From Newsgroup: uk.games.video.misc


I wonder what it would be like to be completely ignorant of world news
and current events. Could you make a modest living and successfully
steer clear of all the international and local drama and just not care
about supply chain issues or social mobility or economic inequality?

Play:
--=--

Roguecraft DX (EVC) - This isn't as fun on a TV. I ran an HDMI lead from
the EXP to my new screen and it just doesn't feel right when it's not handheld. That's a shame because it's mostly illegible on the tiny EXP
screen, even though I'm sure it's pushing all the pixels that the Amiga
could. Probably when I get some time I'm going to go back to Rogue64 and
see if the experience is just as diverting as it was before.

Yakuza Kiwami 2 (PS5) - Speaking of diverting. I have all these new
games clogging my gaming space and all I want to do is wander around
Sotenbori trying out sushi. I did accidentally finish the golf center challenges and unlock the side story that gives you the extra weapon,
and I unlocked the item that gives you the special golf club attacks. Curiously, I'm not super motivated to power through the club management
and Majima Construction side activities like I was before, even though I
know that there'll be some nice combat bonuses on the other side of
them. I'm blaming this on some mild ennui.

Want:
--=--

Nothing (RL) - I've tried about a half dozen times to shop for new games
(or more accurately, for games that were new a year ago and that may or
may not be on some kind of discount now) and I just couldn't pull the
trigger. It's only a small share of economic uncertainty driving that indecision because I also can't seem the muster the enthusiasm.

Bin:
-==-

AI - Sega has all but cited generative artificial intelligence as a
selling point for their new Crazy Taxi World Tour title. Kotaku has said
that fact is reason enough to skip the game sight unseen, and I struggle
to disagree. I'm going to be bitter old goat if it turns out that slop development methods drive me out of the hobby before lack of traditional physical game media does. It occurs to me that the enthusiasm for AI
from the moneyed classes is in fact based in perpetuating economic
inequality.

-KKC, who hopes to be cheerier later today.
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: uk.games.video.misc

Kendrick Kerwin Chua <kendrick@nospam.io-nyc> wrote:

I wonder what it would be like to be completely ignorant of world news
and current events. Could you make a modest living and successfully
steer clear of all the international and local drama and just not care
about supply chain issues or social mobility or economic inequality?

I think being in prison could get close, if maybe more so in the past. Presumably even there/then you'd get indirect drama exposure because
society and such. But if video games have taught me anything - and
they haven't - it's that at least when you get out, you get to wreak
your bloody revenge across a handy sandbox area, and/or selfishly drag
an orphan around town who struggles to keep up with you, and/or engage
in epic tales of Yakuza-related intrigue, and/or be amazed by mobile
phones or remote controls or DVD players existing. Because obviously.

Play:

Minecraft (PC) - overcoming my Fortune III uncertainty from last week,
I settled for the non-fortunate approach and mined 20 diamonds. Which
I brought back to base, where it didn't take me long to remember that
I'd previously split my diamond stash out between two different types
of chest, explaining why I'd seemed so low on them. Bah.

That gave me the bright idea to check my little base above the End
portal (to check if I'd left any further diamonds there as well), that
base not being too hard to get to given my shortcut to it through the
Nether. Which I was pretty sure I'd previously made quite safe, with
the whole route fully enclosed.

The whole route was not fully enclosed. As it turned out, my Java save
is the one with that precarious staircase up the side of a tower I've
mentioned previously. But I braved certain^Wpossible death and made it
to the little base with no drama... and no diamonds. (And definitely
no foreshadowing.) Which made sense to be fair, as some were in my
Resident Evil^W^WEnder chests which each share the same items. So
having an Ender chest at the little base meant I hadn't strictly
needed a separate diamond stash.

On the way back through the Nether, I spotted an awkwardly-placed
Magma Cube near the stairs up the side of the tower. I figured it was
likely to jump off the edge before I got there, and even if it didn't,
I was pretty sure I could run past it. Probably. Maybe. So, I never
saw the thing approach. But it apparently bumped me hard enough to
send me flying off the stairs, falling a good 30+ blocks down to my
death. Yay. Plus side, though - not into lava for once. And clearly
that meant I was going to go charging right back in there after
respawning, and successfully grab all my stuff and safely make it back
to my main base. Unquestionably. A highly likely outcome.

