Yes to version control with LLM tools! FYI, Aider already has it integrated. In fact, use version control for everything you do on the computer. (Then backup your repo, need I say it?)

"This validates the 'do more with less' strategy..." I would say, "supports," not "validates." I use LLM tools every day to support my code development. It is not so easy to ensure good enginering while using an LLM based tool. While I feel that I am faster with the tools, I doubt it is really possible to produce quality output wrangling 5 sessions on 5 pieces of system code at a time.

"Microchimeric cells should trigger rejection but do not. Higher-than-typical concentrations have been found in people with autoimmune conditions..." Well, then, that could be evidence that they DO trigger rejection, just not at a 100% rate. I really wish writers would be more careful about their wording.

"The vacuum of space also means that contaminants can't sneak in." Not true. The vacuum just means they aren't floating around suspended by an atmosphere. Contaminants can still move around; they will simply move balistically, like tiny rocks that got kicked or thrown. They can "sneak in" if they are moving in the right direction.

Good points! Besides, the article points out "one week" of AI and "38.5 hours" of human work.. In my book, that is one week of a person sitting and using a tool. The peraon is responsible here, and did that person practise good engineering? If the goal was to boot first time, then yes. If the goal was to satisfy significant reliability and manufacturability requirements along with form and function... probably not, I imagine.

.. if i search for blue diode lasers?? what app will it make? blue-ray mastering? thin film gold micromaching? this seems rediculous. what if I search the form and structure of fossilized diatoms? What app does it make for that? schedule your pool maintenance visits? this is just as useless as targeted advertising...

The point is, whatever truth there may be behind that statement, the author 1) did not jusyify it (when he could have), and 2) used vagur wording that evokes emotional connotations (though he knows very well how to be precise).

"Ruby, you mightâ(TM)ve guessed, is dynamically typed. Python and JavaScript are too, but over the years, those communities have developed sophisticated tools to make them behave more responsibly." Make? Them (..who)? Behave? Responsibly? The heck is this guy saying there, with neither example nor description? Attacking his target audience savagely while excusing his (larger) favored readership without explanation, in vague language... Classic flamebait. I think this author knows what his goal is, and makes it clear.

your point resonates with AI as a research discipline, and many of its outcomes. but here we have an LLM/GPT based product being marketed as "AI", or even as "an AI", which is different than talking about AI as a research area.

it's not an AI, it's an LLM being marketed as an AI. on the other hand, if it /were/ an actual AI, it could simply be /convinced/ to spy, steal and damage!

Gamer PCs can run inference on an open source LLM locally with good response time... As a society, we really only need datacenters for the model contruction (still an important issue). The idea that this infrastructure is necessary for end use is just corporate greed trying to prevent you learning the truth. If the world economy holds on another few years, our [new] laptops will have the right chips. The inference side of the game is going to fan out away from datacenters anyway.

I still have two 32 bit systems in constant use. One I use as a RAID NAS for personal archive. It has 500MB, works great but has not been able to run web browsers for a long time. This impacts only when I want to install something, which is rare but happens. The other is a laptop with 4GB, and runs (er, ran) Firefox great. I use this laptop all the time, it is my main "walk around" laptop. Sorry to see Firefox leave it.

Thanks! My point of view, LLMs are a narrow part of natural language processing, and I don't understand them to get context sensitivity right, in the sense of extracting or expressing any meaning. Even though they produce strings that may conform to a context sensitive grammar. (Also, I don't understand the term "AI complete" to be rigorous...) But I DO understand LLM operation to be a type of computation, somewhat useful in a small subset of the ways that an "AI" would be useful. LLMs have nothing to be ashamed of. There is no reason to conflate them with intelligence.

They are "language models".. it is in the name. Models that simply predict likely text to follow example text - based on statistics, not "thought". You can do a lot of tricks, like have your example text say that the next text will describe a thought process... but it does not; this is simply a trick of language. I encourage everyone to refuse to repeat the rather misleading marketing term "AI", and just use one of the true terms "LLM", "GPT", or the actual brand name (Copilot, Gemini, ...) when discussing these machines.

At the end of the article this advice is given: Never use your primary WiFi network for IoT devices. I don't need an article to know that. What I think is worthwhile, clearly a lot of effort went into this simply for the sake of investigation, to find EXACTLY what the situation was, after the basic facts were obvious. (Also clearly for a lot of hacking fun!) I would have enjoyed a more technical dive in the article...