* Posts by Michael Wojcik

12096 publicly visible posts • joined 21 Dec 2007

We talk to W3C board vice-chair Robin Berjon about the InterPlanetary File System

Michael Wojcik Silver badge

Re: Clueless

It wasn't a great interview for someone who's not already familiar with IPFS.

Personally, I'm not particularly fond of IPFS. The switch from authority-based locators to content-based locators is a different trade-off — it's not automatically superior. In practice, it's very useful to me to know what the authority portion of a URL is before I attempt to follow it (which is one reason why I don't use URL shorteners, and I rarely follow shortened URLs). The authority portion doesn't deserve a lot of trust, but it's a significant update to the model. With a CID you lose that completely.

And we already have a distributed peer-to-peer content-chunk distribution protocol that works quite well: BitTorrent. Yes, BT got a bad reputation because it was widely used for pirating, but that doesn't make it technically inferior. There are arguments for IPFS over BT, but are they compelling?

In general I haven't seen a lot of convincing general use cases for IPFS.

Michael Wojcik Silver badge

Re: Blockchain

Merkle DAGs can scale just fine. It's proof-of-work (and to some extent using naive degenerate Merkle graphs) that doesn't scale. But zfs, say, or git, don't have the sort of scaling problems that are typical of cryptocurrencies.

The problem with IPFS is the use of the term "blockchain", not the use of a Merkle graph. In fact IPFS does not use blockchain, because blockchain is a lousy degenerate Merkle graph rather than a general DAG.

India quickly unwinds requirement for government approval of AIs

Michael Wojcik Silver badge

removing the requirement for government permission, but adding obligations to AI service providers. Among the new requirements for Indian AI operations are labelling deepfakes, preventing bias in models

So the Indian government removed a burdensome requirement on "AI service providers" and replaced it with an impossible one?

I'm ... OK with this. Yeah. More of this, please.

Michael Wojcik Silver badge

Oh, wait, we have a Secretary of Commerce too?

US secretary of commerce Gina Raimondo last week visited the Phillipines and Thailand

... and Speaker of the House Mike Johnson announced the House will begin impeachment proceedings against Raimondo, on the grounds that he was reminded the position of Secretary of Commerce exists. "Look, we're trying to impeach everyone, but sometimes a name slips through the cracks," he told the handful of reporters still listening to him.

US Congress goes bang, bang, on TikTok sale-or-ban plan

Michael Wojcik Silver badge

No, it's not odd. Of course where the stakes are low it's much easier to get agreement. The cost of supporting the Death to TikTok Act is quite low; even people who care about either TikTok or Trump's flip to pro-TikTok the other day will have mostly forgotten about who voted for the thing, or decide it's not enough of a provocation to change their vote,1 or simply not be likely voters in the first place.

And there's a decent chance some court will stay this pending appeal anyway, which means it won't go into effect until some future administration quietly removes TikTok from the list of Evil Foreign Organizations. So no ban will actually happen, and the TikTok fans will forget this was ever a thing.

1Particularly since most wouldn't change their vote if their candidate broke into their house, ate all their food, and shot their dog. We're polarized to the point that few voters will change their minds for any reason.

Michael Wojcik Silver badge

Re: Protecting Americans from Foreign Adversary Controlled Applications Act

Circumscribed corporate personhood under the law goes back to the Middle Ages. A bit late to be complaining about it now.

Michael Wojcik Silver badge

Re: Clone

I'm sure most are. That doesn't mean there aren't some gems which are informative, interesting, and/or entertaining.

I don't watch video (of any sort) every day — as I've noted before, I find synchronous media in general requires significant effort, whereas I can read comfortably for hours — but there are some YouTube creators I watch occasionally. Lock Picking Lawyer, Legal Eagle, CPG Grey are all a decent use of my time, I think, when I'm up for a few minutes to perhaps half an hour of watching.

Michael Wojcik Silver badge

Re: Clone

Yeah, if Jellied Eel and Mike Masnick agree something is a bad idea, I'm updating heavily in the direction of "bad idea".

JFTR, I have never looked at TikTok, and from what I've read about it, and the amount of time I've seen people spend looking at it, I suspect it's dreadful. And, sure, the Chinese government are all over the thing. But this is a bad law trying to do a bad thing. I don't have much hope that SCOTUS will shoot it down, but they ought to.