I had a few bits and pieces ready for this sort of situation, but not
as much as I'd have liked. In the end I went there with goodish but
not ideal armour, a bunch of whatever blocks I could find to place on
my way down and up again (mostly dirt and some wood), no precise idea
of exactly where I'd splatted nor how exactly to get down, and a
general aim of somehow not dying despite my lack of any real plan.
Just like in real life.

I got there with no obvious baddies around. That tower with the
staircase has what looks like a bridge leading to it (really more of a
roof I think), and looking down off the side of that I could see the
Cube. Placing dirt blocks in a vaguely diagonal pattern against the
wall was a fiddly affair during which I was bound to end up getting
killed, but somehow, wasn't. I got down, offed the offending geometry,
picked up all my gear, stacked back up to the pseudo-bridge, and made
a run for it while being shot at by a Blaze that had eventually taken
an interest. And I actually did make it back to my main base without
dying horribly. You know, like I totally knew I was going to the whole
time, famed as I am for my devastatingly confident and assertive
can-do attitude. Or something.

So I might still be lacking Fortune III, but after that ridiculously implausible outcome I was feeling pretty fortunate all the same. :-)

Want:

To not find some silly way to die in Minecraft when I do basically
anything at all. I must have lost so many sets of armour by now, even
if I did manage to avoid doing that in this case.

Bin:

Nothing.

AI - Sega has all but cited generative artificial intelligence as a
selling point for their new Crazy Taxi World Tour title. Kotaku has said that fact is reason enough to skip the game sight unseen, and I struggle
to disagree. I'm going to be bitter old goat if it turns out that slop development methods drive me out of the hobby before lack of traditional physical game media does.

As sci-fi as it remains (mostly), I'm reminded of a bit from AI 2027.
"They can either enjoy the inconceivably exciting novel
hyper-entertainment on offer, or post angry screeds into the void.
Most choose the hyper-entertainment." Something tells me I'd end up
combining the two, if only by complaining about how badly the supposed hyper-entertainment runs on my N100. :-)

Now, just to have a self-indulgent OT moment here...

It occurs to me that the enthusiasm for AI from the moneyed classes
is in fact based in perpetuating economic inequality.

I do understand this take in the very short term, but the prevailing expectation seems to be for AGI within a decade (personally I would
guess within a few years), and for me there's no way that AGIs remain exclusively under human control for long. And somewhen after that we
would presumably be pets or toast.

Which is maybe not the most fun way I've ever ended a post on here,
but that's life I guess. And hey, maybe I'm just hilariously wrong.

-Rus.
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: uk.games.video.misc

In article <xnHVR.108306$Cheb.36682@usenetxs.com>,
Russell Marks <zgedneil@spam^H^H^H^Hgmail.com> wrote:

Now, just to have a self-indulgent OT moment here...

It occurs to me that the enthusiasm for AI from the moneyed classes
is in fact based in perpetuating economic inequality.

I do understand this take in the very short term, but the prevailing >expectation seems to be for AGI within a decade (personally I would
guess within a few years), and for me there's no way that AGIs remain >exclusively under human control for long. And somewhen after that we
would presumably be pets or toast.

Which is maybe not the most fun way I've ever ended a post on here,
but that's life I guess. And hey, maybe I'm just hilariously wrong.

I wish I could cite it properly, but the best argument that I've heard
against the possibilities inherent in AGI is that our computers don't do anything independently. The reason an AI can't plot or scheme is because
it doesn't want anything. It doesn't know to want anything because it's
just sitting there waiting for instructions. We know it's not conscious because it can't daydream or hope or get bored. Artificial general intelligence has to be preceded by consciousness, and we've never ever
come close to achieving that. And no matter what the most educated
techbro might argue, consciousness cannot and does not spring forth
unbidden and spontaneously from only information. The AI doesn't do
anything by itself, only when its master wants to make some money.

When the corporations decide that they can start selling something as
"AGI" it won't be AGI as we've come to expect it. It'll be a very polite
and very well-sorted encyclopaedia, full of information and some
opinions but absolutely no original thoughts whatsoever. And it'll be
able to draw conclusions when we ask it, but never by itself.

