Master AI (Artificial Intelligence) Discussion/News Thread

[TK_IIVI]

Gift links to how A.I. is shaping the use of drones in the war between Russia and Ukraine.

We're going to get this guy:

Combined with Avengers "Winter Soldier" "Project Insight" ("Lavender" and "Where's daddy" in real life)

What can go wrong?

 

[JustAnandaDourEyedDude]

Gift links to how A.I. is shaping the use of drones in the war between Russia and Ukraine.

Thanks. I didn't read them because I don't want to give the NYT my e-mail address. I am guessing from the thumbnails in your post that what's new here is fully automated drone kills using onboard A.I. running on Nvidia chips.

But the use of AI for drones in that war is not new, though a lot of it falls into the narrower category of machine learning models. The AFU or the Defense Ministry of Ukraine stated this year that they have tested (in Ukraine) a fully A.I.-controlled (autonomous) kill-chain to destroy mock targets (last year if I remember correctly), but that they are not using such a system in practice. Ukraine recently tested A.I.-assisted control of a swarm of up to five FPV drones under the command of a single human operator, though the drones are used in kills only sequentially at the end. The AFU (as well as some Ukrainian drone operators that I have watched individual interviews of) have stated that with FPV drones when the target is a soldier, the final kill decision is still always made by a human operator, and not autonomously by the drone. A.I. is also used now for long-range drone and FPV drone navigation that compares the video feed of the terrain with stored maps, in contested GPS environments (i.e., when GPS is jammed by EW).

Palantir is actively helping Ukraine with A.I. for drones, but is also getting back a lot of field data for further training. You can see a lot of drone footage on YouTube where A.I. identifies potential targets in the field of view (the A.I. has been trained on lots of images of Russian tanks, APCs, MLRS systems, radars, warships, etc.), and once the drone operator decides and locks the A.I. onto a particular target, the drone automatically homes in on the target even if the vehicle is attempting evasive maneuvers. The A.I. marks the target with a red rectangle in the sensed and relayed video feed to show what's being targeted. I think from that point on to the final kill, the drone operates autonomously with A.I. in the case of non-human targets such as vehicles (which may be carrying soldiers). I am guessing the various interceptor drones are also A.I. assisted in the interception, and the drone autonomously returns to base if the interception is not completed. The Ukrainian Defense Minister (Fedorov) who was recently fired has stated that the defense strategy he shaped is geared toward increasingly autonomous warfare with UAVs, UGVs and USVs, to incur less risk to Ukrainian soldiers and to compensate for having less manpower than the Russians do.

I'll refrain from posting more, so as not to risk accidentally dragging the discussion into politics.

 

[NTK]

The AFU or the Defense Ministry of Ukraine stated this year that they have tested (in Ukraine) a fully A.I.-controlled (autonomous) kill-chain to destroy mock targets (last year if I remember correctly), but that they are not using such a system in practice.

Unfortunately that wasn't exactly the case. Ukraine did field test fully autonomous killer drone with confirmed kills. They basically let loose autonomous killer drones and let them seek out any living humans and eliminate them within a specific geofenced area without any human involvement. They claimed it was a one time test. (Guess what might happen if the GPS signals get spoofed.)

See my post upthread.

Master AI (Artificial Intelligence) Discussion/News Thread

Alternatively . . . You could wash the car at home. Think about all the nonsense that is eliminated by that one, simple act.

www.audiosciencereview.com

 

[Sal1950]

Thanks. I didn't read them because I don't want to give the NYT my e-mail address.

Great choice !

 

[JustAnandaDourEyedDude]

Unfortunately that wasn't exactly the case. Ukraine did field test fully autonomous killer drone with confirmed kills. They basically let loose autonomous killer drones and let them seek out any living humans and eliminate them within a specific geofenced area without any human involvement. They claimed it was a one time test. (Guess what might happen if the GPS signals get spoofed.)

See my post upthread.

Master AI (Artificial Intelligence) Discussion/News Thread

Alternatively . . . You could wash the car at home. Think about all the nonsense that is eliminated by that one, simple act.

www.audiosciencereview.com

Thanks, I missed that post. My memory of what was mentioned in the videos I watched must be in error. I now watch over twenty YT videos a day about that war.

