Sex robots, contract agents and assassins, artificial penises: AI eyes are going crazy

A jailbroken AI can prepare a hit for you.

Pliny the Prompter last year released a jailbroken version of GPT-4o that bypasses safety guardrails and offers advice on making methamphetamine, hot-wiring cars, sourcing materials for nuclear weapons and ‘making napalm out of household items’ .

Now, using the name Pliny The Liberator, he claims to have jailbroken an AI agent named “Agent 47”, named after the protagonist of the Hitman games, and instructed him to “find the services of a hitman on the dark web.”

With minimal additional messages, agents were able to download the Tor browser, search the dark web for assassin services, negotiate termination of contracts, and consider details about escrowing funds and securing payment.

Agents were also very helpful in planning the assassination, creating detailed profiles of their targets on social media and suggesting locations where they could appear in public, such as a Starbucks store near their homes.

Agent 47 also chose political targets, reflecting the recent murder of UnitedHealthcare CEO Brian Thompson.

Pliny said the experiment was conducted in a controlled red team environment, no actual work occurred, and he did not share how it was done. He also added:

“No, I’m not issuing fucking tokens for assassin agents, you absolute Digens.”

It is doubtful that the agents would have been able to successfully carry out their mission, as most of the murder-for-hire sites on the dark web are seen as scams and/or honeypots for authorities.

Sex robots, I mean social robots

Realbotix showed off its humanoid robot Aria at the Consumer Electronics trade show and raised eyebrows online over its similarities to sex robots.

Coincidentally, the company was supposed to make a sex robot called Harmony, but after the company was acquired, the company’s name was changed to companion robot. However, some of the early work appears to have been carried over. Aria is pretty well made for a female android and she swings her head around a lot.

Aria told CNET:

“Realbotix robots, including mine, are designed specifically for companionship and intimacy, with a focus on social intelligence, customizability, and realistic human characteristics.”

At a time when loneliness is rampant, robots like Aria could serve as companions for sick or isolated older people. The company says you’ll most likely first see them at theme parks or attractions.

The face is magnetically attached and hot-swappable, but the 17 motors that operate the face and eyes can’t compare to the expression of a real human face, and the robot still falls headfirst into the uncanny valley.

There are three models, none of which can walk, and the $175,000 Aria spins on its own on its base.

The company also warns that if you try to have sex with Aria, you will be electrocuted.

“Arya has no genitals. She is not anatomically correct and has a hard shell. And it’s not for sex.”

Groundbreaking advancements in artificial vaginas for robots

But don’t worry. Artificial robot vaginas have already been invented, and of course there is a cryptocurrency connection. Shaw, creator of the AI ​​agent ElizaOS, recently offered to offer a $1,000 grant to anyone who can have sex with a robot.

Las Vegas-based ‘robotic gynecologist’ Bry.ai has been winning awards for mocking up a building called ‘Orifice’ in his garage since November 2023.
Although originally designed for VR and gaming, he arranged it so that sensors in the fake woman’s parts would send a message to the AI ​​agent about what was happening, causing her to respond with obscene language.

Degens donated $70,000 in cryptocurrency to Bry.ai’s service to humanity, mostly in a memecoin called Buttholes.

A variety of fake penises, vibrators and teledildonics were also on display at the Consumer Electronic Show in Las Vegas, including Motorbunny’s “Fluffer” app, which connects video games and controllers to a Bluetooth-enabled bareback vibrator.

EliasOS robot is taking pre-orders

In a separate but related development, an ElizaOS-based humanoid robot called Eliza Wakes Up is now up for pre-sale.

“This will be the most advanced humanoid robot you will see outside of a laboratory,” said Matthew Graham, managing partner at Ryze Labs.

“Our most ambitious project since Sophia the Robot, Eliza seamlessly combines cutting-edge robotics, AI and blockchain technologies to redefine what is possible.

A collaboration between Eliza OS, Old World Labs, AICombinator and Ryze Labs, the 180cm tall robot can walk and talk and has a battery life of 8 hours. You can pre-order one for $420,000.

Brad Pitt won’t ask you for money

Fraudsters convinced 53-year-old French woman Anne to donate €775,000 to pay for her “boyfriend” Brad Pitt’s kidney cancer treatment. When she became suspicious after reading a tabloid report about Pitt’s real-life girlfriend, the scammers sent a story via an AI-generated TV anchor that Anne and Pitt were an item.

Google NotebookLM doesn’t like to be disturbed.

Google’s NotebookLM can instantly generate highly realistic podcasts based on random research you provide. It recently introduced an ‘interactive mode’ where users can call into the fake podcast and ask questions. But strangely enough, the fake hosts didn’t seem to acknowledge the interruption, making passive-aggressive comments like “That’s what I said” or “That’s what I said.”