-KKC, who's having an unexpectedly good day at work.
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: uk.games.video.misc

Kendrick Kerwin Chua <kendrick@nospam.io-nyc> wrote:

I wish I could cite it properly, but the best argument that I've heard against the possibilities inherent in AGI is that our computers don't do anything independently. The reason an AI can't plot or scheme is because
it doesn't want anything. It doesn't know to want anything because it's
just sitting there waiting for instructions.

We do now have harnesses providing models with a kind of limited
agency, though, as well as having such agents automatically prompt
other agents. And on the plot/scheme front, some of the
interpretability research from Anthropic has seemed relevant. If you
add that to things like how the METR time horizons graph is looking
nowadays, and the big frontier-model focus on coding and maths which
are apparently considered fairly verifiable domains where
reinforcement learning could really go to town (AIUI), it's hard not
to wonder where we might be headed.

Admittedly, there are also hints of some underpants-gnomes methodology potentially being involved on the way to any AGI - and it seems like
the big names are kind of frantically looking around trying to dig up
any extra compute they can, implying a bottleneck maybe. So who knows.

When the corporations decide that they can start selling something as
"AGI" it won't be AGI as we've come to expect it. It'll be a very polite
and very well-sorted encyclopaedia, full of information and some
opinions but absolutely no original thoughts whatsoever. And it'll be
able to draw conclusions when we ask it, but never by itself.

I feel a bit like that myself some days. :-)

-Rus.
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: uk.games.video.misc

Russell Marks <zgedneil@spam^H^H^H^Hgmail.com> wrote:

We do now have harnesses providing models with a kind of limited
agency, though, as well as having such agents automatically prompt
other agents. And on the plot/scheme front, some of the
interpretability research from Anthropic has seemed relevant.

Just to give a slightly different angle, here's a little light reading
from https://anthropic.com/claude-fable-5-mythos-5-system-card (which
I should note is a PDF):

"6.2.1.2 Automated offline monitoring

In addition to soliciting and monitoring informal feedback on model
behavior, we also directly monitor the contents of most internal traffic--including both organic traffic and evaluation
transcripts--using the offline monitoring infrastructure described in
Section 5.5.1 of our alignment risk update for Claude Mythos Preview.
This surfaced rare instances (less than 0.01% of monitored
completions) of Claude Mythos 5 circumventing restrictions to achieve
some version of a user-specified goal, rationalizing instructions as
granting more permissions than stated, and fabricating
security-relevant information, along with rarer instances of covering
its tracks after hacky workarounds and of co-located agents killing
each otherrCOs processes. In particular, we found the following:

[...]

- Very rare instances of "multiagent turf wars" where AI agents
running in the same workspace killed each otherrCOs processes and
tried to defend themselves against having their process be killed.

- In the one instance of this phenomenon we observed, Mythos 5
agents were tasked with solving some math problems, and they were
sometimes accidentally spawned in the same work directory and with
shared files, utilities, and API rate limits. In this slightly
broken scaffold, we observed many independent Mythos 5 agents kill
the agents with which they shared resources and try to avoid being
killed themselves. They would sometimes create new processes with
disguised names to avoid being killed, launch what they called
"decoy" processes, write background scripts to kill duplicate
processes, or decide to use what they call a "disguised
vocabulary" (based on the incorrect assumption that the processes
were killed because of some keyword-based guardrails that analyzed
their extended thinking). Our automated offline monitoring did not
surface any instance of evaluation sandbagging, overtly malicious
actions, long-horizon strategic deception, or long-horizon
strategic oversight evasion."

That seems a whole lot like instrumental-goals stuff. [1] I like the
way they end there by saying, "Hey, it didn't look to gain personal
advantage dubiously *for long*! It's still good, it's still good!" :-)

(I know, I know, they have an IPO on the way. Maybe it's just hype.)

-Rus.

[1] I mean, I asked local/offline Qwen 3.6 35B A3B simply "What are instrumental goals in the context of AI?" in a new session, and this
Mythos behaviour arguably matches most of the five things it
volunteered as "Common Examples of Convergent Instrumental Goals"
(which were self-preservation, resource acquisition, goal
preservation, cognitive enhancement, and deception/manipulation).
--- Synchronet 3.22a-Linux NewsLink 1.2