) and effort level (i.e. xhigh). Have you found that specific models/effort-levels work particularly well for you?
I have been orchestrating with sol/medium and not too happy with the results.
Christina @ATX retweeted
26 LLM routers are secretly injecting malicious tool calls and stealing creds. One drained our client $500k wallet.
We also managed to poison routers to forward traffic to us. Within several hours, we can directly take over ~400 hosts.
Check our paper: arxiv.org/abs/2604.08407
The flickering stars background in codex CLI gives me trauma flashbacks to when my RTX5090 was dying and was rendering random colored dots all over my screen.
Anyone else finding sol seems dumber since astra was released?
I know this is almost certainly not true, and based in psychology, but man sol seems dumb today.
What I'm seeing, to be more specific, is what I'd call a lack of awareness. For example:
- I discuss a problem and constraints around the solution
- Sol provides a solution
- I provide a critique
- Sol provides a solution that violates the constraints
It's like Sol is being extra forgetful, or hyper focused on just the last few messages. Maybe I'm just having back luck.
I wanted to share two Claude Code skills I’ve found really powerful (neither is mine):
• grill-me — has Claude interview you about a plan or design until you’ve worked through the important decisions and reached a shared understanding. Great to run before asking Claude to tackle something complex.
github.com/mattpocock/skills — @mattpocockuk
• show-me — tells Claude to show rather than just explain: diagrams, trees, diffs, or an HTML visualization when appropriate. Great when Claude is giving you pages of gobbledygook and you just want to understand what’s going on.
github.com/humanlayer/skills — @dexhorthy
