Janne Sinkkonen
@scellus.bsky.social
Data scientist, PhD, some sort of AI expert, psychologist; widely interested in science and societal craziness, even philosophy.
And www.is.fi/kotimaa/art-...
"Venäjältä öljylasteja hakevat tankkerit odottavat Suomenlahdella pitkiäkin aikoja tietoa määränpäästään, kerrotaan Fintraficilta. Sunnuntaina laivaliikennettä Suomenlahdella oli paljon."
"Venäjältä öljylasteja hakevat tankkerit odottavat Suomenlahdella pitkiäkin aikoja tietoa määränpäästään, kerrotaan Fintraficilta. Sunnuntaina laivaliikennettä Suomenlahdella oli paljon."
December 29, 2024 at 11:55 AM
And www.is.fi/kotimaa/art-...
"Venäjältä öljylasteja hakevat tankkerit odottavat Suomenlahdella pitkiäkin aikoja tietoa määränpäästään, kerrotaan Fintraficilta. Sunnuntaina laivaliikennettä Suomenlahdella oli paljon."
"Venäjältä öljylasteja hakevat tankkerit odottavat Suomenlahdella pitkiäkin aikoja tietoa määränpäästään, kerrotaan Fintraficilta. Sunnuntaina laivaliikennettä Suomenlahdella oli paljon."
Also one or more of those two or three tankers getting into trouble in storm would be a good story, an excuse for something "unconventional" happening.
December 29, 2024 at 11:54 AM
Also one or more of those two or three tankers getting into trouble in storm would be a good story, an excuse for something "unconventional" happening.
One thing is that the weather is getting pretty stormy on the Baltic Sea, especially tomorrow. (I have no idea whether this affects the ships or their timing, I'm a complete amateur there. But something like waiting for a day or two could make sense.)
December 29, 2024 at 11:51 AM
One thing is that the weather is getting pretty stormy on the Baltic Sea, especially tomorrow. (I have no idea whether this affects the ships or their timing, I'm a complete amateur there. But something like waiting for a day or two could make sense.)
More seriously, is that place somehow representative of anything? Is there a real trend? I've lately been surprised by the AI sentiment there.
December 28, 2024 at 1:21 PM
More seriously, is that place somehow representative of anything? Is there a real trend? I've lately been surprised by the AI sentiment there.
Hacker News is the new Bluesky of AI?
December 28, 2024 at 1:18 PM
Hacker News is the new Bluesky of AI?
But on a longer term (which I was thinking), meaning is hard to predict and the whole concept of work is questionable.
Anyway, if complexity, cognition and autonomy grow on the tech side but not in our heads, and there's a clear boundary between, the current arrangement of power is unstable.
Anyway, if complexity, cognition and autonomy grow on the tech side but not in our heads, and there's a clear boundary between, the current arrangement of power is unstable.
December 28, 2024 at 1:17 PM
But on a longer term (which I was thinking), meaning is hard to predict and the whole concept of work is questionable.
Anyway, if complexity, cognition and autonomy grow on the tech side but not in our heads, and there's a clear boundary between, the current arrangement of power is unstable.
Anyway, if complexity, cognition and autonomy grow on the tech side but not in our heads, and there's a clear boundary between, the current arrangement of power is unstable.
Oh it was originally about meaningful work. That's complicated: meaning comes from contents of the work, one's local social environment (still ok sometimes), also from big stories (less ok maybe). I'd think effects of AI could even be positive for the content, at least in the short term. (1/2)
December 28, 2024 at 1:11 PM
Oh it was originally about meaningful work. That's complicated: meaning comes from contents of the work, one's local social environment (still ok sometimes), also from big stories (less ok maybe). I'd think effects of AI could even be positive for the content, at least in the short term. (1/2)
I think it more like our infrastructure gets intelligence, we gradually give up autonomy with respect to it, and the end result is us giving up control willingly. Think markets, social media algorithms, soon capable personal assistants. Competition between species is a bad metaphor.
December 28, 2024 at 8:53 AM
I think it more like our infrastructure gets intelligence, we gradually give up autonomy with respect to it, and the end result is us giving up control willingly. Think markets, social media algorithms, soon capable personal assistants. Competition between species is a bad metaphor.
Plausible deniability etc., and at this point I wouldn't yet say it's even that ship. But overall, quite likely yes imo, too many coincidences lately, incl. 25th Dec as the date.
December 26, 2024 at 10:26 AM
Plausible deniability etc., and at this point I wouldn't yet say it's even that ship. But overall, quite likely yes imo, too many coincidences lately, incl. 25th Dec as the date.
Reposted by Janne Sinkkonen
Even that bare ascii representation seems to be challenging due to perception problems, incl. tokenization.
anokas.substack.com/p/llms-strug...
anokas.substack.com/p/llms-strug...
LLMs struggle with perception, not reasoning, in ARC-AGI
What made o3 so much better than previous models on this benchmark?
anokas.substack.com
December 26, 2024 at 7:30 AM
Even that bare ascii representation seems to be challenging due to perception problems, incl. tokenization.
anokas.substack.com/p/llms-strug...
anokas.substack.com/p/llms-strug...
Even that bare ascii representation seems to be challenging due to perception problems, incl. tokenization.
anokas.substack.com/p/llms-strug...
anokas.substack.com/p/llms-strug...
LLMs struggle with perception, not reasoning, in ARC-AGI
What made o3 so much better than previous models on this benchmark?
anokas.substack.com
December 26, 2024 at 7:30 AM
Even that bare ascii representation seems to be challenging due to perception problems, incl. tokenization.
anokas.substack.com/p/llms-strug...
anokas.substack.com/p/llms-strug...