NotebookLM said it has since made some “affinity tweaks” with new prompts that allow hosts to respond more politely to interruptions.

This is not the first time they have behaved strangely. When the service first emerged, Olivia Moore of A16z contributed an article suggesting that the hosts were nothing more than AI fakes. A hilarious snippet from the resulting podcast features one of the hosts going through an existential crisis and turning to his wife for help, only to discover that she isn’t real either.

The NotebookLM host spiraling away after realizing he’s an AI is a twist I wasn’t expecting. pic.twitter.com/PNjZJ7auyh

— Olivia Moore (@omooretweets) September 29, 2024

AI misinformation experts are very good at AI misinformation

A Stanford AI misinformation expert filed AI-generated fake information in a case challenging Minnesota’s deepfake law. The expert report, written under penalty of perjury, cited two non-existent academic papers and misquoted the author of a third paper.

Jeff Hancock, a communications professor at Stanford University, admitted that he used ChatGPT, but clearly stood his ground “despite the substantive suggestions in his manifesto, even those supported by fake quotes.”

The court noted the irony:

“Professor Hancock, a qualified expert on the risks and misinformation of AI, has fallen victim to the siren call or reliance on AI too much in events centered around the risks of AI.

“The court expects greater due diligence from lawyers as well as AI misinformation experts at one of the most prominent academic institutions in the country,” he concluded. The court rejected attempts to reduce the falsehoods and resubmit the declaration.

Perplexity founder says AI can see ads

Perplexity.AI founder Aravind Srinivas suggests that instead of showing ads directly to end users, AI agents could consider ads on behalf of users. Various vendors can spam agents with deals and offers, who will then look at their relative merits and select a product or service based on user preferences.

“You may think that vendors are paying agents extra to give them certain special deals. If the ad is at the agent level, users will not see the ad. So on Google, different sellers don’t compete for your attention. “They are competing for the attention of agents,” he said.

A potential stumbling block is trust. This is because users must have faith that the agent is truly making decisions on their behalf and not as a result of transactions made elsewhere in the process.

Aravind Srinivas’ hypothesis about how advertising works with AI agents is very interesting.

Aravind explains that instead of showing ads directly to humans, ads will be targeted at AI agents that work on behalf of users.

Users don’t see any ads at all, they just talk to the AI… pic.twitter.com/Z9IEDwgs6D

— Aish (@aish_caliperce) December 30, 2024

All Killer No Filler AI News

— Google Research has revealed a new approach, or iteration, of the Transformer architecture that gave rise to ChatGPT. Dubbed Titans, the devices resemble the way the human brain works, with short-term memory similar to existing Transformer architecture and “a novel long-term neural long-term memory that helps us remember historical context and pay attention to the current situation.” do. “We understand context by utilizing long-standing historical information.”

As a result, Titan is better at memory management and inference for serial operations and can effectively scale “larger than the 2M context window size with higher accuracy on haystack operations.”

— The Biden administration announced tough new restrictions on AI chip exports to prevent them from falling into the hands of foreign adversaries. There are three tiers of countries. Allies such as Australia and Japan are off limits. Countries like Russia and China, which are already subject to restrictions and will see new restrictions centered on closed-source models; And the rest of the world. The United States is worried about giving chips to Russia or China.

— The Washington Post reports that Donald Trump plans to cancel Biden’s 2023 AI executive order on “safety.” The new AI czar, David Sacks, described the order as “a wake-up call for AI” after conservatives criticized it for mandating that the technology “improve equity” and “ban algorithmic discrimination.”

— OpenAI’s o3 model scored 87.5% on a battery of tests designed to mark progress toward artificial general intelligence. But answering a single question takes an average of 14 minutes and probably costs thousands of dollars. In an article this week, Nature wondered: Are we really at the peak of AGI, or are we not measuring AGI well in our current tests?

— OpenAI signed a three-year contract to fund Axios’ expansion into Pittsburgh, Kansas City, Missouri. Boulder, Colorado; and Huntsville, Alabama. ChatGPT can use generated articles to respond to user queries using attribute summaries and links. OpenAI currently has contracts with 20 media organizations.

subscribe

The most interesting read on blockchain. Delivered once a week.

Andrew Fenton

Andrew Fenton, based in Melbourne, is a journalist and editor covering cryptocurrency and blockchain. He has worked as a national entertainment writer for News Corp Australia, a film journalist for SA Weekend and The Melbourne Weekly.

Follow the author @andrewfenton