Forget TikTok – Chinese spies want to steal IP by backdooring digital locks

Michael Wojcik Silver badge

Re: Why bother...?

To be fair, LPL videos are pretty good for quelling worries of whatever sort. So soothing.

FTC goes undercover to probe suspected antivirus scam, scores $26M settlement

Michael Wojcik Silver badge

Re: Another evil doer escapes with paltry fine and "admits no wrongdoing".

And no one would be in jail for this, regardless of what the FTC did. You pay taxes to support feasible enforcement, not magic.

Microsoft forges One Teams App To Rule Them All

Michael Wojcik Silver badge

Well, I don't, since the only reason I'd ever use Teams is because I absolutely have to, and fortunately I only absolutely have to work, where I have a single account.

For me, there are any number of bigger problems with Teams. The lack of a setting to stop it from converting text to fucking emoji, for example. Its horrible inability to format anything well is another. The crap search facility.

Of course, adding anything "AI" to it would make it enormously worse, so OP's question really doesn't make sense.

Michael Wojcik Silver badge

Use case: It's a lot more convenient to kill the fucking thing when it's running in its own process.

Also I don't need to pollute my browser with Teams, thanks.

The end of classic Outlook for Windows is coming. Are you ready?

Michael Wojcik Silver badge

Re: "Microsoft has warned that those days will be coming to an end"

The problem with that is Microsoft have enough momentum to keep them going for decades yet, regardless of what mistakes they make. Sure, my grandkids may celebrate Microsoft's self-destruction (though honestly they're unlikely to recognize it when it happens, unless they decided to become IT-industry historians), but in the meantime those of us unfortunate enough to have to use Microsoft applications for work reasons will suffer with each new idiotic decision out of Redmond.

Michael Wojcik Silver badge

Re: Rebranded Mail & Calendar

I'll say it. If I have to have stupid not-actually-a-browser-but-just-as-bad runtimes, I'd like to at least be able to choose among them.

Dedicated PWA runtimes are a horrible idea to begin with, but locking an application into one of them just adds to the wrongness.

Michael Wojcik Silver badge

Re: Rebranded Mail & Calendar

It's more a case of it being the web version, spun out as a sort-of standalone* program

Ugh. OWA is even more loathsome than Outlook. And dedicated web runtime containers like WebView2 and Electron are horrible too — bloated and with a huge attack surface for no advantage.

Michael Wojcik Silver badge

Re: what Outlook?

I've considered switching to some other client, but here people use Outlook for all sorts of things, and any sort of brokenness could prove hugely disruptive to getting actual work done. I hate Outlook, but I wouldn't much like working around mysterious process failures from not using Outlook either.

Michael Wojcik Silver badge

Re: I need classic outlook

My advice is to get your email out of that proprietary format as soon as possible, and into something easier to work with, like plain text.

It's a beautiful dream, but Outlook's export mechanisms are rubbish (let's see — I have a choice of CSV or ... another PST?). And while plain text is fine for, well, plain text (assuming you can get it exported in Unicode, because Not Everything Is ASCII), it's not a ton of help when half a zillion of your interlocutors insist on using RTF. (Let's assume we can lump HTML, or The Thing That Microsoft Office Calls HTML Even Though It Really Isn't, in with "plain text".)

And then there's the problem of attachments.

And after that, there's the problem that Outlook actually stores a bunch of not-actually-email things, such as calendar items, task items, journal entries, contact-list information, and so on. Maybe the user needs to keep some of that stuff too. I have a ton of historical information in Outlook PST files which I occasionally have reason to refer to. It's important perhaps a few times a month — which is often enough to be quite important indeed. And, yes, you can represent those in plain text, but you'll want to know how they're going to be represented so you can find them later and do useful processing on them.

Outlook's capabilities for organizing and retrieving information are moderately to severely terrible, but it does have them. Any export process is going to need to be able to duplicate those. Sure, if I have everything exported to, say, mbox format,1 I can happily grep and awk and whatnot all the day long. But that does rely on my knowing how those messages have been exported. And it probably involves doing some scripting to split mbox files up and that sort of thing.2

So exporting all that historical Outlook data — mine goes back to 1996 (including, I am amused to note, "to do" items from at least as far back as 2002; must get on those soonish) — looks like a fairly major job.

