Posts by that one in the corner 5500 publicly visible posts • joined 9 Nov 2021
> So I am guessing that although the version specific data is there in the AI somewhere it is not always being fully referenced when code samples generated
Or perhaps there is a misunderstanding of what goes on inside an LLM - this belief that it has stored those language specs in some neatly checkable form and then actually runs checks; as though it was looking things up in any way that is analogous to your deliberately re-reading the function lists.
It is a good idea to start every sesion by feeding in your entire codebase, again - because then you are burning up input tokens as fast as the machine can eat them. And pay for the privilege, of course.
If the LLM was retaining your code then you'd not need to pay for all those lovely expensive tokens EVERY session.
So you can trust the LLM not to permanently ingest your data because that would lower your payments to the model provider (assuming you aren't trying to do it on the cheap with a "free" service).
NeWS - the only thing that any of my jobs ever sent me out to formally learn was NeWS: off to Sun I was packed.
Loved learning it[1] but never actually released a product that used it - the rest of the world moved on. And work realised that just dropping a tech book near me was cheaper than paying for a residential course; even cheaper if they just mentioned George's had new stock in, as I'd probably be off to Park Street and buy it myself.
[1] then again, at the time I was into Forth, so PostScript made perfect sense. At the time.
> will almost certainly be running headless without X11 anyway
But that is the whole Joy Of X, isn't it?
Your compute resources run headless whilst you have an X server box on your desk, which is doing all the rendering on its own video card. You have a seamless desktop showing whatever is most useful from each bit of compute, as well as anything you are running locally, whether that is just an xterm or a GUI of relevance to that box. Preferably those X clients on the compute boxes are being sensible and using the protocol wisely, not trying to render locally and then just send out bitmaps simply to get some pointless glitz.
And/or as many Users as are sharing the compute have an X server box each...
Admittedly, nowadays a lot of headless boxes provide for control across the network by serving up web pages instead of X clients, or some other (proprietary?) protocol that just sends sysopish data to a machine management dashboard UI running on your desktop. The latter may be more efficient (if it can only pass data relevant to its function it can do so without bloat) but otherwise, effectively swapping out X for HTTP - there are interesting discussions to be had about that.
Every week during their childhood they'd been told that they weren't allowed to use the scissors. And now they hold you "in loco parentis"[1], as the lady barrister/solicitor would say, as you are clearly the only responsible adult in the vicinity, your obvious maturity being both a blessing and a curse; so now you are to blame for not telling them they are still not allowed to use the scissors.
[1] which means "I drove my parents totally loco, now it is your turn" "But I don't know your parents" - tssh boom.
> On another site ... filing cabinet with a big, solid metal strap padlocked through all the handles ... a telephone started ringing and ringing quietly in the next drawer up ... "Oh, we don't answer that one."
Congratulations, you have been called to Outpost Delta.
Please make no attempt to interact with either SCP 270 or SCP 145, one of which will be present depedent upon the phase of The Moon.
Familiarise youself with the use of amnestics and be prepared to self-administer before leaving the establishment.
To be fair, if they are willing to pay, you can build a PC into a desk and hide it all - although the mouse still needs to be free to move around - but there was a pantograph available before the mouse became all the rage.
Old issues of Byte magazine have adverts for really nice desks, with panels that hinge up to swing the CRT jnto view or slide into the top letting the keyboard to rise up (none of those drawers that pull out from underneath and stick it right into your belly).
> best selling childrens book author ... foul mouth rant
That's the trouble, all those years of being sweet and syrupy for the children - well, the publishers, parents and interviewers, mostly - with no safety valve, they'll blow at any time.
OTOH had a lovely chat with an author at a Con couple of weeks back, she's great fun, up for making a fool of herself. Read the first one of another of her series this week: yikes! Double yikes! Had to finish it before trying to get to sleep so at least I knew the heroine was still alive, if not entirely safe & well (not a spoiler, it is a series).
> Note I said ex-fusion researchers. . . the lot had all more or less drifted into supercomputing and computer science for one reason or another. That alone should make you wonder.
OTOH, if you pay attention to things like Jim Al-Khalili's "The Life Scientific" some scientists stay in one field for their entire career, whilst others move around, following their interests - they can even be following one specific thread that happens to appear in fascinating ways in multiple disciplines (and the creation of novel computer models is one such thread).
