lambdaone

The problem with all of these, even the most recent one, is that they have the "AI look". People have tired of this look already, even for short adverts; if they don't want five minutes of it, they really won't like two hours of it. There is no doubt the quality has vastly improved over time, but I see no sign of progress in removing the "AI look" from these things.

Rust programmers are far more likely to have the vigilant mindset than C programmers, or they wouldn't be using Rust.

You can get an awful lot done very quickly in C if you aren't bothered about security - and traditionally, most of the profession has done exactly that.

And even better if someone can implement the whole massive spec securely...

Python type hints are static - at the moment, they are advisory only, but there is an obvious route forward to making Python an (optionally) fully statically typed language by using static type checking on programs before execution.

Python is a typed language. Perhaps you were trying to say something different?

And we'd be wrong about that. Different domains are showing wildly different characteristics, with some ML models showing superhuman or expert level performance in some domains (chess, face and handwriting recognition for example) and promising but as yet just not good enough in other domains (radiography, self-driving cars, question answering, prose writing). Currently coding is somewhere in the middle; superhuman in some ways, disappointingly unusable in others.

I don't we can make any conclusive verdict about the promise of ML for radiography right now; the life-critical nature of the application it's in the unusable middle, but it might get better in a few years or it might not. Time will tell.

The archive link above is broken: this is an earlier archived copy of that page with content intact:

https://web.archive.org/web/20230713101725/http://bactra.org...

Some pages have 'semi-protection' or 'extended semi-protection', where a user must have a Wikipedia account which must have made a certain number of edits and be older than a specified time period.

However, this isn't one of them.

You can edit it immediately, either with or without creating an account.

You are absolutely correct that the difference between y against x and x against y fitting perfectly demonstrates why the bias exists, but the correct way to remove the bias is not normalization, but to use a coordinate-independent regression technique.

See the other comments by many other commenters for details.

Also relevant to this topic:

Fred Gruenberger's Measure for Crackpots (1962):

https://www.rand.org/content/dam/rand/pubs/papers/2006/P2678...

John Baez's Crackpot Index (1998):

https://math.ucr.edu/home/baez/crackpot.html

This looks to me like it's "I Can't Believe It's Not Helvetica."

Arguably, all we do is something similar to hallucination; it's just that hundreds of millions of years have selected against brains that generate internal states that lead to counter-survival behavior.

I recently almost fell on a tram as it accelerated suddenly; my arm reached out for a stanchion that was out of my vision, so rapidly I wasn't aware of what I was doing before it had happened. All of this occurred using subconscious processes, based on a non-physical internal mental model of something I literally couldn't see at the moment it happened. Consciousness is over-rated; I believe Thomas Metzinger's work on consciousness (specifically, the illusion of consciousness) captures something really important about the nature of how our minds really work.

"Hallucinate" is a term of art, and does not imply a philosophical commitment to whether LLMs have minds. "Confabulation" might be a more appropriate term.

What is indisputable is that LLMs, even though they are 'just' word generators, are remarkably good at generating factual statements and accurate answers to problems, yet also regrettably inclined to generating apparenly equally confident counterfactual statements and bogus answers. That's all that 'hallucination' means in this context.

If this work can be replicated, it may offer a way to greatly improve the signal-to-bullshit ratio of LLMs, and that will be both impressive and very useful if true.

The title as posted is misleading and sensational: we should use the actual paper title, "H-Neurons: On the Existence, Impact, and Origin of Hallucination-Associated Neurons in LLMs", which is far less sensational.

It could be made much less vulnerable by dividing the gas storage balloon structures into a set of multiple smaller structures, ideally with some further internal partitioning so a single punctured surface does not release all the gas in that structure. Everything else is as 'hard' or 'soft' as any other piece of industrial plant, and can be hardened further by putting it inside a reinforced building or set of buildings, internal redundancy etc.

Very few people are excited by the "the potential of the AI PC" that Microsoft is trying to push, but rather infuriated by the attempt to buffalo them into replacing their perfectly good working hardware for no apparent gain in function. Microsoft clearly feel that the price of being hated for forcing this change is worth the revenue gained. They might be wrong.

I see this as complementary to other energy storage systems, including sodium ion batteries; each will have its own strengths and weaknesses. I expect energy storage density cost will be the critical parameter here, as this looks best suited to do diurnal storage for solar power systems near out-of-town predictable power consumers like data centers.

Very unlikely. All the technologies involved work best at scale; for example, the area-to-volume ratio of the liquid gas storage vessel is a critical parameter to keep energy losses low.

A theoretical study shows 77%, which is in the same ballpark: https://www.sciencedirect.com/science/article/pii/S136403212...

A few percent here of there is not that important if the input energy is cheap enough.

Fantastic detail, thank you.

This seems almost too good to be true, and the equipment is so simple that it would seem that this is a panacea. Where are the gotchas with this technology?

Clearly power capacity cost (scaling compressors/expanders and related kit) and energy storage cost (scaling gasbags and storage vessels) are decoupled from one another in this design; are there any numbers publicly available for either?

There are lots of Stratum 0 servers out there; basically anything with an atomic clock will do. They all count seconds independently from one another, all slowly diverging over time, with offset intervals being measured by mutual synchronization using a number of means (how is this done is interesting all by itself). Some atomic clocks are more accurate than others, and an ensemble of these is typically regarded as 'the' master clock.

To quote the ITU: "UTC is based on about 450 atomic clocks, which are maintained in 85 national time laboratories around the world." https://www.itu.int/hub/2023/07/coordinated-universal-time-a...

Beyond this, as other commenters have said, anyone who is really dependent on having exact time (such as telcos, broadcasters, and those running global synchronized databases) should have their own atomic clock fleets. There are thousands and thousands of atomic clocks in these fleets worldwide. Moreover, GPS time, used by many to act as their time reference, is distributed by yet other means.

Nothing bad will happen, except to those who have deliberately made these specific Stratum 0 clocks their only reference time. Anyone who has either left their computer at its factory settings or has set up their NTP configuration in accordance to recommended settings will be unaffected by this.

At this point, web design 'ecosystems' are essentially money whirlpools. They're complex, so they require programmers skilled in using them, who in turn make more sites which need more programmers, and so on, and the network effect takes over and cements this feedback loop in the structure of the jobs market.

And the frameworks are churned continuously and are also bug-ridden nightmares, so that continuous development and support is needed to keep websites functioning and secure.

Any reduction in framework complexity threatens the whole edifice.

SVG without <script> would do just fine.

I have always loved this work. A classic example of the effective application of mathematical principles to a computing problem.

I'm sure when this is all over there will be interesting stories to tell. Does anyone know why they failed?

Sadly such things do happen:

https://metro.co.uk/2017/02/20/gang-of-beggars-pictured-gett...

I look forward to you donating that $200 to a homeless charity, KingMob.

I have on several occasions offered to pay for food at a nearby sandwich shop or fast food place for beggars who were asking for money for food. None of them accepted the offer.

Claude is insanely good at grunt-work maintenance coding, which is a fairly formulaic exercise that mostly requires RTFM and simple code changes that look a lot likw the surrounding code. Designing new things from scratch based on human specs is something which Claude still struggles with.

It's presumably only a matter of time before there's a Wikipedia article for "Brightness war"