1And mbox is horrible too. It really only exists because BSD Mail was invented before BSD had a fragging filesystem, and disk space was a really scarce resource then, so small files were excessively expensive, and emails tended to be small because accurséd MIME hadn't been invented yet. Really each message should be its own file, and the collection should be organized using a hierarchy of directories; that's the whole damn point of a hierarchical filesystem. Putting multiple messages in a single file these days is just stupid.

2Oh, I'm sure there are any number of open-source Perl and Python packages to split mbox files. Unfortunately, those would require either Perl or Python. I'd rather write my own stuff from scratch than use someone else's Perl or Python code.3

3This.

Michael Wojcik Silver badge

Re: I need classic outlook

Hey, you leave Arthur Ransome out of this. If Nancy Blackett were here she'd give SatNad a good kicking over it. Or steal his parrot feathers, anyway.

Michael Wojcik Silver badge

Re: I need classic outlook

Glad to be in that 10% who is happy to not have email on his phone. .... I only want to read email on my one device.

A different reason for the same conclusion: I don't want any work documents on my personal devices. Ever. Nor am I willing to give corporate IT any control over my personal devices, which is what I'd have to do if I wanted to have work materials on them.

Sunak's defunct SaaS scheme spent seven percent of budget designed to help 100,000 SMEs

Michael Wojcik Silver badge

Re: Unicorn Kingdom, huh?

Yes, I might be able to get my older granddaughter interested in The Wee Free Men this summer; she's of an appropriate age. Never hurts to encourage the development of First Sight and Second Thoughts.

From quantum AI to photonics, what OpenAI’s latest hire tells us about its future

Michael Wojcik Silver badge

Ugh, what a lot of rubbish

I don't know what specifically OpenAI think Bartlett is going to do for them, much less QC (if that's actually what they hired him for). I'd be interested to hear Aaronson's take on this, but he hasn't posted anything to his blog about it yet.

Frankly photonics seems more likely than QC as a role for Bartlett at OpenAI. Still pretty much speculative primary research, but at least it might be applicable to what OpenAI do.

The stuff mentioned in the article regarding QC is rubbish, frankly. QAOA and other "quantum optimization" approaches have yet to be shown to achieve anything useful; there's no proof QAOA is better than classical approaches, despite years and many papers published. "Optimization problems" have attracted a lot of attention in the QC community because there's money behind them. The results, despite much noise, have not been "promising". Or if they've promised, they haven't delivered.

the unique attributes inherent to quantum computers allow them to explore these factors simultaneously: No. They. Do. Not. Please stop repeating this misconception.

Nothing from DWave is worth the space taken to quote them; they don't do general QC (just adiabatic QC, which is utterly irrelevant here). That comparison to neurons is embarrassing.

Michael Wojcik Silver badge

Re: The universal principle of no free lunch.

Quantum computers already work, so that's your position disproved right there.

Has anyone actually demonstrated quantum supremacy? Debatable. Aaronson's blog has some good posts on the question, and none of the results thus far are in areas where we can have really high confidence that classical methods couldn't be improved.

Will we resolve QC scaling and error-correction sufficiently for some practical problems, such as "crack[ing] encryption" (by which I'll assume we mean factoring reasonable-sized RSA keys and computing FF and ECC discrete logs of reasonable size)? Unclear.

Will QC ever be economically feasible for anything besides very specific applications, where we put a few very expensive systems to work on a handful of hard problems, most likely quantum-physics simulations? I suspect not, at least not in my lifetime. Physics simulations still look like the best application.

Will QAOA actually prove useful? Will we find a lot of other useful algorithms in BQP? I'm not holding my breath.

But quantum computers most definitely do work.

Michael Wojcik Silver badge

Re: "it'll take [..] about a million physical qubits just to compete with modern GPUs"

We have been repeatedly told that the quantum computer resolves all possible values in one go.

Not by anyone who understands how QC works. This is utterly, completely wrong.

When the best anyone can do at the moment in a lab are 1000 qubit computers

Even that claim is pretty dubious. The New Scientist piece is just marketing fluff; they don't cite any independent verification.

Your PC can probably run inferencing just fine – so it's already an AI PC

Michael Wojcik Silver badge

An AI PC

... is one some idiot paid extra money for, as far as I can tell, in order to indulge in intellectual laziness.

And on a related note, Intel's alleged position that application developers will soon infuse all software with AI is patent rubbish. First, of course, a great deal of software isn't applications. And a great deal of it has no use for any of the technologies (mostly LLMs and diffusion models) currently being lumped under "AI".

