Posts by Michael Wojcik

Re: Really?

This is blatant Neo-anti-neoism and we of the People's Neoist Front will not stay silent in the face of it. For shame, sir. For shame.

Re: Look North !!

The problem is data centers in daft locations like Phoenix, or elsewhere in the Great American Desert (as it used to be known), using up scarce freshwater resources. If that water then precipitates down over wetter parts of the country, or the ocean, that's still a big problem.

Using seawater would avoid this, but 1) you'd have to build data centers near the sea, rather than in Arizona¹ or Texas² or other terrible choices; and 2) evaporating seawater leaves behind all sorts of tricky stuff to deal with, particularly salt.

As for "captur[ing] the humidity": That would require either a great deal of additional energy input to move the heat somewhere else, or magical engineering. That steam will condense when it gets high enough in the atmosphere to dump its heat through radiation. How big a cooling tower do you think you can build? Again, that water will return to earth somewhere far away. (And, no, a bunch of data centers along the Pacific coast of the US will not put enough steam into the atmosphere to make a difference to the drought.)

¹Seriously, Arizona. What, was the Sahara not available?

²Where you can have simultaneous water and energy crises. Yay!

If we can waste enough of their time they become unprofitable.

Unfortunately I don't think that's practical. Few people will do it, and thanks to the low overhead of VoIP, bots, and minimally-compensated people working from home, these scams are profitable even with a very low hit rate. It might make some recipients feel better to waste the callers' time – and if so, go right ahead – but I doubt it has any noticeable effect on the bottom line.

I don't know when they started with macros

The important date here is 1999, when Kwyjibo wrote the Melissa virus for Word and Smith released it. Melissa was a pretty major event, hitting ~1M users and making the national news.

That it took Microsoft over two decades to properly restrict macro execution in Office products (after years of ineffective half measures) shows just how resistant the Office product-management team is to curtailing "features" that are actually serious security vulnerabilities.

Re: time diff

Yeah, it's not clear from the summary in the article what exactly the test was. I expect it was actually in effect "find the passage in the context window that doesn't match what you would expect for the next token", so it wasn't diffing Real TGG against Modified TGG; it was running Modified TGG against the entire model, which included somewhere in its parameter space a gradient matching Real TGG but not the actual text in a literal representation.

Not a hugely interesting experiment, as far as I'm concerned. Exactly what I'd expect a really large model to be able to do. So what?

Now, if the underlying model had been trained on a data set from which all copies, excerpts, and references to Real TGG had been removed, and it still caught the offending passage, that would be a slightly more interesting experiment. (It's feasible for a transformer LLM to do this, if there's enough similarity between the world of the novel and the world of the training set for most-probable completion to get a strong disagreement on the altered passage.) A better test would be to use a freshly-written unpublished novel, of course, so there's no possibility of data-set contamination. But even then, all you've confirmed is that the surface of parameter space contains a gradient that diverges sufficiently at the point where the out-of-place passage appears.

And that's a big problem with LLMs. They converge on a middle ground of expectation. They seek to reduce surprise, which is another way of saying they reduce information entropy in the output. They're bland. They have no style. They have no conversation, as we used to say of uninteresting people. They regurgitate the most likely continuation, in a dull fashion. You can anneal them into slightly higher valleys with prompting, but the existing models and their architectures fundamentally lack the inconsistency of human discourse. And that's what makes us interesting.

Counterpoint: Putting a PC on every middle-manager's desk destroyed the typing pool. Typist was a skilled trade, and replacing the typing pool with a PC and word-processing software cost those jobs, and it cost managers time because dictation to a skilled human is faster than hunt-and-peck typing into Microsoft Word or the like, and it reduced the quality of business prose because it was no longer trained professional writers producing the final copy.

There have been a number of studies which suggest the "PC revolution" was actually fairly expensive in terms of productivity.

On a similar note, giving spreadsheet software to bookkeepers and others who understood how to use paper spreadsheets was productive. Giving them to people with no idea how to use them correctly? Quite possibly not.

A CACM article on the 20th anniversary of Powerpoint (which presumably was published around 16 years ago, but I'm not going to go look for it) noted that in the '80s, similar presentations were generally either B&W overhead transparencies¹ or carefully-orchestrated multimedia presentations with synchronized slide projectors and tape decks that took many hours to create. Now Powerpoint Rangers generate zillions of fancy presentations every day with graphics! and animation! and mind-numbing stupidity! – which, yes, is a lot more output, but is it more value?