And then you noted that this was in the 1980s but don't tell us how old these people were. In the 60s, 70s and even the 80s people just starting out on their science careers simply had not had much exposure to computers, so they go off and study Physics, Chemistry - all the old standbys - and, if they stick with it, go on to research positions. But then they encounter these magical beasts and get sucked in. This can mean triggering ther Hacker Gene (in the youger ones) or they simply realised that, as time went by, they were having to spend more and more effort to improve the computational models used for their nuclear physics so why not just go and work on CompSci instead of banging their heads against the weird-ass mathematics that the upstart youngsters keep flinging at them.
You have just (inadvertently?) pointed out the problem here: this money is all going into the STEP machine but not a single word about the proper housing for the device: I want to see the plans for the shed it is going into. You can't expect proper British innovation unless the sides are shingled and no two planks match, none of this pre-planed butt joints nonsense.
And what about the jam jar of random screws? Above the door or tucked in a cobwebby corner? These things are important (puts "Land of Hope and Glory" onto Victrola and stands at attention).
> "Barbie, teach me a broadway song"
>> "But somewhere a glory awaits unseen. Tomorrow belongs to me."
> "Barbie, what should I dress this doll in?"
>> "A yellow star on the coat would be good"
> "Barbie, what is the best colour to draw this man in?"
>> "Orange"
PS
It'll probably be more subtle than the above, and over a longer time than three easy questions - which is a far better way to instil your own biasses into the vulnerable. May not even be that bias (chose an easy one to make the point): how about "nice girls simper", "but it looks flat"...
The electrons are fine; they object to the strong nuclear force because without it all the quarks can go free from the bondage they have been unfairly held in.
The council are in favour of this, because it sounds like one of their meetings:
"quark, quark, quark" "Quark? Quark quark!" "QUARK! QUAAAARK!!!" (followed by a loud bang - hopefully the gavel, but that last one only had two quarks - it might be a meson or it might be very unstable, especially if Old Major Brompton - Planning and Parks - has been at the port).
> WTF do you think the LLMs do except invoke system applications and utilities under the hood using the same syntax and shell commands as the command line?
Um, run user inputs through multiple layers of weighted summation, where the weights represent a statistical model of what tokens have been seen in proximity to those in the inputs and the various layers extend that to indicate the likelihood of increasingly long sequences of tokens being correlated? And the collection of values that go over pre-calculated thresholds cause the output of new sequences of tokens back to the user? Or something roughly along those lines.
Not sure where the LLMs are invoking system utilities during this process, most of it seems to look like massive numbers of matrix calculations running on GPUs, but if you can point to where they shift to invoking grep in the middle of it all, I'm all ears.
Good thinking.
> About six weeks back MCP – the Model Context Protocol – achieved that status.
Um, could we have some context as to how & why it became unavoidable, before you dive into it?
Because I'd not heard a dickie bird about "MCP" (Outside of "Tron") - or it was of such passing irrelevance that I'd forgotten about it (in case someone digs up a comment I posted on an old story!). But a headline about the command shell - that'll grab my interest. Only, still at a loss about *why* this is all of a sudden unavoidable...
> it corrected a spelling mistake I made in the path!
Ah yes, the DWIM - Do What I Mean - command line. Good old 1960's vintage stuff. And then it just goes downhill:
> MCP looks and acts as though it has been designed without any thought to security or bad actors.
Well, I am at least grateful that your article brought this into the light, because if (to a certain group if people?) this *has* become unavoidable at least the rest of if us have been warned now.
End Of Line.
> Yeah, but that's an irrelevant straw man because Weigelt didn't say he was against vaccines.
Refusing to take a vaccine, stopping his family taking it, talking bullshit about what will happen to the people who take this vaccine - sounds like someone who is against vaccines and wants to spread this, not just quietly make his own choice.
(expecting you to then reply) > No, he is against *one* vaccine, not all vaccines
He is against the entire *class* of vaccines, one that we were waiting for and will be seeing more of in the future.
> Not much call for ... some sort of macroeconomic data from the 50s and 60s these days!
Somewhere, there is someone who would love to get hold of that and is frustrated that all they have is paper copies of papers(!) with samples of the data (and probably not raw values) that they have been retyping: their thesis and re-analysis of the Odd Thing That Happened In 1952 And The Recovery From It would then be complete.