And for most of the rest of it, it's dubious that "AI" brings anything of value. There's some evidence that GAI is useful in, for example, exploring the space of possible chemical compounds for various applications. Other than that, I haven't seen a methodologically-sound study that convinces me of actual utility. And that definitely includes the "ask questions about a contract" example from the article.

Intel under Gelsinger is desperately searching for a reason why anyone should consider it an innovator rather than just a manufacturer of commodity processors. No doubt they'll ride the AI hype until it crashes, but for mass-market PCs, this emperor has no clothes.

Michael Wojcik Silver badge

Re: the pachyderm in the parlour

That kind of spec is not unreasonable for a desktop

It's unreadable if you don't need it to do real work. I certainly don't.

Of course, I haven't had a desktop in decades either. (And despite having been using PCs more more than half a century, I've only ever bought two for myself, and they've both been laptops. The best price to pay for a computer is nothing, IMO.)

Michael Wojcik Silver badge

There's a whole pile of literature on the effects of offering rewards in prompts. Promising tips (in the monetary sense) is one technique that worked well with the popular models, for example — though now that SoTA public models are starting to keep more session context, people have reported models rejecting tip offers on the grounds that they didn't get the promised tip the last time.

(This is amusing but not surprising. The training corpora no doubt contain many references to being paid or not being paid tips, so there are plenty of gradients that a tipping prompt could direct the model toward.)

Michael Wojcik Silver badge

Re: No wonder hardware vendors are on board

... which is less than 21 and more than 23, as implied by the post you're responding to.

Michael Wojcik Silver badge

Re: So I can run a local chatbot

when privacy or legal concerns means you don't want to or cannot let the information leave your system

I can accomplish that without using an LLM.

Oracle AI buzz means Larry Ellison's worth $15B more today

Michael Wojcik Silver badge

I suspect quite a few investors see this as a bubble and intend to try to time it, and get out while they can make a profit. I've seen some people explicitly recommending that. And, indeed, if you can tolerate the risk and don't see that as an unethical or undesirable investment strategy, it makes some sense.

(I personally have no appetite for investing in the stock market or other speculative investment. My 401(k) has a broad portfolio including stocks, but I ignore that other than checking my statements to see that my elections are performing more or less as I hoped. With a pretty high probability that is costing me money — I have disposable income and there's a decent probability that I could make at least a modest return if I actively invested — but I don't want to spend the energy and attention on it.)

Attacks on UK fiber networks mount: Operators beg govt to step in

Michael Wojcik Silver badge

Nah, they all were. That's why we don't have any medieval 5G towers any more.

Trump 'tried to sell Truth Social to Musk' as SPAC deal stalled

Michael Wojcik Silver badge

Re: Why would Musk want to buy that

Trump has never yet displayed gratitude. Only a fool would look for it in the future.

Michael Wojcik Silver badge

SPACs are nearly always cons, true, but it's hard to see offhand how regulators could forbid them. What's the legal criterion you'd use to distinguish them from the normal case of a listed firm acquiring an unlisted one?

I suppose the solution is really just for the SEC to increase reporting requirements for acquiring an unlisted company to be more or less on par with those for an IPO.

Michael Wojcik Silver badge

Re: Sorry????

Right. Who's successfully ridden Trump's coattails? At the moment, Melania and Ivanka are the only ones I can think of. The boys are clinging on but he'd cut them loose in an instant. His various lawyers over the years have mostly suffered for the privilege, except for the ones smart enough to bail out while they were still in a decent position, like Tacopina. (I don't think anyone was surprised that Tacopina proved to be the brightest of that bunch.) He's happy to avoid paying those he owes money — something Trump and Musk have in common — and to throw his co-conspirators and former allies under the bus.

Sure, Trump got a bunch of morons elected to Congress and other political office, but only until they cross him in any way, so it's lapdog or you're out on your ass.

Trump's motto isn't even "what have you done for me lately?", it's "what will you do for me now?".

Even Musk seems to be sensible enough to realize that.

Airbnb warns hosts who use indoor security cameras they may face eviction

Michael Wojcik Silver badge

An empty set?

good news to those who were worried about privacy in their Airbnb short-stay

Why is anyone who is "worried about privacy" using Airbnb at all?