And I recall a Byte article from many years ago (obviously) about the "Fat Bits" option in Mac Paint (or whatever it was called): the zoom function, basically. The author suggested that having a zoom function, and being able to do pixel-by-pixel editing, led to people wasting a vast amount of time fiddling with details that no audience member was likely to notice, and thus offered essentially no return on investment.

Information technology has severe revenge effects, especially when it attracts a lot of attention² and triggers obsessive behavior in users.

¹Or "foils", if you worked at IBM, the Land of Our Own Damn Nomenclature, Live With It.

²One of the great ironies of the current LLM fervor is that it was touched off by a paper titled "Attention is All You Need". The use of "attention" as a term of art in transformer algorithms is an accidental gesture toward the greatest problem they currently cause.

I would know; I roomed with one senior year of college.

All AC posters are idiots. I would know; I just read your post.

Oh, is generalizing from a single example not reliable?

Re: re: Have faith - he seems like he knows what he’s doing.

These are antivaxxers. Jumping to unjustified conclusions is their specialty.

Re: Security fail by design

Who thought that auto linking, fetching and executing in mails was a good idea?

Borenstein and Freed started us down this particular crumbling cliffside path.

Admittedly, RFC 1341 was inspired partly by the need to support character sets outside ASCII, which is a legitimate problem. And 7.4.2 manages to list a surprising number of security issues with "active" content, for 1992; unfortunately it's clear few implementers gave this much thought.

With 5 of 48 orders analyzed. So it's probably more like $250M wasted,or 1/7. That sounds pretty unreasonable.

And that's just wasted in this fashion. How much waste for overpriced products? How much for systems that are not fit for purpose, or are significantly less productive than they should be?

That's up to the shareholders, and the shareholders apparently have decided not to do so.

I'd say I'm surprised that Apotheker has been appointed to a number of boards (at least two as chair) since the debacle, except really I'm not. Everyone knows that corporate boards are a club and you have to offend the other members to get kicked out. Merely being terrible at your job is regarded as a quirk.

Re: Psst Microsoft...

And they'll be happy to pay you once you deliver a working one. COD.

As someone noted above, this is not a risk. Microsoft has just promised to buy a little (for them) electricity at a reasonable price in the future, should it be available. Unless the price of electricity drops enormously by then, they're not taking on any risk.

I got an Android phone from Motorola and it came with the Facebook app preloaded.

Re: I suppose after ....

Trust experience, question everything else.

An impressively foolish maxim.

Personal experience is by definition anecdotal. The sample size of personal experience will be much too small to justify any generalizations for most categories of experience.

Humans are prey to a large number of well-documented perceptual and cognitive limitations and traps. Our ability to observe situations and draw rational conclusions from them is severely limited. That's why we have epistemological protocols for mitigating those limitations and not trusting personal experience.

Learning from experience is both necessary and unavoidable. But "trusting" it is the hallmark of uncritical thought.

Asynchronous media beat synchronous media.

Re: Productivity

Nor is the problem "time spent communicating". That can be just as productive as any other activity. I've known plenty of programmers who could take four hours to accomplish something that could have been done in ten minutes if they'd asked the right question of the right person.

This is a rubbish study, based on rubbish data and rubbish premises. Pure marketing fluff.

Re: No responsibility.

Secure Boot was always vulnerable to the theft of a private key. That's true for any security feature that relies on a secret.

Not that I'm saying Secure Boot was a good idea – I believe there are legitimate concerns with it. But this isn't due to a flaw in the design of Secure Boot; it's due to a flaw in MSI's security which let the private key be discovered and exfiltrated by attackers. It's not, in fact, a Microsoft bug at all. It's just exploited by malware written to attack Windows, and Microsoft are therefore providing a patch for it. (And that patch is problematic because key revocation is a hard problem.)

Re: What else is new?

Exactly what people have said about every other WordPress plugin vulnerability.

No one has to use them. But people do. This is not the fault of the WordPress developers, except that they opened the door.

There's no cheap, simple fix for this problem. "Don't use plugins" is not a fix, because the problem is other people deciding to use plugins. It's all just part of the tremendous mess the industry has made of the Web, starting with Netscape's decision to stick LiveScript into the browser, and Microsoft's to invent DHTML (compounded by Microsoft's invention of XHR, and Google's popularization of it).