ninohardt.bsky.social
Marketing scientist, market researcher, choice modeler.
web: ninohardt.com
51 posts
58 followers
320 following
Regular Contributor
Active Commenter
comment in response to
post
So if unemployment↑ and inflation↑ what will Fed do?
comment in response to
post
Training for a new domain, with verifiable answers / solutions
comment in response to
post
@weightsbiases.bsky.social about to rename?
comment in response to
post
MarkText can help with cleaning up markdown, it's a good editor. But I just use Cursor, even on .md files in an Obsidian vault.
comment in response to
post
Notion? It's plain markdown, you can link between the markdown notes, and you can share.
comment in response to
post
Lex Fridman-esque
comment in response to
post
In consumer good context, 10.1287/mksc.2022.1371 show that generational share gap is mostly due to experience, and less due to intrinsic differences.
comment in response to
post
Yes, like that. Maybe you could categorize the controls.
comment in response to
post
With enough controls, can you make the cohort effect vanish?
comment in response to
post
I understand the sentiment, but can we just all stop anthropomorphizing AI? It's not helpful.
comment in response to
post
"applying theory of mind" sounds odd indeed. Just what do cognitive scientists make of that example? And what are the latest ToM & LLM papers worth reading?
comment in response to
post
"summary" is idiosyncratic yet people pretend it's not
comment in response to
post
In written, it works as expected.
comment in response to
post
NVIDIA reported INT8 for Jetson Orin, now FP4 for Digits. Who are they kidding?
comment in response to
post
FP4 ....
comment in response to
post
"good prompts for your data" is the new "let's train on the test set"
comment in response to
post
"democratization" ... with a censored model. Oh, the irony.
comment in response to
post
Interesting: when you ask for illegal drug recipes, deepseek behaves more like other models, it just writes out that it won't help. It's a different mechanism.
comment in response to
post
On a number of predictably "sensitive" topics, the model first generates and then abruptly deletes already generated text to say "beyond scope" or something.
comment in response to
post
What is "most"? >50%? And what is the list of "tasks"?
comment in response to
post
quoting from that page "Passing ARC-AGI does not equate to achieving AGI, and, as a matter of fact, I don't think o3 is AGI yet. o3 still fails on some very easy tasks, indicating fundamental differences with human intelligence."
comment in response to
post
Instead of just quoting "breakthrough", maybe also quote "Passing ARC-AGI does not equate to achieving AGI, and, as a matter of fact, I don't think o3 is AGI yet."
comment in response to
post
That was their intention, I guess. Well, it's still a useful device for edge applications, but I dislike that nvidia sometimes omits those kind of details ...
comment in response to
post
*INT8
comment in response to
post
veo veo ... que ves
comment in response to
post
The number of humans on the planet is at an ATH. They'll come up with some actually new things. It may take a little longer than the average time between new models on HF, but still doesn't mean we are at "peak" data.
comment in response to
post
Isn't that kinda insulting to anyone writing a book (or a paper, or ...) today? The data they generate doesn't count?
comment in response to
post
I will say that personas, especially those in the paper, are quite shallow. No way you get the same entropy as sampling real people. But is that really an issue for their use cases?
comment in response to
post
It is about diversity, not recovering actual personas or representativeness. Just don't anthromophize it so much.
comment in response to
post
Just using aistudio.google.com/live, sharing a lichess screen with it, and I can discuss my next move. Amazingly responsive.
comment in response to
post
Ok, running it again it beats Nova Pro to it:
comment in response to
post
comment in response to
post
comment in response to
post
comment in response to
post
comment in response to
post
Got it in regular "Plus", not sure what the limit is yet.
comment in response to
post
Wasn't AI supposed to be the great equalizer?
comment in response to
post
AI is also good at helping to cut through BS, and we can see Gemini clearly identifies the danger in the fake leadership rules.