Kremlin accuses America of plotting cyberattack on Russian voting systems

Michael Wojcik Silver badge

Re: Shocked... shocked I tell you... well not that shocked

Oh, I think it's legit. Donald Trump actually won the 2018 Russian election, but the NSA didn't want him to run both the USA and Russia at the same time, so they stole it for Putin. Sydney Powell told me so in a dream.

Microsoft waited 6 months to patch actively exploited admin-to-kernel vulnerability

Michael Wojcik Silver badge

Re: Thought Experiment

That just happened last year.

We also have last month's Azure breach, for example, which was against Microsoft systems, even if it didn't focus on Microsoft data.

Microsoft is certainly a constant target. Any organization of any significant size is.

Michael Wojcik Silver badge

The problem is triaging an exploit like this. Sure, you can say that in general if the attacker's code is already running in some sense in the TCB then there's nowhere to go in terms of privilege escalation. But if, say, the attacker has a data-based attack against existing TCB code, rather than their own code in the TCB, then arbitrary code execution in the kernel is most definitely an escalation. Or if the attacker is otherwise limited to what they can do within the TCB but can escalate that to arbitrary execution through a chained vulnerability, and so on.

It's the same way with other data-based exploits. Take, oh, HTTP response splitting. At first blush you're just corrupting an HTTP response and producing invalid data, but with a bit of effort you can get the recipient to do quite a few things that are valuable to an attacker.

Rules of thumb for evaluating vulnerabilities are just that. Often they give a good approximation of the risk; occasionally they fail.

Trying out Microsoft's pre-release OS/2 2.0

Michael Wojcik Silver badge

Re: Microsoft Presentation for the 1989 IBM PS/2 forum

Fixed (more or less) in OS/2 Warp 4 Fixpack 17.

(I thought I remembered they had finally done something about it, and some searching turned that up.)

Michael Wojcik Silver badge

Re: Very Different

Yeah. The Palm Pilot wasn't a tablet in any sensible definition of the term either, even if you wrote on it with a stylus.

Michael Wojcik Silver badge

Re: It was very nearly ready

But it would have made no difference to Apple, phones, Mp3 players, tablets etc, and eventually IBM & MS would have fallen out and there would have been NT, but delayed.

It certainly could have made a difference to Apple. (Since we're talking hypotheticals, it's hyperbolic at best to claim it wouldn't have.) IBM's OS/2 marketing was horrible, and Microsoft had the (anti-competitive) OEM deals; delaying the Microsoft-IBM breakup could have led Microsoft to do a much better job of getting OS/2 sold, and that in turn could have driven ISVs to putting out more and better applications, which could have pulled even more revenue away from Apple in the mid-1990s and caused it to fold completely or get bought during that period's acquisition rush.

It's fairly daring speculation to claim that it would have destroyed Apple, but it's at least as tenuous to state that it would have made no ifference.

what if IBM had used DRDOS (CP/M 86), and eventually multidos and GEM?

DR-DOS was not CP/M-86. DR-DOS was descended, in a sense, from CP/M-86, via Concurrent CP/M, but since it aimed at MS-DOS compatibility it was a substantially different OS from any CP/M variant.

GEM, originally developed for 16-bit systems in the early '80s and growing out of the earlier GSX, was limited by that legacy. One of the best things about OS/2 was its replacement of the decidedly quick-and-dirty APIs of QDOS1 (and PC BIOS) with a fairly well-thought-out kernel API for applications. Facilities like threading and IPC were done properly, not just in the kernel implementation but in the abstraction presented to applications. OS/2 2.0's HPFS was a big improvement over FAT. And OS/2 had key features for those big enterprise players, such as SNA (with LAN Manager) and TCP/IP (in Extended Edition and later Connect); the revenue brought in by those add-ons helped fund development. Bringing GEM up to the same level, however nice the kernel might have been (I've never looked into it), would have been a major undertaking.

Also, look at the internal politics in IBM. Even though Future Systems died a horrible death in the mid-70s (though some elements were preserved in the S/38 and then the AS/400), IBM took decades to shake off the conviction that success lay in tying all its systems together. OS/2 fit neatly into SAA, in a way that GEM never would have. OS/2 was enterprisey and, despite its innovations, conservative. GEM was cool and forward-looking. An outfit like SGI or Borland might have tried to sell GEM as a PC OS; the suits at IBM are unlikely to have supported it.

