Alexander

@Building_Josh Ever turned your phone into an “always on” mic as a part of your observation layer? Curious to hear best solutions

seems like all human knowledge needs to be restructured to better feed LLM/multi-agent systems, analogous to soul.md files, as the new UI keeps changing. just as the UI recently shifted from blue links to LLM chats, the next evolution many layers of abstraction higher is the intent-based system: multi-agent systems interpreting our intent through voice, facial recognition, and eventually neural interfaces that can tap into the power of human intuition (biggest step towards post-humanism). for now seems like there’s a big opportunity in bespoke and adaptable UIs, calibrated by the user’s technical level and intent, which attempts to bridge the data processing discrepancy between man and machine: bits -> base pairs though the immediate behavioural hack is to just develop “agent empathy”. to understand the model from it’s perspective, and feed it accordingly

@maxleedev cool

@LoicReco how?

you can use claude code to improve your wifi connection

traversing from the particular to the universal

"God" as the totality of energy within the universe. The "collective unconscious" being the bridge between the two, and the human-centric abstraction of the energy

@lazaro45ive Thanks !

Music: Hiro Sekai - It was written

I feel lost in this world

The race for LLM "cognitive core" - a few billion param model that maximally sacrifices encyclopedic knowledge for capability. It lives always-on and by default on every computer as the kernel of LLM personal computing. Its features are slowly crystalizing: - Natively multimodal text/vision/audio at both input and output. - Matryoshka-style architecture allowing a dial of capability up and down at test time. - Reasoning, also with a dial. (system 2) - Aggressively tool-using. - On-device finetuning LoRA slots for test-time training, personalization and customization. - Delegates and double checks just the right parts with the oracles in the cloud if internet is available. It doesn't know that William the Conqueror's reign ended in September 9 1087, but it vaguely recognizes the name and can look up the date. It can't recite the SHA-256 of empty string as e3b0c442..., but it can calculate it quickly should you really want it. What LLM personal computing lacks in broad world knowledge and top tier problem-solving capability it will make up in super low interaction latency (especially as multimodal matures), direct / private access to data and state, offline continuity, sovereignty ("not your weights not your brain"). i.e. many of the same reasons we like, use and buy personal computers instead of having thin clients access a cloud via remote desktop or so.

@spolu Is any improvement needed in the current tech stack to enable mass market educational agents to hugely improve the educational system? My intuition is that the last product feature needed is robust long term memory (some improvement on RAG)

Every year we take an intern from high school for 1-2 weeks. It has been mind blowing to witness the evolution of what you can make them do with 0 prior knowledge in CS over such a short period. 2 years ago: ~nothing, a bit of JS 1 year ago: a fairly complex self contained projects based on a research paper (in English, a language they didn’t speak :p) (Per Bak neural networks implementation was it) This year, it’s been 3 days so far and I’m hopeful they will be shipping non trivial code in production before the end of the week; learning the linting, type checking, PR creation, and deployment procedure in the process with minimal very high signal interactions with the rest of the team. The rule of the game is simply: if you don’t understand anything, ask an agent. Ask us only when blocked for longer than 30mn. The key change is that, if they are willing to play the game, they are capable with the help of agents to figure out high level instructions independently, just vibe learning and coding all day long. And agents have just become such great tutors. So I agree with the prediction, the workforce that is about to come online is going to be unstoppable.

@bentossell How do you think Claude Code’s terminal-based AI coding could evolve to address Andrej Karpathy’s concerns about jagged intelligence in LLMs, especially for real-time debugging in complex codebases

🙃 today's intro

1st client secured!

Life should not be a journey to the grave with the intention of arriving safely in a pretty and well preserved body, but rather to skid in broadside in a cloud of smoke, thoroughly used up, totally worn out, and loudly proclaiming "Wow! What a Ride! ~ Hunter S. Thompson @lexfridman

Deeply amusing that the greek “logos” means both “word” and “reason” considering humans almost never use words to convey reason 🤣

@archi_reum It’s always Sunday outside the simulation

Another monday in the simulation !

The ultimate vision is to radically improve how humans interface with knowledge, particularly young minds. Above all else education should install a Philosophy of Curiosity. Learning should be addictive and inherently engaging. A bespoke AI Agent for each learner will enable this. It will have memory of all of their interests and innate skills and guide them accordingly. The smartest generation of humans is soon to come… @karpathy

Interested to see what problems I encounter as the memory scales. I’d predict the mainstream LLMs release this functionality in the next 6 months

I’ve noticed LLMs lack robust long term memory - and considering context is everything and token limits are prohibitively small, i’ve had to build external memory for our system (for the agents and my Custom GPTs). I’m using a Vector Data Base with semantic memory retrieval.

Excellent - youtu.be/0jinOTQ9BaU?si…