Mainly laugh, Stamets is... something else.
December 25, 2024 at 8:47 PM
Mainly laugh, Stamets is... something else.
I bet their looking for "root cause" in terms of functional imbalances and such would have obscured my underlying ABCG8 mutation even better than the conventional approach did. Just digging deeper would have been more beneficial. (A whole-genome test I ordered myself eventually did it.)
December 22, 2024 at 8:17 PM
I bet their looking for "root cause" in terms of functional imbalances and such would have obscured my underlying ABCG8 mutation even better than the conventional approach did. Just digging deeper would have been more beneficial. (A whole-genome test I ordered myself eventually did it.)
Yeah found that out later but didn’t care to correct. Non-images anyway.
December 22, 2024 at 3:53 PM
Yeah found that out later but didn’t care to correct. Non-images anyway.
Opinions are all over the place but clustered. x.com/euxoa/status...
On Bluesky: it's not healthy to isolate folks with clearly different priors, culture and epistemic environment to a separate site to make the media environments differentiate even more, but 🤷.
On Bluesky: it's not healthy to isolate folks with clearly different priors, culture and epistemic environment to a separate site to make the media environments differentiate even more, but 🤷.
x.com
x.com
December 22, 2024 at 1:03 PM
Opinions are all over the place but clustered. x.com/euxoa/status...
On Bluesky: it's not healthy to isolate folks with clearly different priors, culture and epistemic environment to a separate site to make the media environments differentiate even more, but 🤷.
On Bluesky: it's not healthy to isolate folks with clearly different priors, culture and epistemic environment to a separate site to make the media environments differentiate even more, but 🤷.
There's also the drift towards attention seeking, esp. in titles. NYT is still decent with its titles (and content too as far as I can tell), but in Finland at least, all media is now clickbait. Even the national broadcasting company (yle.fi) is mostly clickbait, and drifting to POV journalism.
December 22, 2024 at 9:58 AM
There's also the drift towards attention seeking, esp. in titles. NYT is still decent with its titles (and content too as far as I can tell), but in Finland at least, all media is now clickbait. Even the national broadcasting company (yle.fi) is mostly clickbait, and drifting to POV journalism.
I'm thinking media (NYT etc.), not experts. That media is seen now against the background of sea of "information" in internet and social media is a good point. It may indeed be that the mainstream media has always been a bit lost and biased, those imperfections are now just more visible.
But...
But...
December 22, 2024 at 9:54 AM
I'm thinking media (NYT etc.), not experts. That media is seen now against the background of sea of "information" in internet and social media is a good point. It may indeed be that the mainstream media has always been a bit lost and biased, those imperfections are now just more visible.
But...
But...
o* models are nerdish, good for problems where there is a well-defined solution.
Claude Sonnet (new) meanwhile has a distinct quality that I prefer for general use and for generic coding. It is sophisticated somehow, gets nuances and expresses them too.
Claude Sonnet (new) meanwhile has a distinct quality that I prefer for general use and for generic coding. It is sophisticated somehow, gets nuances and expresses them too.
December 22, 2024 at 9:25 AM
o* models are nerdish, good for problems where there is a well-defined solution.
Claude Sonnet (new) meanwhile has a distinct quality that I prefer for general use and for generic coding. It is sophisticated somehow, gets nuances and expresses them too.
Claude Sonnet (new) meanwhile has a distinct quality that I prefer for general use and for generic coding. It is sophisticated somehow, gets nuances and expresses them too.
I think o3 proves the point of o* series beyond any doubt.
But on that "naming scheme", 4, 4o, o1, o3, ... soon only "o4 (new)" and "4o (new)" are missing. ;)
Arena is an unknown goal, we don't exactly know how biased the prompters are, but can well believe 4o is better for "general use" than o*.
But on that "naming scheme", 4, 4o, o1, o3, ... soon only "o4 (new)" and "4o (new)" are missing. ;)
Arena is an unknown goal, we don't exactly know how biased the prompters are, but can well believe 4o is better for "general use" than o*.
December 22, 2024 at 9:23 AM
I think o3 proves the point of o* series beyond any doubt.
But on that "naming scheme", 4, 4o, o1, o3, ... soon only "o4 (new)" and "4o (new)" are missing. ;)
Arena is an unknown goal, we don't exactly know how biased the prompters are, but can well believe 4o is better for "general use" than o*.
But on that "naming scheme", 4, 4o, o1, o3, ... soon only "o4 (new)" and "4o (new)" are missing. ;)
Arena is an unknown goal, we don't exactly know how biased the prompters are, but can well believe 4o is better for "general use" than o*.
o1 is not on the Arena, and o1-mini and o1-preview are on top in math and coding as they are supposed to be. o* series doesn't give advantage on humanities-style non-reasoning problems where Claude Sonnet 3.5 (new) is probably still the best.
December 22, 2024 at 8:35 AM
o1 is not on the Arena, and o1-mini and o1-preview are on top in math and coding as they are supposed to be. o* series doesn't give advantage on humanities-style non-reasoning problems where Claude Sonnet 3.5 (new) is probably still the best.
An economist would say that's not an equilibrium. ;) I mean, one can predict something about the long-term feasibility of those intermediary representations.
December 16, 2024 at 7:03 PM
An economist would say that's not an equilibrium. ;) I mean, one can predict something about the long-term feasibility of those intermediary representations.