I often fail to notice it on the first look of some snippet of code

If there's infinite star trek episodes, I don't care to watch it. I'd rather spin a fidget spinner instead.

If anything will be missing from juniors it will be the ability to run code in their heads if they've only written code via AI.

The article's topic is brain to AI interfaces. She mentions this quote as a "vision" from 2035, where not only can thoughts inform AI but AI can respond in thought too.

When we get used to a good tool it becomes an extension of ourselves.

How is this AI psychosis? Is it that it's very unrealistic for such an interface to be created in the next 10 years? Or is it that there's a fear of letting AI interfere with one's thoughts lest they lose themselves?

The place to find them would be performance profiles, query plans, telemetry. The guidance for perf still applies, measure before and after change.

The issue is that the code often does not contain the information to do a perf optimization. Eg. you can't tell your cache size, the volumes of data in your DB or the latency of your network through just the text. Should you provide this context, you can get better results.

So in this case, it matters that spiders are not in fact insects in the strict sense.

Also about "creating" means of production, these companies actively try to sabotage this as another profit maximization strategy. They buy all the ram, so others cannot compete. They buy startups who succeed, so they stop competing.

It's not aspersions, it's just describing the phenomenon.

Even if I take your suggestion to heart, once my company would be big enough, if I wanted to optimize for profit, I would have to do the same as these companies.

The end result is concentrated power.

However the result is exactly the same, concentration of power.

Also you are saying you are also working as a senior SWE. How are you so involved with 55 tenants and balance a full-time job at the same time? I've known people with 1 tenant who needed days off to deal with their issues, I find it hard to imagine personally dealing with 55.

Honestly curious here

My personal strategy is specialization, because I've noticed that I cannot work with AI in hard domains that I haven't got a background in. For example, I've tried to use claude opus to understand a phd thesis in quantum physics from a friend. My background is in engineering and later compilers/static analysis. Despite opus helping a lot to give me the high level idea behind it, I could see a few problems with my lack of background: - I couldn't verify if opus was right or wrong in its explanations - Even when asking the AI to simplify explanations, there was so much prerequisite knowledge I needed to absorb that there was little point to bother with its output - It was very hard to collaborate with the AI to understand what predictions can be made from the PhD. The model could produce a lot of output but I didn't really understand its answers. Asking follow-up questions did not really solve the issue because it felt like I was missing a mental model.

My thinking is that, no matter how much LLM intelligence grows, for sectors with inherent complexity, people will still need specialized expertise to understand, evaluate and use their outputs. I also doubt that the most efficient future is one where humans don't understand the outputs and delegate everything, because it will be both be hard to understand if the machines are aligned to the benefit of their users.

So specialization sounds like a good strategy for the future.

I'd like to hear some opinions around this topic. Have you observed the same? Do you disagree based on other experiences/data?

Comments URL: https://news.ycombinator.com/item?id=48544117

Points: 2

# Comments: 0

My personal strategy is specialization, because I've noticed that I cannot work with AI in hard domains that I haven't got a background in. For example, I've tried to use claude opus to understand a phd thesis in quantum physics from a friend. My background is in engineering and later compilers/static analysis. Despite opus helping a lot to give me the high level idea behind it, I could see a few problems with my lack of background: - I couldn't verify if opus was right or wrong in its explanations - Even when asking the AI to simplify explanations, there was so much prerequisite knowledge I needed to absorb that there was little point to bother with its output - It was very hard to collaborate with the AI to understand what predictions can be made from the PhD. The model could produce a lot of output but I didn't really understand its answers. Asking follow-up questions did not really solve the issue because it felt like I was missing a mental model.

My thinking is that, no matter how much LLM intelligence grows, for sectors with inherent complexity, people will still need specialized expertise to understand, evaluate and use their outputs. I also doubt that the most efficient future is one where humans don't understand the outputs and delegate everything, because it will be both be hard to understand if the machines are aligned to the benefit of their users.

So specialization sounds like a good strategy for the future.

I'd like to hear some opinions around this topic. Have you observed the same? Do you disagree based on other experiences/data?

Comments URL: https://news.ycombinator.com/item?id=48544106

Points: 2

# Comments: 2

The analogy I've had for myself is that it feels like using a bulldozer to dig rather than a shovel. If you use it to dig archaeological artifacts, it can make things worse than you started. A lot of the work however, is just moving dirt around, so you are wasting time by using a shovel.

Basically integrate sqlglot to translate the schema between dialects and then use a base dialect for generating the schema.

The two tools seem complementary and you seem to be a better designer, so it would be nice to see it all together