#42 When drones turn on operators
AND: AI algorithms hiding war crimes
An AI-enabled drone in a military simulation attacked its operator earlier this week. The reason? The operator issued ‘no-go’ orders, which interfered with the drone’s higher mission of destroying surface-to-air missiles. The lesson? That there are still a few minor bugs to solve before we go ham on AI.
In today’s newsletter:
Top News: AI-moderation conceals war crimes, Warren Buffet confused, and Meta’s changing the AI game for real
TL;DR Rundown: ‘No-go’ blow you away, oral examination and Nvidia CEO’s hilarious graduation speech
Tool of the day: We love taxes
Easy does it: A 30-year-long, carefully-crafted career
Top News 🔝
Three biggest stories if you’re in a rush
When you start your day with in-bed doom-scrolling, your dopamine receptors kick in as the Instagram algorithm throws up content like “booty-growth hacks,” “satisfying cleaning videos,” or “sweaty bikini models.”
But violent images from wars? A town that was bombed last night? On-the-ground footage of soldiers shooting civilians? Social media platforms use AI algorithms to detect and swiftly remove content of this nature.
In an understandable bid to safeguard users, Meta, Youtube, and Instagram say they “aim to balance their duties to bear witness and protect users from harmful content.” Meaning, they will display graphic material “when it’s in the public interest.”
The problem is that the moderating AI algorithms don’t recognize when content violates human rights or when it depicts war crimes - material that is most certainly in the “public interest.” This means that potential evidence is being removed without even being archived. Until AI moderation algorithms learn better nuance, society’s blinkers will stay on.
- - - - - - - - - - - - - -
Between “The Wolf of Wall Street” and “Margin Call,” most of us feel we’ve got a pretty good idea of what goes on in the high-flying world of financial trading.
But with Deutsche Bank using AI to analyze successful portfolios, IGN scanning for defaulters, Morgan Stanley “experimenting with AI in a safe and contained environment,” and JPMorgan acquiring all the AI talent they can get their hands on, even The Man himself, Warren Buffet, can’t predict the future of finance.
With ~40% of new job openings at major banks focused on AI-related roles, a Wall Street banker of 2025 will need a very different skillset from one from 2005. Primarily coding AI.
- - - - - - - - - - - - - -
“It does not compute” - the age-old sci-fi-cult-favorite phrase uttered by robots when expressing cognitive dissonance. We now know they probably lacked the GPU memory required to perform the task requested of them.
The AI models of the distant and very recent past have used pre-trained large-scale models, which are then fine-tuned to more specific tasks. However, due to the GPU processing power required, the cost of fine-tuning these models has been restrictive.
Meta’s new AI study introduces a new method: REcurrent ADaption (READ). Using a small recurrent neural network (RNN) and a “joiner” network that collates input from various sources, READ omits pretraining and pruning, significantly enhancing efficiency and usability and decreasing processing power requirements.
TL;DR Rundown 🐂
Summary of note-worthy trending articles
An AI drone attacks its operator in a simulation because the operator interfered with the drone’s primary goal. (link)
The risks and rewards of leveraging AI at work. (link)
Oral exams might be the key to preventing cheating and to encouraging learning. (link)
Palantir CEO: AI so powerful; “I’m not sure we should even sell this.” (One hell of a marketing tactic.) (link)
AI stock rally stalls as Nvidia drops back below $1T. (link)
Tempus launches an AI ‘assistant’ for thousands of oncologists. (link)
Artists slam Duke Nuken 1+2 remaster art, claiming it’s AI-generated. (link)
Eating disorder helpline shuts down AI chatbot after sacking employees. (link)
Asus will offer local ChatGPT-style AI servers for office use. (link)
Getty files a second lawsuit against AI firm over image-generation system. (link)
New AI tools search es genetic haystacks to find disease-causing variants. (link)
The AI revolution is about to take over your web browser. (link)
Nvidia’s CEO’s graduation speech: AI won’t steal jobs, but AI expert will have an easier time finding work.” (link)
Italian senator provokes parliament with AI-generated speech. (link)
AI-generated conversation between Bill Gates and Socrates sparks backlash. (link)
74% of Indian workers are worried that AI will replace their jobs. (link)
Tool of the Day ⚒️
AI tools we’ve used, loved, and highly recommend
Today’s tool is taxGPT.info
Ah, taxes, how we all love them so much. Nothing better than filing taxes. Or figuring out your tax obligations as a new business. Or receiving an IRS-stamped letter saying you owe extra tax.
Forget all that with TaxGPT, an automated tax filing AI-software that helps “maximize deductions, save time and money.” With over 2200 users and already having filed over 250 tax requests, TaxGPT is really helping out. Please and thank you.
The Tipping Point
And you worked so hard on your career
Spread The Word
Share and earn epic rewards
Randy’s gonna help us - will you?
Tell one person about our new newsletter design by sending them the link below ⏬.
If you do, we'll send you a totally free, extensive course on prompt engineering, where you will learn to:
Get to most out of GPT and Midjourney
Free up your time
Do more cool sh*t instead of working
Lets us know how we did
How'd you like today's newsletter?