(Weird historical note: What eventually became DR-DOS started development at Digital Research Europe, in Newbury, Berkshire, UK. Later DR-DS was acquired by Novell, and then Novell by Attachmate Group, and then Attachmate Group by Micro Focus — headquartered in Newbury.)

1Hence the name.

Michael Wojcik Silver badge

Re: Very Different

OS/2 really was quite a solid OS for the hardware it ran on. I worked for a number of years on a middleware product that had versions for MS-DOS, Windows (from 3.0 in real mode on up, as successors appeared), OS/2 from 1.1 EE onward, various UNIXes, Linux, OS/400, OS/390 (with variants for batch, CICS, IMS, and TSO), and CICS/VSE. We had quite a few customer installations of our OS/2 product, running distributed applications with other components on various combinations of those other OSes. There were few reports of stability issues.

I built a distributed source-code control system using RCS and our middleware platform with an OS/400 client and OS/2 server, so that we could have source code control on the AS/400. Worked a charm.

Prior to (IBM) OS/2 2.0, it's true that it was a PITA if the single DOS Box hung, because while you could generally switch back to OS/2 applications, whatever was in the DOX Box was stuck until you rebooted. But that minor quibble aside, it worked quite nicely. And having TCP/IP in OS/2 1.2 EE around 1990 was a big improvement over Windows with its assortment of fragile third-party TCP/IP implementations. A lot of our networking was SNA (because that's what a lot of our customers were using), and OS/2 was good1 with SNA as well, but particularly for interop with UNIX having a good TCP/IP stack was great.

1Well, as good as things got, with SNA. Many's the hour I spent pouring over SNA traces through the long winter nights. And short summer nights. And the rest of the time, too.

AI models show racial bias based on written dialect, researchers find

Michael Wojcik Silver badge

Re: Bias that matches society?

the issue, to me, isn't that the LLMs are biased but that society itself is biased

It's not a useful distinction. If LLMs are employed as part of a decision system, their biases are a concern for the operation of that system. The origins of those biases do not mitigate the problem.

Nor is it a helpful observation for explicability. For one thing, it's the observation anyone would make; I don't think anyone believes these biases are sui generis in the models. For another, identifying them as originating in the training corpora doesn't help, because we already knew they were there, and we can't remove them (as I noted in an earlier post).

And, finally, we know this is a hard problem, since it's been formally proven that you can't satisfy all of even a small set of intuitively-plausible fairness constraints in any decision system.

So ultimately the question of "where does the bias originate" isn't interesting.

Michael Wojcik Silver badge

Re: Broadcaster Presentation / Linguistic Standards

In history? Well, obviously yes; that's how we can have languages that are not mutually intelligible in the same linguistic group. Indo-European didn't start as a bunch of people each speaking a different language.

More generally, this entire line of thought — "linguistic degradation", dialect formation, propagation of dialects and so forth — is just far too simplified to be useful, I would say. The evolution of language is much more complex, particularly as communication technologies become increasingly sophisticated and available to speakers of a language.

It's possible that the promulgation of RP, say, in radio and television broadcasting in the UK for a large part of the twentieth century had some effect on dialect formation and adoption, and increased consistency and mutual intelligibility to some extent within the reception area. But there are so many confounding factors, such as attempts at language regularization in schools, that I suspect someone would have a hard time of demonstrating it to general satisfaction.

Natural languages have always evolved, diverged, intermixed, and so on, with elements such as pronunciation, spelling, and idiom coming into and falling out of fashion. This is particularly true for English, which is very widely spoken, internally highly inconsistent, and lacking any generally-recognized central prescriptive authority.

Michael Wojcik Silver badge

Re: Not at all surprising really

If a human were able to review the entire training data set they would probably find that such biases are deeply embedded in it.

That's unnecessary. Even a cursory review of a representative sample will tell you that. If you want details, well, we have entire academic fields that have been producing the details for decades or centuries (depending on the field).

There's no such thing as a significant piece of human discourse which doesn't incorporate some bias. It's impossible to present an argument without incorporating some bias, and as Toulmin demonstrated, everything's an argument.

Of course that doesn't mean results like those discussed in the article aren't important — I'd argue they're very important indeed.1 But the fix isn't "remove bias from the training corpora" because that isn't A Thing You Can Do.

