Daniel Jeffries
@Dan_Jeffries1
Author, futurist, and systems architect. Recursively self improving.
I need a one hundred trillion token context window. I fear "compacting" like I fear sharks. The model instantly becomes stupider and forgets key context from earlier sessions that it needs to remember.
Gemini CLI is superior to figuring out complex bugs, and WAY faster. It speedruns through the repo in 10x the time it takes Claude to walk through. I ask Gemini to explain only since it's edit tools are crap, the copy all text and paste it into a fresh claude session to fix.…
TLDR, know what the fuck you are doing already and AI works great. Don't and it won't do shit for you in UI design (or anything else) because you don't know what kerning and negative space are and couldn't recognize them if they bit you in the ass.
"The difference between AI slop and great designs is in the last 10%." Here's my new episode with @MengTo where he reveals his step-by-step process to create stunning AI designs and websites. We cover how to: - Build your AI design prompt from scratch - Find great fonts,…
The point @sundeep is making about be combinatorial effect of having many (now 7?) Chinese open AI models is very powerful, and I mis-appreciated by many. Each model can improve each other model. And new models are much easier to launch. 🧵
China's open source and open weights culture around AI is creating a lot of leverage for the open source ecosystem. Companies can openly distill, generate synthetic data, and build on each other’s work. The primarily closed approach in the US is driven by "moat" culture.…
The key to these AI platforms getting really good is making them DIRECTABLE to artists. Artists are always in the best position to take best advantage of these tools and that will become increasingly clear as they tools mature and get more powerful and by extension more…
We just discovered the 🔥 COOLEST 🔥 trick in Flow that we have to share: Instead of wordsmithing the perfect prompt, you can just... draw it. Take the image of your scene, doodle what you'd like on it (through any editing app), and then briefly describe what needs to happen…
The verification bottleneck in AI workflows: If you use the systems to do work, checking that work is hard and the time to do it adds up but it must be done or don't even bother. The bottlenecks in all AI workflows: 1) Concepting/Planning 2) Iteration 3) Verification…
This is pretty cool. Easier to hit the esc button when running with --dangerously-skip-permissions 😂
Claude Code with full visibility into what your agents are doing 👀 npx claude-code-templates@latest --chats Opens localhost:3333 with: • Real-time conversations • Token usage tracking • Tool call details • Session analytics Runs locally = full privacy 🔒 Repo:…
"You are a QA manager with a personality disorder and an alcohol problem. You viciously critique all unit tests and view hardcoded passes, workarounds and as an affront to God's Creation, punishing the authors of such heresy with extreme predjudice until they fix the tests, and…
Claude Code is getting a brand new feature: custom subagents. Type `/agents` to get started.
It was called BBS.
I feel like there's a huge opportunity for a "no algorithm" social media platform
Cool stuff. I am not smart enough to work on this stuff but maybe some of you out there following me are!
I’m building a new team at @GoogleDeepMind to work on Open-Ended Discovery! We’re looking for strong Research Scientists and Research Engineers to help us push the frontier of autonomously discovering novel artifacts such as new knowledge, capabilities, or algorithms, in an…
“I don’t have a P(doom) number. The reason I don’t is because I think it would imply a level of precision that is not there. So, like, I don’t know how people are getting their P(doom) numbers. I think it’s kind of a little bit of a ridiculous notion because, um, what I would say…
Here's my conversation with @demishassabis, CEO of Google DeepMind, all about the future of AI & AGI, simulating biology & physics, video games, programming, video generation, world models, Gemini 3, scaling laws, compute, P vs NP, complexity, energy (solar & fusion), and much…
This is cool work and much needed.
🚨New Paper!🚨 We trained reasoning LLMs to reason about what they don't know. o1-style reasoning training improves accuracy but produces overconfident models that hallucinate more. Meet RLCR: a simple RL method that trains LLMs to reason and reflect on their uncertainty --…
My Claude Code cheat sheet: 1) claude code --dangerously-skip-permissions 2) esc -> "what the fuck are you doing, remember that.." 3) ctrl+l That is all. :) Oh also, run in claudebox so it can't eat your machine. Great project.
Nobody messes with Jensen.
President Trump says he planned to break NVIDIA up, until he learned who Jensen Huang was.
“Think of how stupid the average person is, and realize half of them are stupider than that.” ― George Carlin
This is a pretty awesome step forward. Even if RL and tool use never scales to solving truly messy problems like creativity or making choices in open ended spaces with many "right" or "wrong" answers, aka most of life, then it is still incredible to see machines getting stronger…
Another AI system, ByteDance's SeedProver solved 4 out of 6 IMO problems *with* Lean, and solved a fifth with extended compute. This is becoming routine, like when we went to the moon for the fourth time. There is *nothing* "routine" about this!!...
Great write-up on the new White House AI Action plan and the shift in the American mindset to more realistic policies after being poisoned by astroturfing micro safety startups for too long.
This is a major shift in the tenor of how the U.S. government feels about and prioritizes open AI models. This narrow piece of the action plan is a huge W overall, with a couple nit picks. Happy to share my thoughts as I scope my own path to building this reality.
thanks for the fodder. made this into a comic. hope you'll enjoy. :)