Definitely. Trying to use a cutting edge new product that is similar (but better) than an old product is a struggle. For that, you have to do it yourself basically. And we don't have evidence that this will improve.

Oh, yeah I agree on that. I think all we can say with fairly high certainty is that they will be a part of the workflow (like assemblers, compilers, intellisense, etc before it), but the standard agentic methods to improve performance are not practical.

That said, yeah very skeptical of agentic workflows. I don't let my AI save (or delete) anything I haven’t at least looked at.

It would be shocking to me if in 5 years we can’t buy a machine which can locally in real time perform better than the current state of the art models for less than 10k.

I think for a software engineer, the gains offered by AI are worth 10k.

Sure, but it's not super hard to run an LLM locally, except that the models with the best performance need a rather expensive machine to run. Those fundamentals will get better. Whether or not the AI service companies can be profitable is a different issue.

Services DO get cheaper regularly as long as they are compute bound. Computers do get faster regularly. That said, providers will charge as March as possible, so they won't lower costs if there are no alternatives.

I'll grant you agentic worklows tend to be expensive cash hemorrhages with limited results. But as coding assistants where everything is interactive with a programmer, that SHOULD get cheaper as hardware gets cheaper and models get more efficient.

Yeah.
1. Since when is accurately simulating a whole human brain not a huge achievement?
2. What about current technology makes a reasonable person think AIs are remotely similar to human brains?

Well there are time bounded variants that are not incomputable... but they do take exponential time which is practically uncomputable.

Ugh, but this weekend is vacation AND I need to get a camera version of my ITCS paper time and space efficient list decoding. It's just hard finding the energy. Editing papers is always the most tedious step and I'm a but burnt out from the tedium of my day job.

I mean, grant money. You don't have freedom if you don't have money.

But left to its own devices, current AI will not get very far. It just can't remember the whole code base, will write overly verbose code, will keep copying the same code instead of abstracting out common functions, etc.

I mean, AI is legitimately useful as a software engineer. It's very much like using advanced auto complete.

But you it’s got lots of problems. You have to manage the context window, and go's forbid you're to use anything cutting edge (especially if the API resembles an older product).

That reporter did a bad job. I don't know how big or small that is. Compared to what? The same area for farmland? The whole region? I don't have a point of reference.

We know livestock, especially beef, is very water intensive. We don’t have to eat meat anymore than we have to use the internet.

I did read most of it. All they did was increase the amount of water in the waste water system a tiny amount. (If rain goes into the waste water, it literally is the exact same problem).

The problem is how they handle the waste water.

I mean, this would be like blaming rain for the dirty lagoons. Like yeah more rain will cause more runoff (and this is frequently the thing which triggers deadly algal blooms). But the issue is the lagoons.

Basic argument: short people are smaller on average, use fewer resources. There's also less torque on your body, so can probably reduce large groups of health issues.

... and today I learned congress members get pensions at 5 years? I'm all for bigger salaries for elected officials, but as someone who practically has to build their own 401k for retirement, that seems unfair. My employer only matches up to 1% of my income.

... I think the current congress statistics disagree.