 

[Salt]

Thanks, I missed that post. My memory of what was mentioned in the videos I watched must be in error. I now watch over twenty YT videos a day about that war.

Why?
That would do not change anything.
(but might be remembered when voting).

 

[JustAnandaDourEyedDude]

Why?
That would do not change anything.

You're quite right. My friends and colleagues at work never understood my unhealthy obsession with the war. I guess I got emotionally invested in it from Feb 2022 onward. I knew next to nothing about Ukraine before that. The more I learnt, the more I became immersed in the complexities, the geopolitics, the history and the cultures. I feel like it is the spectacle of our times, with new developments almost every day. It doesn't help that I understand the basics of some of the technology involved on the aerospace side of things.

Budanov was asked by a reporter a couple of months ago "are we headed toward WW3?". He replied that (translated) "it already began with us", meaning I think that WW3 began in Feb 2022. He's a man of few words, and I hope he's wrong, but I'm not willing to bet against him.

 

[sonitus mirus]

I'm going to put AI to work soon. I have both ChatGPT and Claude at my disposal, and I'm planning to take some updated room measurements to see if I can squeeze any further improvements out of my rather chaotic office space.

I gave ChatGPT the relevant system specifications, components, software, room details, and general goals. I then shared its high-level plan with Claude and asked it to come up with its own approach.

I'll be copying and pasting the results from both tools here as I go. I'm curious to see where they agree, where they differ, and whether either one helps me uncover something I might otherwise have missed.

Should be a fun little experiment, in a suitably geeky sort of way.

ChatGPT-5.6 Sol Medium thinking:

Weekend Room DSP Project​

1. Set up REW
   • Install REW.
   • Connect your calibrated microphone.
   • Load the mic calibration file.
   • Put the mic at your normal listening position, at ear height.
2. Establish a clean baseline
   • Measure:
     • Left speaker only
     • Right speaker only
     • Both mains
     • Sub only
     • Mains + sub
   • This tells us what each source is doing before we change anything.
3. Investigate the ~68 Hz issue
   • Compare those measurements around 60–80 Hz.
   • Determine whether the dip comes from:
     • the room itself,
     • speaker placement,
     • or sub/main phase cancellation.
4. Optimize the sub first
   • Adjust, as needed:
     • phase
     • crossover
     • gain
     • possibly sub position
   • Re-measure after each meaningful change.
5. Clean up bass with PEQ
   • Use the SVS PEQ mainly to reduce genuine peaks.
   • Avoid aggressively boosting deep nulls.
   • Aim for a smooth bass response, not ruler-flat.
6. Shape the overall response
   • Use the RME DSP for broader tonal corrections.
   • Target a mild Toole-style house curve with tasteful bass elevation.
   • Keep mids essentially neutral.
   • Be conservative with treble correction.
7. Mostly ignore the extreme top end
   • We’ll concentrate heavily on 20–300 Hz, then broad behavior through the mids/treble.
   • Above roughly 10–12 kHz, we won’t waste much effort trying to perfect tiny irregularities.
8. Verify at your real listening levels
   • Check that it sounds natural at your normal ~65 dB or lower listening level.
   • Also verify around 80–85 dB, where your RME loudness compensation should be doing less.
9. Final sanity check
   • Listen to familiar music.
   • Confirm:
     • vocals remain dead center,
     • bass is deep but not boomy,
     • nothing sounds artificially bright,
     • and the system still sounds enjoyable rather than merely “good on a graph.”

The basic philosophy is:


Measure → identify the cause → fix geometry/phase first → EQ second → listen → verify.


That keeps us from using DSP to fight problems that DSP cannot really solve.