1The only argument to the contrary that I've seen which I'd even entertain is the "existential risk is more pressing" one made by the doomers. I'm not convinced by the doomers at this point; my P(doom) is pretty low, and my P(adverting-doom|ASI) is also low, so I'm not putting my energy into worrying about that scenario.

Michael Wojcik Silver badge

Re: @LionelB - Reasoning

The SoTA LLMs are being tuned after the initial training process with RLHF, which is a type of reinforcement learning.

Bias, while a concern, is by no means the whole story, of course. For example, it's been formally proven that a small set of features which most human judges consider intuitively necessary components of "fairness" cannot be jointly satisfied, even approximately, by any decider (human or machine). See Raghavan, "What Should We Do When Our Ideas of Fairness Conflict?", CACM 67.1 (2024), for an introduction to the problem.

So we have, on the one hand, demonstrable issues with bias and algorithmic unfairness in current mechanical systems for making or influencing decisions; and on the other, the demonstrable impossibility of creating any system which — even given perfect information about the past — will be fair under all the obvious criteria ("obvious" meaning "typically described by a large number of people"). And then we have all the various issues of explicability, inner and outer alignment, auditing, trust, and so on.

Now we know we have all those problems with human deciders too. The question becomes, at what point do we decide it's ethical to — in whatever situation we pick — substitute a flawed mechanical decider for a flawed human decider? And if we do, qui bono?

Michael Wojcik Silver badge

Re: Reasoning

Yes. That's why we do see a statistically-significant correlation between IQ scores (and the results of similar tests) and success in some other classes of intellectual tasks, as well as some things that some people argue are proxies for intelligence;1 but we don't see a meaningful correlation between IQ and many other intellectual and cognition-dominated tasks, such as, say, scientific productivity, or response to instruction.

Indeed for creative tasks you want a certain degree of breakdown in pattern-matching, in effect to anneal the system and get you out of local minima. LLMs and diffusers implement a form of this with the "temperature" parameter which injects some noise into the system, but that's a fairly crude mechanism.

Personally, I think we won't get anything close to human-like machine intelligence without 1) significantly larger models, and 2) systems that aggregate heterogeneous models which compete internally for attention. The latter could be simulated (deliberately or accidentally) by a much larger unified model, but I suspect that's outside our scaling capabilities currently.

It is worth noting that current SoTA LLMs have been proven to incorporate world models and some other features that are probably necessary for cognition; descriptions such as "just predicting the next token" are no longer useful in understanding the functions implemented by these models. And that stands to reason, since a large enough language model is an abstracted world model, because our (natural) languages incorporate information about entities and structures in the world. But analytic techniques such as SAEs and linear probes still — as far as I've seen — suggest they're not doing anything that I would say merits the label "thinking". (Though, like Searle, I'm not ready to pin that down yet either.)

1Though most of the ones I've seen are pretty vague and subjective, such as "professional achievement".

The S in IoT stands for security. You'll never secure all the Things

Michael Wojcik Silver badge

Re: Not sure if that was autocorrect or just me...

That's the thing. It's irrelevant (aside from the small satisfaction of being proved right) that many of us were warning about the dangers of IoT from before that term was even coined. Manufacturers were also aware. They didn't care. They knew perfectly well that a lack of security wouldn't significantly affect profits, whereas with their often razor-thin margins, any efforts to improve security would.

The market won't fix this. Security vulnerabilities are an externality for consumer-device manufacturers, medical-device manufacturers, and the others responsible for this mess. If we want the situation to improve, we'll need regulation.

Trump, who tried kicking TikTok out of the US, says boo to latest ban effort

Michael Wojcik Silver badge

Re: US politics is basically

the best candidates

You misspelled "only viable". Like it or not (and most of the electorate doesn't), when the Presidency is decided by a simple majority of the Electoral College and nearly all the states deliver all their electors in a basket, this is the situation you're going to end up with.

Incumbent first-term presidents nearly always run again; Polk is the only one I can think of who didn't (aside from those who died before they could). And successful demagogues are hardly a new phenomenon.

Michael Wojcik Silver badge

Re: Tik Tok Youth Brigade

"The reverse vampires, working in conjunction with the RAND corporation..."

Michael Wojcik Silver badge

Re: Tik Tok Youth Brigade

A post that contained only facts

It opened with speculation, dude. Regardless of my opinion on that post's content (JFTR, I didn't vote on it either way), it most definitely was not "only facts".