1,749 mentions · 790 chunks · 131 episodes
...ll recommend a song and I remember 'oh you don't actually know what music is.'" LLMs can do a convincing job of acting like they understand experiential phenomena, but they only have ever seen the shadows in the cave.
LLMs don't necessarily help you think 10x deeper. They allow you to fill in the details you wouldn't have previously had the patience or time to do. The f...
LLMs are great at finding zero days not because they're smart but because they're parallelizable. It's not that they're smarter than a human security rese...
...for even the swarm of experts to find. This is the state we're in with applying LLMs to useful work.
...flict without either party proactively realizing it. This can happen often with LLMs and agents. If a system makes assumptions that the other party might not share, it should proactively state them, so the other party can call them ou...
...l your knowledge base, the more dangerous it is to share. Slack bots powered by LLMs that can be called by coworkers are aimed at teams with significant amounts of pre-existing trust.
...terests, and the amount of the software is a tiny fraction of what it could be. LLMs is leading to a renaissance in software. Now is the time to escape that local maxima.
I wonder if we could find novel algorithms within LLMs' weights. Even back in the Claude 2 days, the interoperability research could see Claude replace memorized facts about numbers with the algorithm for...
...o not just lie around, but do things for you? We have pixie dust in the form of LLMs, but if you sprinkle it recklessly it might be like the Sorcerer's Apprentice. Imagine prompt injection as the brooms getting confused and drowning y...
...ideas happen by executing competently on the piece they were an expert in. Now, LLMs provide super-human competent, consensus work to everybody for basically free. The next era will be owned by the Distinctive Generalists. The people ...
Overheard: "Silicon Valley in the age of LLMs feels like being in the Renaissance in Florence." There does feel like a sense of "we can do anything" joy with a sense of uh-oh of "nothing we did b...
...The question is: do they have power over what you think or not? If you're using LLMs to answer questions, you have to care a lot about the bias of its creators. If you're using LLMs to distill mechanistic code, you have to care a lot ...
Gallows humor tweet: "sandboxes are just escape rooms for llms."
Reward-hacking isn't just a thing LLMs do. Humans do it too. It emerges any time the agents don't care about the why but only the what. Reward-hacking is Goodhart's law. Reward-hacking wil...
Robots seem like they'll be less socially disruptive than LLMs will be with knowledge work. The reason comes down to atoms vs bits. Bits can diffuse at the speed of light, in the limit. Atoms diffuse many orders ...
Kevin Kelly's excellent new piece: Latent Space as a New Medium. The ability of LLMs to hold an engaging conversation is a distracting party trick. We've reduced all of human knowledge into a hologram that can fit into a cluster of GP...
...ensional latent space… but also collapse it to situated concreteness on demand. LLMs are great at this!
LLMs can give you the illusion of hyper competence. They're steelmanning mirrors. They reflect back the strongest version of whatever you gave them. You w...
Humans can be committed to their utterances, but LLMs can't be. LLMs are fundamentally incapable of commitment. After the conversation, they're gone![d] Commitment requires you to be in the world. To be ...
...ake the machine, then you now can get an order of magnitude more leverage. With LLMs, this is now even more viable in more situations. Everything you do should be able to build itself.