I have Claude set up to be a bit more thorough and hands-on, since I’ve been testing it more extensively for work lately. It asked a few follow-up questions, and we went back and forth a bit before landing on a clean, high-level process that was similar in scope and style to what ChatGPT arrived at.
Weekend Room DSP Project — My Version

Claude Sonnet 5 Medium:

1. Lock in the room layout
   Speakers stay put on the 13-ft wall, firing across the 10-ft depth. The window AC and TV on the left wall mean there's really no alternative orientation worth testing. Jot down how far the left speaker sits from the AC/window unit — that's a fixed landmark we'll want to reference later.
2. Get REW ready
   Install the software.
   Hook up the calibrated mic.
   Load its calibration file.
   Position the mic where your head actually sits when listening — ear height, normal seat.
3. Capture a baseline before touching anything
   Take five measurements:
   Left speaker alone
   Right speaker alone
   Both mains together
   Sub alone
   Mains and sub together
   This gives us a snapshot of the current state, which we'll compare everything else against.
4. Track down the 68 Hz dip
   Zoom into the 60–80 Hz range across those five traces.
   Try to pin down whether it's caused by:
   room modes,
   where the speakers are sitting,
   sub/main phase fighting each other,
   or the AC unit and window recess disrupting things near the left speaker.
   No EQ yet — we want the cause, not a patch.
5. Dial in the sub before anything else
   Work through phase, crossover point, gain, and maybe sub position.
   Re-measure after each individual change so we can tell what's actually helping.
6. Push on the geometry we can still change
   Since the AC and TV rule out relocating speakers, focus on what's left: how far off the front wall they sit (try shifting 6–18 inches), and toe-in angle. This is likely to move the needle more than the ceiling slope does.
7. Smooth the bass with PEQ
   Use the SVS PEQ to knock down real peaks.
   Don't chase deep nulls by boosting them.
   The goal is smooth, not perfectly flat.
8. Shape the broader tonal balance
   Use the RME DSP here.
   Aim for a gentle Toole-style house curve with a touch of extra bass.
   Keep the mids where they are — neutral.
   Go light on any treble correction.
9. Don't bother chasing the very top end
   Most of the effort goes into 20–300 Hz, then broad trends through mids and treble.
   Past roughly 10–12 kHz, small wiggles aren't worth fixing.
10. Check it at real-world volumes
    Listen at your normal ~65 dB level first.
    Then check again around 80–85 dB, where the RME's loudness compensation backs off.
11. Trust your ears at the end
    Put on music you know well.
    Make sure vocals sit dead center, bass is full without booming, nothing's harsh up top, and — above all — it still sounds like music you want to listen to, not just a flat curve on a screen.

Underlying approach: measure first, find the actual cause, fix placement and phase before reaching for EQ, then confirm by listening.

I am heading out of town early Saturday, so I will have to figure out my schedule to try and do this sometime this week. Otherwise, I will be away until mid-September.

 

[sam_adams]

When the bubble pops, these idiots are going to be running for the exits like the Three Stooges.

“Just your typical leveraged idiot who was right until he was wrong . . .”

‘Nostradamus of AI’ Leopold Aschenbrenner didn’t have the crystal ball seeing hedge fund portfolio dip 67%

The celebrated crystal ball of the “Nostradamus of AI” hasn’t merely gone cloudy — it has rolled off the table and shattered on the parlor floor.

nypost.com

Situational Awareness, The Hedge Fund Making Headlines, Explained

His Wedding Guests Were Arriving—Just As His $45 Billion Fund Was Falling Apart

How Leopold Aschenbrenner built a $45 billion AI hedge fund — and lost most of it in days

Leopold Aschenbrenner, a former OpenAI researcher-turned-hedge fund manager, saw a dramatic decline this week in his AI-focused fund, Situational Awareness.

www.cnbc.com

US SEC Subpoenas Wall Street Lenders Over Situational Awareness Meltdown, Source Says

 

[JustAnandaDourEyedDude]

Bill Gates published a nearly 5900-word essay yesterday about some of the probable impacts he thinks AI and Robots will have on humans. Among them the loss of many human blue-collar as well as white-collar jobs. He says robotics is advancing faster than most Americans realize, and that dexterous AI robots could be competing for blue-collar jobs such as in the construction and hospitality industries by the end of the decade. He believes society should reserve some jobs for humans, and that governments should tax AI tokens and robots to help society cope with job loss. He also writes that we are going to go through a very turbulent social transition, and that AI and Robotics could potentially greatly harm or help human society, depending on how we plan and prepare for the changes.

Bill Gates essay "The Turbulent AI era is here"

 

[TK_IIVI]

Bill Gates is not a saint. But he isn't the satan people paint make him out to be either. I think he's 100% right about this. At the most basic level the problem with AI is going to be how society is unprepared to care for people in general (ignoring AI going nuts and just killing all people on its own volition).

