LLMs unlock frictionless, infinitely scalable knowledge transfer bandwidths. Given the many tailwinds, we should be thinking seriously about new forms of capital based on "knowledge" - and their effective governance.

In the AI race, it is the context producers who will always have the edge. GPUs and researchers can increase context windows, but only our perspectives and experiences can fill vector space.

The OpenAI store was inevitable. Incredible value in custom agents with access to "expanded knowledge". I've been anticipating the formation of knowledge creators/communities around specialized/proprietary data - and GPTs are just the beginning of this. With the mention of rev

Riva Erik Voorhees AI must be decentralised to ensure that the narrative is not controlled by the few. This is a technology that will be able to have a mind lock on people. It’s a control so strong we’ve never seen before. Why? It’s because of the emotional connection people will have to their AI

Great read from Rob Phillips on how individual preferences will be integral to AI agents. However, there is another part of the preference equation - and that is of the people & communities we surround ourselves with Solutions which optimize for social capital > those that don’t

My thoughts on the GPTs move by OpenAI: Beyond the obvious “to create an App Store” or “to collect more data from users” there are a few other (possibly contrarian) considerations. - Like all technology, AI will mature. As we transition to sustaining innovations the moat around

This will be critical as many human-to-human/machine interactions begin transitioning to M2M

An interesting aspect of this discussion is the fact that LLMs will soon start affecting our thoughts, beliefs, mental & linguistic habits, and culture. The idea that we could select a handful of "trustworthy" institutions with the "correct" set of values and beliefs to shape LLM

Not to rain on the Q*Anon parade but those AlphaGo self-play references are misguided. You can't do "self play" if you don't have an in-built reward signal. AlphaGO had the rules of the simple closed form game embedded as the external reward. What does an AGI-wannabe LLM have?

Some thoughts on Self-Play and self critiquing in LLMs (responding to replies to the tweet below--doing a quote tweet so I can consolidate the responses). [A mini Post-thanksgiving Teach-in] 1. If by self-play we mean LLMs generating and critiquing their own ideas by themselves,

Musk is king troll 😅 However, I feel like most people are forgetting that we don’t actually need AI to have common sense - symbiotically we serve that function. Understand that the idea of AGI is alluring, but what we have today is already transformational. It’s just how and

Makes me wonder what sort of emergent effects this will have on preferences and trust as people adapt to change. Pretty solid bull case for crypto if you ask me.

Great experiment on GPT-4 for legal tasks finds the increasingly familiar disruption: 1) Boosts quality, especially for lowest performers 2) Big increases to productivity 3) The Jagged Frontier: AI is better at some tasks than others 4) Users are happier papers.ssrn.com/sol3/papers.cf…

You can either resist your vices, and chase virtue. Or give in to your vices, and chase happiness. Unfortunately there’s no half measures.

GenAI is about scaling knowledge distribution. And the most valuable knowledge is actionable

AGI ! Coming to a verifiable domain near you!

Unpopular opinion: The reasoning abilities obtained from RL on arc/gpqa/coding etc that transfer to other domains will not be sufficient to power the level of automation or reliability of tasks that almost everyone seems to expect. Human orchestration, context setting, and

Manually designing/writing/checking reasoning traces really feels like automating myself.