Scythia Marrow
@scythiamarrow.bsky.social
I’m a scientist who loves history and writing. On here I mostly boost people way cooler than me, but I post my own work on occasion.
I freelance as an ML engineer and am currently: open for new part time contracts.
I freelance as an ML engineer and am currently: open for new part time contracts.
Don’t sweat it, and agreed! General intelligence is really new and we don’t know how to optimize it or really where the limits are. Optimists think that we’ll have alphago but for everything, and pessimists think we’ve already peaked.
We can all agree that current methods are not sufficient.
We can all agree that current methods are not sufficient.
December 15, 2025 at 3:13 PM
Don’t sweat it, and agreed! General intelligence is really new and we don’t know how to optimize it or really where the limits are. Optimists think that we’ll have alphago but for everything, and pessimists think we’ve already peaked.
We can all agree that current methods are not sufficient.
We can all agree that current methods are not sufficient.
… I’ve literally worked in a logical AI research lab. Symbolic manipulation is *way* less efficient than language models because of combinatorial explosion.
If we could have solved IMO with symbolic methods we would have solved IMO with symbolic methods. We couldn’t because math requires reasoning.
If we could have solved IMO with symbolic methods we would have solved IMO with symbolic methods. We couldn’t because math requires reasoning.
December 15, 2025 at 3:00 PM
… I’ve literally worked in a logical AI research lab. Symbolic manipulation is *way* less efficient than language models because of combinatorial explosion.
If we could have solved IMO with symbolic methods we would have solved IMO with symbolic methods. We couldn’t because math requires reasoning.
If we could have solved IMO with symbolic methods we would have solved IMO with symbolic methods. We couldn’t because math requires reasoning.
Such a method is actually the core principle behind the LLMs that achieved gold at the IMO, which requires advanced understanding. So yes it does work with log operations and integration.
December 15, 2025 at 2:12 PM
Such a method is actually the core principle behind the LLMs that achieved gold at the IMO, which requires advanced understanding. So yes it does work with log operations and integration.
I’m a literal godamned ML scientist. Here, have evidence directly contradicting you. The first query calls python, the second uses the long multiplication method in pure language. The model succeeds in both cases. g.co/gemini/share...
Gemini - direct access to Google AI
Created with Gemini
g.co
December 15, 2025 at 1:37 PM
I’m a literal godamned ML scientist. Here, have evidence directly contradicting you. The first query calls python, the second uses the long multiplication method in pure language. The model succeeds in both cases. g.co/gemini/share...
That is a wild ass claim. Of course LLMs can do basic arithmetic. Have you ever asked one to do a simple multiplication or solve a simple linear equation? They get it right every time and have for years.
December 11, 2025 at 11:21 PM
That is a wild ass claim. Of course LLMs can do basic arithmetic. Have you ever asked one to do a simple multiplication or solve a simple linear equation? They get it right every time and have for years.
Wow I hate your boss now.
December 10, 2025 at 4:52 PM
Wow I hate your boss now.
Älskling, dom ligger i klädkammaren där dom alltid ligger.
Sortera din tvätt >->
Sortera din tvätt >->
November 26, 2025 at 10:03 PM
Älskling, dom ligger i klädkammaren där dom alltid ligger.
Sortera din tvätt >->
Sortera din tvätt >->
To think is to make choices, which means to trade off between conflicting drives.
Doing so without becoming either catatonic from having nothing to drive you or paralyzed from overthinking is the core ability that meat has and computers struggle with, but will achieve relatively soon.
Doing so without becoming either catatonic from having nothing to drive you or paralyzed from overthinking is the core ability that meat has and computers struggle with, but will achieve relatively soon.
November 24, 2025 at 10:15 PM
To think is to make choices, which means to trade off between conflicting drives.
Doing so without becoming either catatonic from having nothing to drive you or paralyzed from overthinking is the core ability that meat has and computers struggle with, but will achieve relatively soon.
Doing so without becoming either catatonic from having nothing to drive you or paralyzed from overthinking is the core ability that meat has and computers struggle with, but will achieve relatively soon.
They’ll be here tonight too! You haven’t missed them.
November 12, 2025 at 9:57 PM
They’ll be here tonight too! You haven’t missed them.
The previous Trump administration was fascist, this one is a Nazi one. Each meets all the technical requirements so it would be sane washing to call them anything less.
Don’t hesitate to say obviously true things like 1+1=2 or gravity exists and roughly obeys newton’s law of gravitation.
Don’t hesitate to say obviously true things like 1+1=2 or gravity exists and roughly obeys newton’s law of gravitation.
November 8, 2025 at 1:10 PM
The previous Trump administration was fascist, this one is a Nazi one. Each meets all the technical requirements so it would be sane washing to call them anything less.
Don’t hesitate to say obviously true things like 1+1=2 or gravity exists and roughly obeys newton’s law of gravitation.
Don’t hesitate to say obviously true things like 1+1=2 or gravity exists and roughly obeys newton’s law of gravitation.
Do it! At the very least will be a good mark on their performance review.
If you are worried about the social norms you can just praise them in general and recommend them for promotion instead of phrasing it as a request.
If you are worried about the social norms you can just praise them in general and recommend them for promotion instead of phrasing it as a request.
October 28, 2025 at 8:47 PM
Do it! At the very least will be a good mark on their performance review.
If you are worried about the social norms you can just praise them in general and recommend them for promotion instead of phrasing it as a request.
If you are worried about the social norms you can just praise them in general and recommend them for promotion instead of phrasing it as a request.
It’s like 99% access to scale. Without some means to test hypotheses theory stagnates.
The 1% is that people with access to scale don’t want to make a mind, they want to make a product, and the people without access to scale are human chauvinists.
We’ll unwittingly make an artificial mind by 2030.
The 1% is that people with access to scale don’t want to make a mind, they want to make a product, and the people without access to scale are human chauvinists.
We’ll unwittingly make an artificial mind by 2030.
October 27, 2025 at 9:43 PM
It’s like 99% access to scale. Without some means to test hypotheses theory stagnates.
The 1% is that people with access to scale don’t want to make a mind, they want to make a product, and the people without access to scale are human chauvinists.
We’ll unwittingly make an artificial mind by 2030.
The 1% is that people with access to scale don’t want to make a mind, they want to make a product, and the people without access to scale are human chauvinists.
We’ll unwittingly make an artificial mind by 2030.