> We keep hearing that the latest LLMs are approaching AGI.
A gentle reminder if you keep hearing this, it means AGI hasn't happened and so the accusations that "approaching AGI" become more like "fusion in 50 years" than not bear fruit.
There is no inductive reasoning behind "approaching AGI" simply more agglutinative outcomes, which invite the statement.
Back when I did Psychology 101, there were many papers attacking IQ tests on the basis that it depended on which form of intelligence you were testing for. Philosophers cannot agree on how to define intelligence.
On that basis we can accept that "AI" can be good at some forms of intelligence and not so good on others. A human brain needs about 200W of power. AI data centers need a 100 million times that ... each. So next times your lights dim, just think that somewhere an AI is "thinking".
It’s all just cognitive dissonance. Anyone claiming optimism around salaries is just flat wrong. I see non-engineers contributing to production code and it’s simply a matter of time until leadership feels comfortable with downsizing and or lowering salaries.
"writing code" is a tiny part of what "a career in software" means. You haven't begun to understand the complexity of systems if you think the code part is the dominant part of this career space.
coding is thinking. coding is systematising. coding is reasoning. coding is optimising over generic or less optimal solutions. coding is understanding the complexities behind problems. coding is problem solving.
you think the "approaching AGI" is going to magically replace all the myriad ways this manifests in one go?
I agree most of what you said. But I see reasoning from the models. If we give enough context about different systems, I see latest model can able to thinking provide complex solutions.
I enjoy automating other people's livelihoods. I also enjoy figuring out clever ways to skirt around legislation, it's a fun exercise. As a hobby, I like to find new ways to collect more data to make surveillance more convenient.
> We keep hearing that the latest LLMs are approaching AGI.
A gentle reminder if you keep hearing this, it means AGI hasn't happened and so the accusations that "approaching AGI" become more like "fusion in 50 years" than not bear fruit.
There is no inductive reasoning behind "approaching AGI" simply more agglutinative outcomes, which invite the statement.
Back when I did Psychology 101, there were many papers attacking IQ tests on the basis that it depended on which form of intelligence you were testing for. Philosophers cannot agree on how to define intelligence.
On that basis we can accept that "AI" can be good at some forms of intelligence and not so good on others. A human brain needs about 200W of power. AI data centers need a 100 million times that ... each. So next times your lights dim, just think that somewhere an AI is "thinking".
> what keeps you motivated?
Learning new things. Discovering more things about myself. Deepening relationships with other people. Much joy to experience.
It’s all just cognitive dissonance. Anyone claiming optimism around salaries is just flat wrong. I see non-engineers contributing to production code and it’s simply a matter of time until leadership feels comfortable with downsizing and or lowering salaries.
The money is still good compared to breaking rocks.
"writing code" is a tiny part of what "a career in software" means. You haven't begun to understand the complexity of systems if you think the code part is the dominant part of this career space.
coding is thinking. coding is systematising. coding is reasoning. coding is optimising over generic or less optimal solutions. coding is understanding the complexities behind problems. coding is problem solving.
you think the "approaching AGI" is going to magically replace all the myriad ways this manifests in one go?
I agree most of what you said. But I see reasoning from the models. If we give enough context about different systems, I see latest model can able to thinking provide complex solutions.
> If we give
Therein lies your answer.
You have to give things for the model to do something. You have to give the relevant and appropriate things for the model to do the right thing.
I enjoy automating other people's livelihoods. I also enjoy figuring out clever ways to skirt around legislation, it's a fun exercise. As a hobby, I like to find new ways to collect more data to make surveillance more convenient.