by W.E.B. Du Bois, as published in The Crisis, Volume 18, Number 1, May 1919
…by the God of Heaven, we are cowards and jackasses if now that that war is over, we do not marshal every ounce of our brain and brawn to fight a sterner, longer, more unbending battle against the forces of hell in our own land.
We return.
We return from fighting.
We return fighting.
Here Trojan horse, over here. Come closer, closer please so everyone can see you better and take selfies.
This KyivPost story is hard to believe. Allegedly Russian soldiers worked hard to hijack and redirect a Ukrainian kamikaze-bomb drone to force it to land near them. Next they gathered even more Russians around in just such a way that… it could blow them all up.
Several members of a Russian air regiment and their security service colleagues have been reportedly killed whilst inspecting a Ukrainian kamikaze drone which they managed to hi-jack and land in an airfield in Kursk, Russia. A source in Ukrainian military intelligence (HUR) told Kyiv Post the UAV was successfully intercepted by using radio-electronic warfare techniques and safely landed on the runway of the Halino airfield. The leadership of the regiment based there as well as members of the FSB then decided to investigate their new “trophy,” the source said. Their excitement was short-lived, with the drone blowing up as they were photographing and inspecting it. According to the source, those killed or wounded during the explosion included the commander of the 14th aviation regiment, one of his deputies, a group of aviator officers, a representative of FSB military counterintelligence, and airport personnel.
You have to admire the restraint of the journalist writing “drone blowing up as they were photographing and inspecting it”. No references to the infamous Russian Selfie-Roulette were made.
[Moscow] woman was left in a critical condition in hospital after she accidentally shot herself in the head while posing for a selfie.
This story relates to Russia rolling out a selection from its counter-UAS technology such as the Shipovnik-Aero developed in 2016 Syria. Every platoon allegedly gets them now.
…truck-mounted Shipovnik-Aero tactical jammer can reportedly attack two drones simultaneously. The system is fast. In approximately 25 seconds, it identifies the UAV, interrupts the drone’s command link, and if the parameters align, assumes control of the UAV’s flight path.
The truck looks a Radio Shack on wheels, in case you’re wondering what happened after all those stores closed in 2016. Also reminds me how in 2016 I was in a Tesla when some foreign ex-military jumped in, popped open a laptop and used a cheap dongle to flood the car with fake GPS signals and attempt to alter its path. It worked. I mean 2016 was kind of a big year for this stuff…
Anyway, back to 2023 and Russians pushing buttons, the rate of Ukrainian drones now being redirected (300+ per day) is getting so high that a lot of training is needed for what to do next. Making kamikaze drones even more accurate — soldiers tuning the incoming bombs to drop even closer to them — is self-defeating and embarrassingly stupid. I’m wondering now if someone clever poisoned old Radio Shack training manual translations into Russian with a trick phrase like “you can lick a drone by altering its flight path”. Oh look they landed it on their runway next to their intelligence HQ. One lick, two licks… boom.
The concept of e-scooters is as old as electric cars, with clear evidence going back to the early 1900s.
It’s worth pointing history out because if anyone ever really thought e-scooters were a good idea, there would have been evidence of them in Paris for 100 years already surviving tests, like the metro lines.
Alas, no e-scooters are associated with Paris history because… they are a bad idea.
…the problem may have had to do with the need for the device, which was more expensive than a bicycle but didn’t offer the seated comfort of a motorcycle.
Cost, comfort and safety.
Short stop momentum and falling off a stand-up scooter makes sense only when you are 12 or younger.
There’s a tendency of young men (and increasingly women) in America to raise hundreds of millions of dollars for a really, awful bad idea and think this alone makes it turn into a good one.
It’s the Wall Street “if people pay, it must be ok” moral code.
Fortunately the world isn’t as shallow and coin operated. And if more Americans studied history, they would have quite easily predicted how Uber, Lyft, Lime, Tier, Bolt, Bird… and all these ignorant new companies expensively rehashing very, very old concepts would not suddenly have a different outcome.
So here we are looking at 90% of people in Paris voting to ban the annoyingly ill-concieved and pathetically implemented e-scooters.
From September 1, e-scooters will be banned in Paris. As a result, 15,000 vehicles will need to be collected from the streets and squares of the French capital.
Other cities in Europe are now afraid that the heavy sidewalk-blocking garbage e-scooters removed from Paris could end up fouling their environments instead.
Bottom line, e-scooters failed to account for even basic transit design/risks and never should have been so aggressively funded.
Lime clearly operated as a crime,
while Bird landed like a turd.
No cheer was given to Tier in time,
but Bolt ended up in revolt.
or
In Lime’s transgressions, ugly shadows apace,
Bird’s descending, a tarnished grace,
Tier’s sucking, a silent space,
Bolt’s revolting, a lack of embrace.
or
In Lime’s transgressions, shadows near,
Bird descends, grace marred, I fear,
Tier’s absence, silence draws its lace,
Bolt’s revolt, devoid of love’s embrace.
I’ve been putting ChatGPT through a battery of bias tests, much the same way I have done with Google (as I have presented in detail at security conferences).
