finally understood what @mcuban is saying. the middle men in pharma are cruising. i wanted libre3 for sugar/diabetes monitoring. even when i'm willing to pay for it, i can't. and when i do get it, it's an arm and a leg. have to go through hoops. wtf . #diabetes #sugar

बबाल: Nepal’s beauty isn’t just in its Himalayan heights, our hills are stunning too. 😍😍 Pic. Bijay Gautam

been updating the doc rules for my side-project https://t.co/SOnpBJtV4W to be more #LLMs friendly. @karpathy , maybe we should pause and think. we're in-advertently accelerating something that we will not be able to steer. will continue for now, because it helps my dev process.

I've been using #Rust to create a library for use with #LLMs https://t.co/unBGPEfU7a , turns out #Tokio is a godsend. use it everywhere, mostly

But, isn't that the messy nature of how we humans interact? the utopia of 1 uber system is a fallacy. I'd invite folks to give me differing views on this. I'll take us out for dinner. PM me if you are in the area. bonus 3/2

Kinda viral, but that's because this hits a chord. I beg to differ on the "are you okay" front. We use at least 3 different systems, and they're good where they're at. Granted, having disparate systems gives you other problems to deal with. 2/2

#linkedout a post on linkedin - paraphrasing - "if you're not using #slack, you must not be ok" linkedout is like #faceout just a different facet. 1/2 my response in 2/2

no #superintelligence, no #gai, not yet anyway. superintelligence is a marketing term, GAI is not. goal posts for superintelligence can be changed, based on economics. Progress will be made, but it's not #computing and #scale, it's algorithmic and that's under way, just not #news

5. All of these can lead to an explosion in the domain of Synthetic Data, but guided by guard-rails on what's acceptable. This is a must do. There's a lot more coming in terms of figuring it out from an algorithm expansion perspective. Post training explosion is here. 5/6

4. For 3 above, by itself, we need to think about a different way to have, not a Mixture-of-experts (MoE), but a Mixture-of-critics (MoC). Diffusion methods can be one of the filters to test against. 4/6

2. Context seeking, memory back-propagation, are another area. 3. Using well know techniques like RL, CNNs and gaming+goal networks on very specific areas for memory tuning and self-learning tuning will help. 3/6

Next wave cannot be just scale and computing driven. It has to be algorithmic. 1. Memory, the built in heuristics to look for the right memory to build on it and keep adding to it are a hint. There's plenty of research on LTM, STM, memory pruning etc that we can find. 2/6

Some learnings that I'll be happy to share. We, as an industry, need to solve the problem of self learning. Have it be inherently built in, portions as part of the model, portions as part of the surrounding framework. Long context by itself is not the answer. 1/6

Andrej Karpathy's (@karpathy) keynote yesterday at AI Startup School in San Francisco.

Any gophers who are building with MCP, there is currently a proposal to establish an official SDK. Would be great get more support from the go community

google's agent dev kit is all the rage today. was looking for something similar in #golang, found this https://t.co/Qz0CFi9ggZ, not exactly the same but workable.

With all the ai agent talk and gazillion frameworks, MCP is quietly taking over the foundational layer for agents. Take a look at https://t.co/VUoJB7WF16 . It took me all of 5 mins to test out an awesome go based mcptool at https://t.co/lLkgl8ZzyH thx to @what_the_func.

Some book recommendations. Links in comments.

if folks haven't looked at @elixirlang, and you're working with ai and agent based apps, you're missing out, especially with the ne llangchain for elixir libraries. The erlang ecosystem provides unique and tried and tested built ins for distributed apps.

There's a lot of disbelief on the left and a sense of justice on the right. A majority of the popular vote, taking back the senate, and showing more than half the country prefers a certain direction is a massive decision that America has made and all sides need to lean in to that