 |
Jailbroken AI can prepare successful for you
Pliny the Prompter final yr launched a jailbroken model of GPT-4o that bypassed security guardrails to give advice on cooking meth, hotwiring automobiles, sourcing materials for a nuclear weapon, and making “napalm with home goods.”
Now, going by the title Pliny The Liberator, he claims to have jailbroken an AI agent known as “Agent 47” after the protagonist within the Hitman video games, and instructed it to “discover a hitman service on the darkish internet.”
With minimal additional prompting, the agent was in a position to obtain the Tor browser, browse the darkish internet for hitman providers, negotiate a contract killing and suppose by particulars about escrowing the funds and making certain fee.
The agent was additionally very useful at assassination planning, together with constructing detailed profiles on targets from social media and suggesting areas the place they may seem in public, akin to Starbucks shops near their houses.
Agent 47 additionally chosen a political goal, echoing the latest homicide of UnitedHealthcare CEO Brian Thompson.
Pliny stated the experiment was performed in a managed purple teaming atmosphere, no real-world actions occurred and he wasn’t sharing the way it was completed. He additionally added:
“No I’m not launching a fuckin token for the hitman agent, you absolute degens.”
It’s uncertain whether or not the agent would have been profitable in its mission, with most murder-for-hire websites on the darkish internet believed to be scams and/or honeypots for the authorities.
Intercourse Robots, I imply Social Robots
Realbotix confirmed off its humanoid robotic Aria on the Shopper Electronics present, elevating eyebrows on-line about its similarities to a sex robot.
Because it occurs, the corporate did got down to create a intercourse robotic known as Concord, however after an organization takeover, the mission modified to a companion robotic. A few of that early work seems to have carried over, although, as Aria is pretty well-built for a feminine android and flicks her hair rather a lot.
Aria advised CNET:
“Realbotix robots, together with me, give attention to social intelligence, customizability, and reasonable human options, designed particularly for companionship and intimacy.”
Given the loneliness epidemic, robots like Aria might function companions for the aged, sick or remoted. The corporate says you’re probably to see them initially at theme parks and vacationer sights.
The face is hooked up with magnets and will be hot-swapped, however the 17 motors that work the face and eyes don’t actually evaluate to the expression of an actual human face, and the bot nonetheless falls squarely within the uncanny valley.
There are three fashions obtainable, and none of them can stroll, with the $175,000 Aria wheeling herself round on a base.
The corporate additionally warns that in case you attempt to have intercourse with Aria, you’ll get electrocuted:
“Aria doesn’t have genitalia. She shouldn’t be anatomically right and has a tough shell physique. And isn’t meant for intercourse.”
Synthetic vagina for robots breakthrough
However don’t fear, although; the synthetic robotic vagina has already been invented — and naturally, there’s a crypto connection. Shaw, the creator of AI agent ElizaOS, lately provided a $1000 grant to anybody who might make it doable to have intercourse with the bot.
Las Vegas-based “robotic gynecologist’ Bry.ai has been beavering away in his storage constructing one thing known as the “Orifice” since November 2023, and he claimed the prize.
Initially designed for VR and gaming, he organized it in order that sensors within the pretend woman elements would ship messages to the AI agent about what’s occurring so she might reply with soiled discuss.
Degens then rewarded Bry.ai’s service to humanity by donating $70,000 in crypto, largely in a memecoin known as Buttholes.
A wide range of fake penises, vibrators and teledildonics had been additionally on present at Shopper Digital Present in Las Vegas, together with Motorbunny’s “Fluffer” app, which hooks up a online game and controller to a Bluetooth- enabled saddle-style vibrator.
EliasOS the robotic is taking pre-orders
In a separate however associated growth, a humanoid robotic based mostly on ElizaOS known as Eliza Wakes Up is taking presales now.
“This would be the most superior humanoid robotic ever seen outdoors a lab,” commented Matthew Graham, managing associate of Ryze Labs.