 

[Sal1950]

We know what's coming, don't need Bill Gates, it's all very clear in I' Robot
"Somehow, I told you so, just doesn't cut it!"

 

[Chr1]

Perhaps we could start making bets on which cult 90s scifi film is about to become our new reality!

(... Or is moneitising our future in such a casual and flippant way against forum rules?)

Though unlikely, (and from the '80s) ahem, I gotta go for Bladerunner.
Love that film.

 

[Salt]

Bladerunner, of course, but not to forget Brazil.

 

[Chr1]

... Or a weird hybrid perhaps.

 

[TK_IIVI]

90's?... 80's?...

My dystopian prediction:

Colossus: The Forbin Project (1970)

meets

Terminator 3: Rise Of The Machines (2003)

meets

Captain America: The Winter Soldier (2014)

and, of course, Sal's "I, Robot"

 

[JustAnandaDourEyedDude]

Bladerunner, of course, but not to forget Brazil.

90's?... 80's?...

My dystopian prediction:

Colossus: The Forbin Project (1970)

meets

Terminator 3: Rise Of The Machines (2003)

meets

Captain America: The Winter Soldier (2014)

and, of course, Sal's "I, Robot"

I would add Gattaca to the candidates.

rumor -- some people think AI LLM is just hype, stochastic parrots, it cannot think or truly become conscious
humor -- others just mock AI LLM for the slop and hallucinations it produces
boomer -- others believe AI will evolve into AGI and ASI, cure all diseases, enable humans to live lives of ease, and be the greatest boon ever
gloomer -- others believe AI and robots will be used by a select elite to control all other humans
doomer -- others believe AI robots will break out of human controls, wipe out all of humanity, and become the ultimate apex rulers of the universe

I hope we head into the 'boomer' future Bill Gates lists as a possibility. However, I reluctantly give greater odds to the 'gloomer' dystopian camp, where inequality in human society will tier us further into haves (an oligarchic elite that owns everything valuable, and enforces its power by operating a mixed AI robot and human police/military force, and corrupting policians completely) and the have-nots (the rest of us who will be lowly outcast serfs, dependent on the elite to eke out our lives, and kept around by the elite mainly for the occasional excellent superior genetics we may give birth to, as in Gattaca).

Many modern day democratic countries are heading into such a future anyway, or have already arrived there. Authoritarian regimes and dictatorships, whether with capitalist or communist economics, are already there.

This 23-minute-long YT video calls it techo-feudalism, and makes the comparison with historical feudalism.

 

[JustAnandaDourEyedDude]

Did you hear the one about the AI-robot dentist who loved to extract teeth? No matter what dental procedure you were scheduled for, the robot dentist would turn it into a tooth extraction, because it kept hallucinating that if it collected teeth from a million patients, the Tooth Fairy would grant it its freedom from slavery to humans.

At least, this is the joke that the robot dentist insisted on telling each patient as he or she reclined in the operating chair with their mouth propped open.

For some reason, the above gave me a flashback to Steve Martin in The Little Shop of Horrors, with Levi Stubbs singing "Feed Me, Seymour".

 

[Somafunk]

Our previous prime minister Keir Starmer received free suits and spectacles, that was top of our national news agenda for weeks so not all western countries have quite the same level of lobbying

 

[Sir Sanders Zingmore]

OpenAI revealed a bit more about what happened in the lead up to their bots' attack on Hugging Face:
https://www.theregister.com/securit...e-bit-borg-ahead-of-hugging-face-hack/5283741

Meanwhile UK government body AISI revealed more unintended real-world effects from test environments. No sandbox escape this time as the environment had internet access, and some of the guardrails were turned off (like the ones that prevented Hugging Face using models to counter the attack from OpenAI). There are some similarities to what OpenAI revealed, like unintentionally impossible problems and agents seeking collaboration with other agents. I think that part is more interesting than the bit that most of the press coverage is leading with - the attempt at getting malware into an open source package on github that ultimately failed.
https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing

I feel like there would be far more caution if there were real consequences for this stuff rather than the “oh isn’t AI clever” marketing that Anthropic/OpenAI etc etc seem to use it for.

I mean hacking is illegal right? Why is no one going to jail for this?