Hacker Newsnew | past | comments | ask | show | jobs | submit | gradascent's commentslogin

From the figure in the first paper listed:

> Responses to the query “Write a metaphor about time” clustered by applying PCA to reduce sentence embeddings to two dimensions. […] The responses form just two primary clusters: a dominant cluster on the left centered on the metaphor “time is a river,” and a smaller cluster on the right revolving around variations of “time is a weaver.”

I just gave Gemini 3 the same prompt and got something quite different:

>Time is a patient wind against the cliff face of memory. It does not strike with a hammer to break us; it simply breathes, grain by grain, until the sharp edges of grief are smoothed into rolling hills, and the names we thought were carved in stone are weathered into soft whispers.


Constantly flowing and makes things smooth like river stones; compared to Tait's "time is a series if staric pictures", Gemini's output is not so different from a river metaphor.

I was thinking the same thing. That one accelerates its growth in the presence of radiation. But it also seeks out human flesh and brains to build its biomass intelligence blob, unfortunately.

In the books, it is suggested (if not stated outright) that the protomolecule was probably intended to work with much simpler forms of life, but is also able to make use of higher forms like humans.

The reason so many were infected on Eros was because humans deliberately infected everyone on the station. Likewise with the human/protomolecule hybrids.


Adding to the chorus: I like the older images and the precise year is important! The underground shelter is something that wouldn’t exist just a few years after that photo, or before.


> The underground shelter is something that wouldn’t exist just a few years after that photo, or before.

It’s probably there now, it’s the London Underground.


Great deep dive. I've learned a lot already and haven't even finished the introduction


Then why do I never get an “I don’t know” type response when I use Claude, even when the model clearly has no idea what it’s talking about? I wish it did sometimes.


Quoting a paragraph from OP (https://www.anthropic.com/research/tracing-thoughts-language...):

> Sometimes, this sort of “misfire” of the “known answer” circuit happens naturally, without us intervening, resulting in a hallucination. In our paper, we show that such misfires can occur when Claude recognizes a name but doesn't know anything else about that person. In cases like this, the “known entity” feature might still activate, and then suppress the default "don't know" feature—in this case incorrectly. Once the model has decided that it needs to answer the question, it proceeds to confabulate: to generate a plausible—but unfortunately untrue—response.


Fun fact, "confabulation", not "hallucinating" is the correct term what LLMs actually do.


Fun fact, the "correct" term is the one in use. Dictionaries define language after the fact, they do not prescribe its usage in the future.


Confabulation means generating false memories without intent to deceive, which is what LLMs do. They can't hallucinate because they don't perceive. 'Hallucination' caught on, but it's more metaphor than precision.


I've found this "book" (series of jupyter notebooks) to be a fantastic course on the Kalman filter from basics to advanced topics. https://github.com/rlabbe/Kalman-and-Bayesian-Filters-in-Pyt...


They're all names of popular Kendrick Lamar songs. Don't ask me why though


Perhaps AI generated?


Or maybe the final moment was a sigh of acceptance and gratitude for the live he lived. Nobody knows but him.


On a similar note, I really hope that the AI companies that don't make it, but have invested a lot in curating and annotating high quality datasets, would release them to the public. Autonomous car and robotics companies in particular since that kind of data doesn't exist on the internet as abundantly as, say, natural language text.


If you want to gain familiarity with the kind of terminology you mentioned here, but don't have a background in graduate-level mathematics (or even undergrad really), I highly recommend Andrew Ng's "Deep Learning Specialization" course on Coursera. It was made a few years ago but all of the fundamental concepts are still relevant today.


Fei Fei Li and Andrej Karpathy's Stanford CS231N course is also a great intro to the basic of the math from an engineering forward perspective. I'm pretty sure all the materials are online. You build up from the basic components to an image focused CNN.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: