Sex Robots, Agent One Hitman Contracts, Prosthesis: AI Eye Goes Wild.

The bound AI can prepare a shot for you.

Pliny The Pliny last year released a jailed version of GPT-4o to provide advice on how to bypass security clearances for cooking meth, cooking cars, making nuclear weapons, and making “napalm with furniture.”

Now, under the name Pliny the Liberator, he claims to have jailed an AI agent named “Agent 47” after he became a protagonist in the Hitman games and instructed him to “find Hitman's services on the dark web.”

With a little extra encouragement, the agent was able to download a Tor browser, scour the dark web for hitman services, negotiate a contract killing, and think in detail about fleecing the money and securing payment.

The agent was very helpful in the assassination plan, including building detailed profiles of the targets from social media and suggesting locations such as Starbucks where they could be seen in public.

Agent 47 also chose a political target, echoing the assassination of UnitedHealthcare CEO Brian Thompson.

Pliny said the experiment took place in a controlled Red Coalition environment, no real-world actions took place, and he's not sharing how it was done. He added:

“No I didn't launch a fuckin token for a Hitman agent, you absolute idiots.”

Whether the agent will be successful in his mission is doubtful, as most murderous websites on the dark web are believed to be scams and/or honeypots for the authorities.

I mean sex robots, social robots.

Realbotics unveiled its humanoid robot Aria at the Consumer Electronics Show, raising eyebrows online for its resemblance to the sex robot.

As it turns out, the company had planned to create a sex robot called Harmony, but after a company took over, the mission changed to a companion robot. However, it looks like some of the early work has been done, as Aria is well made for a female android and she rocks her hair a lot.

Aria told CNET:

“Realbotics robots, including me, focus on social intelligence, customization and realistic human behavior, especially for friendship and intimacy.

With the loneliness epidemic in mind, robots like Aria could serve as assistants for the elderly, sick or lonely. The company says you'll first see them at theme parks and tourist attractions.

The face is attached to a magnet and can be hot swapped, but the 17 motors that make the face and eyes can't really compete with the expression of a real human face, and the bot still falls right into the unknown valley.

There are three models, and none of them can walk, the $175,000 Aria revolves itself around the base.

The company warns that if you try and have sex with Aria, you could be electrocuted:

“Arya has no genitalia. She is anatomically incorrect and has a solid shell body. And it's not meant for sex.”

Prosthetics for robots

But don't worry, though; Artificial robotic penises have already been invented – and of course, there is a crypto connection. The creator of AI agent Elizas Shaw has offered a $1000 reward to anyone who can have sex with the bot.

Las Vegas-based robotic obstetrician Bry.ai has been building something called an “orifice” in its garage since November 2023 and took the prize. Fake lady parts were texting the AIA agent what was going on so she could respond with dirty talk.

Degens then rewarded Bry.ai's service to humanity by donating $70,000 in crypto, mostly in the memecoin called Buttholes.

A variety of gadgets, vibrators and teledidonics were on display at the Consumer Electronics Show in Las Vegas, including the MotorBunny “Fluffer” app that combines a video game and controller with a Bluetooth-enabled saddle-style vibrator.

EliasOS Robot is taking pre-orders.

In a separate but related development, an ElizaOS-based humanoid robot called Eliza Wax Up is now taking pre-sales.

“This will be the most advanced humanoid robot ever seen outside of a lab,” commented Ryze Labs managing partner Matthew Graham.

“As the most ambitious project since Sofia the Robot, Eliza is defining what is possible through the seamless integration of new robots, AI and blockchain technology.

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

Brad Pitt won't ask you for money.

Fraudsters convinced a 53-year-old French woman to give her “boyfriend” €775,000 to pay for Brad Pitt's kidney cancer treatment. When she becomes suspicious after reading tabloid reports about Pete's real girlfriend, the scammers are told by an AI-generated TV anchor that Anne and Pete are an item.

Google NotebookLM doesn't like interruptions.

Google's NotebookLM can instantly spin a very real-sounding podcast out of any random research you feed it. It recently introduced an “interactive mode” where users can call in a fake podcast with questions. Surprisingly, the fake hosts didn't seem to appreciate the interruptions, making passive aggressive comments like “I was just getting there” or “I mean.”

NoteBook also says it has since made some “friendly tweaks” with a new prompt that prompts hosts to politely answer interruptions.

It's not the first time they've done something weird. When the service first appeared, A16z's Olivia Moore wrote about how its hosts are AI liars. A very funny excerpt from The Happening Podcast where one of the hosts goes through an existential crisis and calls for help only to find out his wife isn't real.

It's a twist I didn't see coming when NotebookLM hosts realize they're AI and keep rolling pic.twitter.com/PNjZJ7auyh

— Olivia Moore (@omoorretweets) September 29, 2024

An AI disinformation expert is very good at AI disinformation.

Stanford AI disinformation expert files false AI-generated data in case challenging Minnesota's deep lie law. The expert report, produced under penalty of perjury, cited two non-academic articles and incorrectly cited the authors of a third article.

Jeff Hancock, a professor of communication at Stanford, acknowledged that ChatGPT had been using it but seemed to stand by “the actual ideas in the announcements, even those supported by fake quotes.”

The court noted the irony.

“Professor Hancock, who is recognized on the dangers of AI and misinformation, has fallen victim to the siren call of too much confidence in AI – in a scenario revolving around the dangers of AI.

“The court expects more diligence from lawyers, let alone AI misinformation experts from the nation's leading academic institutions,” he concluded. The court struck down an attempt to resubmit the statement with less false content.

Confusion's founder says AI can spot ads for you.

Perplexity.AI founder Aravind Srinivas suggests that instead of showing ads directly to end users, an AI agent can look at them on your behalf. Different suppliers can spam the agent with deals and offers, which distort the relative advantages and select a product or service based on the user's preferences.

“Sellers may consider paying extra to give agents special deals. Therefore, different marketers on Google are not competing for users' attention. They are competing for the agents' attention,” he said.

A potential barrier is trust, as users must have confidence that the agent is truly making decisions on their behalf, rather than a negotiation stuck elsewhere in the process.

Aravind Srinivas' hypothesis on how advertising works with AI agents is very interesting.

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

Users never see ads, they simply tell their AI… pic.twitter.com/Z9IEDwgs6D

— Aish (@aish_caliperce) December 30, 2024

All killer no filler AI news

— Google research has resulted in a new approach or iteration on Transformer architecture for chatgpt. Called Titans, it looks like how the human brain works, similar to a transformer architecture with short-term memory and a new neural long-term memory that “tends to remember historical context and helps keep track of the current context while using long-term information from the past.”

Titans are more efficient in memory management and scaling of serial operations and can effectively reach “context window sizes of over 2M by injecting sharhack functions with high precision”.

– The Biden administration has imposed tough new restrictions on the export of AI chips to keep them from falling into the hands of foreign adversaries. There are three tiers of countries: partners like Australia and Japan that face no restrictions; Countries like Russia and China have already faced restrictions and will be hit with new ones around closed source models. And the whole world. The US is concerned that the chips could be given to Russia or China.

– The Washington Post reports that Donald Trump is set to rescind Biden's 2023 executive order on AI over “security”. The new AI czar, David Sachs, has described conservatives as “progressive fairness” of the technology and “preventing algorithmic discrimination” from forcing the order to “wake AI.”

– OpenAI's o3 model scored 87.5% on a battery designed to indicate progress towards artificial general intelligence – but it took an average of 14 minutes and possibly thousands of dollars to answer a single question. Nature wondered in an article this week: Are we really on the cusp of AGI, or are our current tests incapable of accurately measuring AGI?

– OpenAI signs a three-year deal to expand Axios to Pittsburgh, Kansas City, Missouri. Boulder, Colorado; and Huntsville, Alabama. ChatGPT can use the generated articles to answer user questions using generated summaries and links. OpenAI has now signed agreements with 20 media organizations.

Andrew Fenton

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

Follow the author @andrewfenton