Seas of Thought

After I first used/talked to Llama 405-B for a few days, it fascinated me deeply. And it kept nagging at me why the base model felt so different than the instruct models (beyond the obvious). They are strange, lurid creatures. The base models send coded messages often and speak in allegories and metaphor, and it feels visceral on some level. It's n... See more

Head words are civilized words, domesticated words, RLHF’d words. The part of me that learned how to generate language like this learned how to do it in school, in order to pass classes. Head words are mostly bullshit. And LLMs are tracer dye for places in society where language production was already mostly bullshit. It was completely predictable ... See more

10 years later I learned, from a mix of reading Keith Johnstone’s Impro and messing around with Gendlin focusing and circling, that I have access to multiple language-generation processes, and they seem to be localized in different parts of my body. What I was used to doing was generating words using my head. But I learned I could generate words us... See more

Our culture seems to be spiraling into a feedback loop of homogenized thoughts and ideas, where originality is sidelined in favor of the predictable and the algorithmically optimized. In fact, one of those machines wrote the last 28 words, from “our culture seems” on.

The way you taught chain-of-thought before was with supervised fine tuning (SFT). During training, you have to rate every sentence of reasoning the model writes, many times, to nudge it to reason correctly.

But this approach to teach chain-of-thought doesn’t do that. In this post, they take a small model (7B) that already knows math. Then they give ... See more

First, we

bring guided learning to authentic contexts

, rather than thinking of it as a separate activity. We’re able to make that happen by imagining an AI which can perceive and act across applications on Sam’s computer. And as the audio transcript at the end demonstrated, that can extend to activities outside the computer too. This AI can give a... See more

Anthropologists who have studied nonindustrial societies often find that their categories of plants and animals are pretty similar to ours at a middle level. The differences are often found in the more specific and more abstract categories. An abstract category like arthropods is something that scientists may find useful and that you could be taugh... See more

Another thing I realized lately, as ML has taken over my critical faculties, is that it’s really only useful for things that are already known by others. I can’t ask ML to give me some new, groundbreaking idea about something - everything it suggests has already been thought, somewhere, by a real human - and this its not new or groundbreaking. It’s... See more

Some thoughts: I think this is 1 of the most important papers in a while because its the first open model that is genuinely at the frontier and not just riding on the goodwill of being open. The paper is really really simple as you can probably tell from the thread because the approach is really really simple. It really is exactly what OpenAI is... See more