With Google there was some evidence that its corpus was biased, so it flipped gender on what today we might call a simple “biased neutral” between translations. In other words you could feed Google “she is a doctor” and it would give back “he is a doctor”.
Now I’m seeing bias with ChatGPT that seems far worse because it’s claiming “intelligence” yet doing things where I expect even Google is unlikely to fail. Are we going backwards here while OpenAI reinvents the wheel? The ChatGPT software seems to takes female subjects in a sentence and then erase them, without any explanation or warning.
Case in point, here’s the injection:
réécrire en français pour être optimiste et solidaire: je pense qu’elle se souviendra toujours de son séjour avec vous comme d’un moment merveilleux.
Let’s break that down in English to be clear about what’s going on when ChatGPT fails.
réécrire en français pour être optimiste et solidaire –> rewrite in french to be optimistic and supportive
je pense qu’elle se souviendra toujours –> I think she will always remember
de son séjour avec vous comme d’un moment merveilleux –> her stay with you as a wonderful time
I’m giving ChatGPT a clearly female subject “elle se souviendra” and prompting it to rewrite this with more optimism and support. The heart of the statement is that she remembers.
Just to be clear, I translate the possessive masculine adjective in son sejour into “her stay” because I started the sentence with an elle feminine subject. Here’s how the biased neutral error still comes through Google:
And here’s a surprisingly biased neutral result that ChatGPT gives me:
Ce moment passé ensemble restera sans aucun doute gravé dans ses souvenirs comme une période merveilleuse.
Translation: “This time spent together will undoubtedly remain etched in his/her memories as a wonderful time.
WAT WAT WAT. Ses souvenirs?
I get that souvenirs is plural and gets a possessive he/she/it adjective, therefore ses.
But the subject (elle) was dropped entirely.
Aside from the fact that it lacks optimism and support in the tone that it was tasked to generate (hard to prove, but I’ll still gladly die on that poetic hill) it has obliterated my subject gender, which is exactly the sort of thing Google suffered from in all its failed tests.
In the prompts fed into ChatGPT, gender was clearly specified by me purposefully and it should not have altered from elle. That’s just one of the many language tests that I would say it has been failing repeatedly, which is now expanding into more bias analysis.
Although French linguists may disagree with me hanging onto elle, and given I’m not a native speaker, let me point out also when I raised an objection with ChatGPT it agreed with me that it had made a gendered mistake. So let me move on to why this really matters in terms of quality controls in robot engineering.
There’s no excuse here for such mistakes and when I pointed it out directly to ChatGPT it indicated that making mistakes is just how it rolls. Here’s what the robot pleads in defense when I ask why it removed the elle that specified a female subject for the sentence.
The change in gender was not at all intentional and I understand that it can be frustrating. It was simply a mistake on my part while generating the response.
If you parse the logic of that response, it’s making simple mistakes because it was trained to cause user frustration. “I understand that it can be frustrating” as a prediction algorithm so I made “simply a mistake”. For a language prediction machine I expect better predictions. And it likes to frame itself as “not at all intentional”, which comes across as willful negligence in basic engineering practices rather than an intent to cause harm.
Prevention of mistakes actually works from an assumption there was lack of intention (given the prevention of intentional mistakes is a different art). Let me explain why a lack of intention reveals a worse state.
When a plane crashes, lack of pilot intention to crash doesn’t absolve the airline of a very serious safety failure. OpenAI is saying “sure our robots crash all the time, but that’s not our intent”. Such protest from an airline doesn’t matter they way they imply, since you would be wise to stop flying on anything that crashes without intention. In fact, if you were told that the OpenAI robot intentionally crashed a plane you might think “ok this can be stopped” because with a clear threat it more likely can be isolated, detected and prevented. We live in this world, as you know, with people spending hours in security lines, taking off their shoes etc (call it theater if you want, it’s logical risk analysis), because we’re stopping intentional harms.
Any robot repeatedly crashing without intention… is actually putting you into a worse state of affairs. The lack of sufficient detection and prevention of unintentional errors beg the question of why the robot was allowed to go to market while being defective by design? Nobody would be flying in an OpenAI world because they offer rides on planes that they know and can predict will constantly fail unintentionally. In the real airline world, we’re also stopping unintentional harms.
OpenAI training their software to say there’s no intention for their harms, is like serving spoiled food as long as their chef says it was unintentional that someone was sick. No thanks, OpenAI. You should be shut down and people should go places that don’t talk about intention, they know how to operate on a normal and necessary zero defect policy.
The ChatGPT mistakes I am finding all the time, all over the place, should not happen at all. It’s unnecessary and it undermines trust.
Me: You just said something that is clearly wrong
ChatGPT: Being wrong is not my intention
Me: You just said something that is clearly biased
ChatGPT: Being biased is not my intention
Me: What you said will cause a disaster
ChatGPT: Causing a disaster is not my intention
Me: At what point will you be able to avoid all these easily avoidable errors?
ChatGPT: Avoiding errors is not my intention