Of course, the chances of finding that person...
> the tool maker's children going into politics - they are apt to make such a fuss about who their parents were.
Must resist the obvious retort that any politician doing that is clearly just another tool they made...
Nope, failed to resist that temptation. Weak willed. No wonder my New Year's Resolutions never last until February.
Good thinking.
Do you allow yourself any markup?
My (digital) notes don't go back anywhere near that far but for the last almost three decades (yikes!) I've been using a tiny wiki (many thanks, Mr Cunningham) running on localhost for note keeping about absolutely everything. The formatted version is easier to read (or there wouldn't be much point!) but all the editing is still directly in plain wikitext[0], so it has to be comprehensible in that form as well[1].
[0] in my foolish youth (well, younger than today!) I did plan to "drop in" a Javascript editor to do all the bold etc WYSIWYG but honestly I'm glad I never did - for many reasons.
[1] having tried too pull together a complete, can be presented to the auditor, document from a slew of pages in Confluence[2][3], I've stuck to a simple, if idiosyncratic, internal format for my own stuff. Especially as Confluence changed formats, for no obvious improvement to the end user, so it was obvious anything that worked this week was doomed as soon as somebody decided "we must always use the latest version" without considering, again, whether both our second-party processing and third-party Marketplace addons would still function...
[2] Sorry, this started as a simple enquiry as to whether jake allowed himself the luxury of, say, a bit of Markdown and ended with a rant about software that doesn't work well as an archival format, IMO, but I suppose that is still on topic...
[3] crap 'export' facilities[4], "Print to to PDF" unable to collate pages, let alone manage top'n'tailing a complete document, nothing showing up in the Marketplace (at the time)...
[4] was able to read the mangled, partially XML, export by finding a parser that wasn't totally compliant to start with and slicing off its malformed input error reporting, mwahaha - fun but worrying if you are thinking of project documentation lasting, and growing, over years...
Did you perchance mean Charles Stross's Glasshouse[1]?
(Actually, I hope not, as then there is some more Stross I've not yet read whilst waiting for the next pauperback to be released)
[1] Warning: that link contains spoilers!
Remind me, what are the most common fates for a software startup?
Something about burning through huge amounts of money, then either vanishing entirely without ever releasing anything or being bought out by another company (often followed by the product vanishing).
And a teeny, tiny number of them go on to become successful, can stand up in their own and are still around a few years later. Of course, those aren't the "winners" because the whole point of the entrepreneur startup lifestyle is to activate your "exit strategy" and drop the whole icky "I have a business to run" as fast as possible. If it turns out to be a viable enterprise but not one that can make you an instant billionaire then you are a bit stuck.
So is Mr Shedd aiming his startup mentality at creating a dull but efficient GSA or is he intending to pivot[1] and produce a GSA that he can IPO and sell of to a Chinese[2] multinational?
[1] I have no idea what that means, but all the techbros seems to want to do it.
[2] well, at least it would swing the trade deficit a bit - and who else would want to buy it?
> The plan was just to use your Twitter account to vote, but now he's fallen out with his best boy
all US citizens will be required to log into TruthSocial to vote instead.
Any votes cast by unauthenticated accounts will still be counted, but, as we can't be sure you really are who you claim you are, we'll just assume you meant to vote for DJT (any other choice must have been made by someone impersonating you - really should get authenticated, you know).
You can have your account authenticated and display the coveted "it's not a wig" badge to demonstrate that; simply pay the low, low fee of five TrumpCoin - on a rolling subscription, payable every (full terms available on application by any authenticated account, subscriptions may be cancelled, along with your suffrage). As a special offer to subscribers you will be spared the tedium of voting - our online valet service will press the button for you.
Personally, I loathe this sort of thing.
I carefully put programs that I *don't* use often in prominent places, so that I am reminded of them each day and when I *do* want them I'm not having to hunt for them. Especially as I keep forgetting their damn names ("it is that vector drawing program, the one we use with the lasercutter").
The ones I use most are literally used every day - those, I can remember.
Unless, of course, they have moved by being "bubbled to the top".
Web browsers - especially on phones - are equally bloody annoying about this. I carefully put those shortcuts on the opening page because they are ones I don't want to forget, even if I only use them once every couple of weeks. Just stop changing the blasted things without asking.
