It was not coding but if an AI cannot understand a simple request to extract one piece of information from a text document then I can guarantee it is going to have problems understanding the far more complex instructions that will be necessary to construct the code needed to do the analysis to extract that information in the first place.

It may be that those limits have expanded in the past few months although that example is only a month old. However, I'd be surprised if my original point does still not hold i.e. that the problem with AI coding is still going to be putting together instructions that are precise enough and that it can understand in order to get it to produce the code that you need. This will be especially true with code that may be doing something less common - like scientific analysis - for which it has either little or extremely poor quality (looking at my own and fellow physicists' code!) training data.

It should be given the choice. Either it gets regulated like a telecom provider in which case it cannot discriminate against content unless it is illegal and, in return, cannot be prosecuted for material on its servers OR it gets regulated like regular media in which case it has editorial control over the content it serves but is then legally liable for that content.

If we just assume that the AI can't write the code we need then the whole premise of the discussion, that we can program code in natural language using AI, is moot.

After speaking to several of my american colleagues a few weeks ago it is very clear that American Science is fighting for its life at the moment and there is definitely no surplus.

It depends on what you are doing and how specific you need your code to be. The more precise the harder, and harder it is to get the AI to do what you are asking in many cases. For example, when I asked ChatGPT what the statistical signfiicance of a specific scientific result in it said "very significant", when I told it to be more precise in its response it responded that it was a "high statistical significance". I then asked it to specify the number of sigma significance, it responded "more than 5 sigma" so I asked it to get the exact number of sigma to one decimal place (since the paper gave it as 5.9 sigma), the AI's response? "More than 5.0 sigma". I never got it to tell me the correct answer of 5.9 sigma.

If I have to fight with a coder-AI like that to get it to do what I want forget it - it will be vastly less frustrating just to write the code myself and traditional coding is fun even if it may be slower overall. Who wants to fight an AI with excellent training but the personality of a pedantic 2-year-old to get the job done?

Paying for the compute time and licensing of the code are not going to be the problems that stop this it will be programming the AI and then debugging the output. Generating the correct prompt for an AI is going to be much, much harder than people think. Sure, it is easy to generate a picture in your head of what you want an app to do but, at least in my experience, often the most consuming part of programming is deciding _exactly_ what you want your code to do i.e. converting the vague floppy idea in your head into a concrete set of instructions where all ambiguity is removed and everything is decided.

Having an AI involved might mean that a developer may not need to know the programming details (although I suspect it will be needed since AI code is far from bugfree) but they will still need to describe the program with complete precision e.g. what input parameters are needed, what the controls for each should look like, where they should be placed etc. because if you do not then not only will the AI make a complete mess of an interface but, everytime you tweak the prompt to fix an issue, the interface will likely completely change unless clearly specified.

Now multiply that by all the small decisions that developers make when coding an app and, while AI may help reduce the burden I doubt it is going to massively reduce it plus you have to contend with the issue that unlike a traditional programming language, one slight change to the prompt can radically shift how the AI codes things meaning that a minor change in one area could radically change another area unless the prompt has everything tied down precisely. So I think that developers are absolutely going to be needed but, if this works well, it may make them more productive as well as change their skillset somewhat.

I'm not a biologist so I have no idea if that would work but I would not call that cloning since it involves combining the genetic material of more than one individual organism to create a new, genetically distinct one.

300 million years is over 2% of the lifespan of the universe. If we take the average human lifespan to be about 80 years that would be a "blink of the eye" lasting 1.7 years which is quite a long blink.

I miss the days when you did not elect them as your president.

Not if there has already been time for supernovae to create heavier elements, like lead, and disperse them which the results indicate is the case here. Stars large enough to undergo supernovae have relatively short lives and, as the summary notes, the presence of heavier elements suggests that this galaxy was around for long enough for supernovae to have already happened meaning that there are even younger galaxies yet to be found.

Strains do have some, albeit very limited, genetic variation. Clones are worse because they have absolutely no genetic variation - assuming no errors in the clning process - making them even more susceptible to single diseases.

Worse, I see absolutely zero point in cloning pets. If we cloned our dog the new clone would lack any of the memories and knowledge of our current dog and since personality and behviour is a combination of both nature and nurture while the clone might end up being similar it would never be the same as our current pet, so in no way can any pet live forever. Identical twins may have similar, but different, personalities and we do not regard then as two copies of the same person!

No, the simplest version is everyone gets an A grade. Equity is all about ensuring equal outcomes and nothing ensures a more equal outcome than everyone getting the same grade.

I guess in reality the technological signularity is a little different from what some people were originally worrying about.

I think one of the things that is hard to tell as someone on the outside is where the failures start to become a serious problem. I get that SpaceX is doing things differently and that what they are doing is insanely complex and hard to get right. However, at some point if the failures continue it will put the program in jeopardy because you cannot keep producing and blowing up expensive hardware indefinitely even when you do learn a lot.

With the best will in the world, I also find it hard to believe any company that says it is all fine since there are many, many examples of companies doing that right up to the point where they collapse. So I hope that you are correct and that this is another of their planned-for losses and that the program will eventually be a success but it is hard not to be a bit concerned although I very much wish them the best of luck!

Just as long as they keep the PCMCIA port for my 56k modem and compact flash reader!