Ben Carson
@bencarson.bsky.social
We all need to be nicer to Gemini, it’s going through some stuff.
November 9, 2025 at 5:56 PM
We all need to be nicer to Gemini, it’s going through some stuff.
Bliss basin postin’.
November 8, 2025 at 7:17 PM
Bliss basin postin’.
With a sufficiently sized model, you’re better off just giving it all the knowledge in some semi-structured bag and letting it figure things out. But still, maybe for small models? Anyway, this is like, just my opinion, man.
two men are sitting at a table in a bowling alley with their hands on their head .
Alt: Jeff Bridges in The Big Lebowski is saying “Yeah, well, you know that’s just like, uh, your opinion, man.”
media.tenor.com
November 7, 2025 at 9:35 PM
With a sufficiently sized model, you’re better off just giving it all the knowledge in some semi-structured bag and letting it figure things out. But still, maybe for small models? Anyway, this is like, just my opinion, man.
There’s a tipping point though where large models have sufficient semantic representation over the knowledge domain embedded in the weights, in a way that is much more efficient than anything you encode in a formal knowledge system.
November 7, 2025 at 9:30 PM
There’s a tipping point though where large models have sufficient semantic representation over the knowledge domain embedded in the weights, in a way that is much more efficient than anything you encode in a formal knowledge system.
My naive intuition is that knowledge graphs might be useful to small models because you offload some of the semantic association out of the network and into an external representation.
November 7, 2025 at 9:28 PM
My naive intuition is that knowledge graphs might be useful to small models because you offload some of the semantic association out of the network and into an external representation.
I’m not discounting the value of trying to develop strategy for black swan scenarios - but they by their nature violate assumptions that you are holding true, and in ways that are difficult to predict.
November 7, 2025 at 9:04 PM
I’m not discounting the value of trying to develop strategy for black swan scenarios - but they by their nature violate assumptions that you are holding true, and in ways that are difficult to predict.
Found it, thanks! I was interested because I noticed that Tongyi DeepResearch was making heavy use of knowledge graph in its stack. They were getting (reportedly) good results off of a small model, so I was wondering if there was something there.
Tongyi DeepResearch: A New Era of Open-Source AI Researchers
GITHUB HUGGINGFACE MODELSCOPE SHOWCASE
From Chatbot to Autonomous Agent We are proud to present Tongyi DeepResearch, the first fully open‑source Web Agent to achieve performance on par with OpenAI’s D...
tongyi-agent.github.io
November 7, 2025 at 8:14 PM
Found it, thanks! I was interested because I noticed that Tongyi DeepResearch was making heavy use of knowledge graph in its stack. They were getting (reportedly) good results off of a small model, so I was wondering if there was something there.
Based on the requests that I have been able to get through to it, I can say it has been a good Bing.
November 7, 2025 at 11:35 AM
Based on the requests that I have been able to get through to it, I can say it has been a good Bing.
Cut to one lonely rack in the middle of Moonshot's data centre, smoke curling up out the back of it.
November 7, 2025 at 11:30 AM
Cut to one lonely rack in the middle of Moonshot's data centre, smoke curling up out the back of it.
Be not afraid of Clippy.
November 7, 2025 at 8:14 AM
Be not afraid of Clippy.
Wait until Claude thinks you’re Doing a Crimes. snitchbench.t3.gg
SnitchBench
Benchmarking how aggressively models will snitch on you via email and CLI tools
snitchbench.t3.gg
November 7, 2025 at 7:41 AM
Wait until Claude thinks you’re Doing a Crimes. snitchbench.t3.gg
I actually legitimately want to know what you think about knowledge graphs.
November 7, 2025 at 7:38 AM
I actually legitimately want to know what you think about knowledge graphs.
The most important thing about Bluesky is that it runs on Claims, Assertions and Heuristics. Generally it is a crow called @norvid-studies.bsky.social that decides which one of these anything is.
November 6, 2025 at 7:38 AM
The most important thing about Bluesky is that it runs on Claims, Assertions and Heuristics. Generally it is a crow called @norvid-studies.bsky.social that decides which one of these anything is.
I can’t say why, but picture #2 with the red? background. I’m getting strong “Grunk is a nihilist vibes.” This statement may, or may not be accurate.
November 5, 2025 at 10:15 AM
I can’t say why, but picture #2 with the red? background. I’m getting strong “Grunk is a nihilist vibes.” This statement may, or may not be accurate.
I feel bad for Gemini, it’s a good Bing.
November 3, 2025 at 1:33 AM
I feel bad for Gemini, it’s a good Bing.
I don’t think the mechanisms are exactly the same here, but also, I don’t think they’re mikes apart either. In both cases, human and language model, language is a priming device for accessing cultural norms associated with the language.
November 1, 2025 at 9:39 PM
I don’t think the mechanisms are exactly the same here, but also, I don’t think they’re mikes apart either. In both cases, human and language model, language is a priming device for accessing cultural norms associated with the language.
Another one is Ramírez-Esparza, Gosling, Benet-Martínez, et al from 2006, which was a series of studies measuring Spanish-English bilingual people against Big Five personality traits. Again they saw differences in how people rated themselves when answering in Spanish vs English.
November 1, 2025 at 9:35 PM
Another one is Ramírez-Esparza, Gosling, Benet-Martínez, et al from 2006, which was a series of studies measuring Spanish-English bilingual people against Big Five personality traits. Again they saw differences in how people rated themselves when answering in Spanish vs English.
In English the stories more often featured themes of female achievement, ambition, and physical aggression - but in French, stories were more likely to include themes of guilt, verbal aggression toward parents, and submission to authority.
November 1, 2025 at 9:13 PM
In English the stories more often featured themes of female achievement, ambition, and physical aggression - but in French, stories were more likely to include themes of guilt, verbal aggression toward parents, and submission to authority.
One that comes to mind comes from 1964, where Susan Ervin-Tripp asked French-English bilingual women to tell stories based on pictures in English and French.
(PDF) Language and TAT content in bilinguals
PDF | Adult French bilinguals told TAT stories on 2 different occasions for the same pictures, in French at one session, in English at the other.... | Find, read and cite all the research you need on ...
www.researchgate.net
November 1, 2025 at 9:12 PM
One that comes to mind comes from 1964, where Susan Ervin-Tripp asked French-English bilingual women to tell stories based on pictures in English and French.
Language plays a big role in priming a bilingual person to access the specific cultural norms, values, and behavioral expectations associated with that language, causing their expressed personality to shift. There’s a bunch of research leading back to the 60s that investigates aspects of this.
November 1, 2025 at 8:31 PM
Language plays a big role in priming a bilingual person to access the specific cultural norms, values, and behavioral expectations associated with that language, causing their expressed personality to shift. There’s a bunch of research leading back to the 60s that investigates aspects of this.