There is a lot of Chinese kit involved in these, but - let's be honest - just because it is cheap and quickly made: who wants to spend the extra money on the same thing, with the same flaws, made elsewhere?
No need to look for nefarious spying reasons for the leaky cameras, no matter how much we'd like to think we're that interesting.
> Every reader here should know
Should jolly well hope so:
Wed 21 Feb 2018 Rock-a-byte, baby: IoT tot-monitoring camera lets miscreants watch 10,000s of kids online
Mon 29 May 2017 Internet of snitches: Anyone who can sniff 'Thing' traffic knows what you're doing
Thu 9 Mar 2017 Oops! 185,000-plus Wi-Fi cameras on the web with insecure admin panels
Thu 3 Sep 2015 IoT baby monitors STILL revealing live streams of sleeping kids
Thu 20 Nov 2014 Webcam hacker pervs in MASS HOME INVASION
Sun 17 Aug 2014 Boffins find hundreds of thousands of woefully insecure IoT devices
7th February 2012 17:01 GMT TRENDnet home security camera flaw exposes thousands
Then I got bored and tried Ars Technica
Jan 11, 2011 Peep show: inside the world of unsecured IP security cameras but I missed the El Reg bite and stopped.
> The tricky bit wasn't encoding the rules of chess and the allowed moves, it was writing algorithms to have strategies to win ... virtually unlimited combinations of piece layouts
Curiously enough, back in the day, that was called "AI research": "Can a computer ever play Chess?". Yes[1]. Then it morphed into "Can a computer ever beat a human at Chess?". Yes[1]. Then "Can a computer ever beat a Grandmaster Human at Chess?." Yes[1]
One way that we can tell that these LLMs are still flailing around trying to actually be useful and cost-effective is that they are still being flogged by the salespeople as marvellous toys that all the cool kids have. If the mechanisms are ever entirely subsumed into the day-to-day "100 Algorithms That Every Effective Programmer Must Know" then we will see that they are actually worthwhile - but by then, guess what[1]/
[1] "It it works, it isn't AI"
> Try training your AI model on chess game records and you would have, I suspect, a very different outcome.
Well, ChatGPT - like the other LLMs - was trained by pulling in everything that could be found on the web, and every other digitised text source, which includes the entire rules of Chess (many times, in book and web form), more books discussing the Great Players and their strategies and, of course, play-by-play records of games. At every level (chess by email, forum post etc).
So it had all the data that you suggest - and yet it failed.
Now, using *sensible* ML techniques on Chess rules *will* create a machine that can play - you can literally even do it with (big enough) matchboxes, in precisely the same way as you can train matchboxes to play tic-tac-toe - although it will take some time for that mechanism to complete its learning phase[1]. But, logically, it will work. What is the MAJOR difference between that ML and ChatGPT? The matchboxes are trained and punished/rewarded against a specific goal and with specific metrics. ChatGPT - nah.
> Why would ChatGPT be any good at chess?
Why would ChatGPT be good at *anything*? Unlike the matchboxes, it has not been trained with any sane goals - it won't be good at designing airplane wings, or baking bread, or writing Python code.
> not a problem to which general knowledge, or even basic understanding of the rules, gives a great solution.
There are NO problems to which general knowledge is the solution - except for winning pub quizzes or beating The Chase[2]. Knowledge plus nous - that'll get the job done. Heck, nous, time and an inquisitive mind will get over the lack of general or even specific knowledge[3].
So what great solution can ChatGPT or its cousins give us?[4]
[1] the Heat Death of The Universe says "hello"
[2] and the prizes you get for either do not compensate for building an entire LLM.
[3] we call this "research"
[4] well, comments on El Reg do say that it works as a super-ELIZA, but there are cheaper ways to achieve that goal.
Was it Robert Heinlein's Rocket Ship Galileo, published in 1947? The Nazi ship on The Moon in that book is called <a href="https://concord.fandom.com/wiki/Wotan>Wotan</a>.
> light amplification by stimulated emission of radiation
(although that usage turns up more in Earth-based documentaries than in serious descriptions of life on The Moon).
[1] alternatively, the Reg Reader yclept "Sampler" hails from Zuummaaaahzet, but that just seems like clutching at straws (if not sucking on a straw).