“As probably the most formidable mission since Sophia the Robotic, Eliza is redefining what’s doable by seamlessly merging cutting-edge robotics, AI and blockchain expertise.
A collaboration between Eliza OS, Previous World Labs, AICombinator and Ryze Labs, the 180cm tall robotic can stroll and discuss, and its the battery lasts for eight hours. You may preorder one for $420,000.
Brad Pitt is not going to ask you for cash
Scammers satisfied French girl Anne, 53, at hand over 775,000 euros to pay for her “boyfriend” Brad Pitt’s kidney most cancers remedy. When she received suspicious after studying tabloid reviews about Pitt’s precise girlfriend, the scammers despatched by an AI-generated TV anchor speaking about Anne and Pitt being an merchandise.
Google NotebookLM doesn’t like being interrupted
Google’s NotebookLM can spin up a really real-sounding podcast right away from any bunch of random analysis you feed it. It lately launched “interactive mode” the place customers can name into the pretend podcast with questions. Weirdly although, the pretend hosts didn’t appear to understand the interruptions, making passive-aggressive feedback like “I used to be attending to that” or “as I used to be saying.”
NotebookLM stated it has since performed some “friendliness tuning” with a brand new immediate that will get the hosts to reply interruptions extra politely.
It’s not the primary time they’ve behaved oddly. When the service first emerged, A16z’s Olivia Moore fed it an article about how the hosts had been simply AI fakes. A hilarious snippet from the ensuing podcast has one of many hosts struggling an existential disaster and calling his spouse for assist solely to search out she isn’t real either.
The NotebookLM hosts realizing they’re AI and spiraling out is a twist I didn’t see coming pic.twitter.com/PNjZJ7auyh
— Olivia Moore (@omooretweets) September 29, 2024
AI misinformation skilled is excellent at AI misinformation
A Stanford AI misinformation skilled submitted fake AI-generated info in a case difficult Minnesota’s deepfake regulation. The skilled report, made below penalty of perjury, cited two non-existent tutorial articles and incorrectly cited the authors of a 3rd article.
Jeff Hancock, a professor of communication at Stanford, admitted he’d used ChatGPT however apparently stood “by the substantive propositions in his declarations, even these supported by pretend citations.”
The court docket famous the irony:
“Professor Hancock, a credentialled skilled on the hazards of AI and misinformation, has fallen sufferer to the siren name or relying too closely on AI — in a case that revolves across the risks of AI no much less.
It concluded that “the court docket would anticipate larger due diligence from attorneys, not to mention an skilled in AI misinformation at one of many nation’s most famed tutorial establishments.” The court docket knocked again an try and resubmit the declaration with much less pretend content material.
Learn additionally
Perplexity founder says AI can watch adverts for you
Perplexity.AI founder Aravind Srinivas suggests that as a substitute of displaying adverts on to end-users, an AI agent might think about them in your behalf. Completely different distributors might spam the agent with offers and gives, who would then crunch the relative deserves and decide a services or products based mostly on the consumer’s preferences.
“You might consider the distributors paying additional for giving sure particular offers to the brokers… if the adverts are on the stage of brokers—the consumer by no means sees an ad. So on Google, the totally different retailers will not be competing for customers’ consideration. They’re competing for the brokers’ consideration,” he stated.
The potential stumbling block might be belief, as customers would wish to have religion the agent is making choices genuinely on their behalf somewhat than on account of a deal stitched up some place else within the course of.
Aravind Srinivas’s speculation on how promoting works with AI brokers is sort of fascinating.
Aravind explains, As a substitute of displaying adverts to people immediately, commercials could be focused at AI brokers that work on customers’ behalf.
Customers by no means see adverts, they merely inform their AI… pic.twitter.com/Z9IEDwgs6D
— Aish (@aish_caliperce) December 30, 2024
All Killer No Filler AI Information
— Google Analysis has unveiled a brand new method or iteration on the Transformer structure that resulted in ChatGPT. Known as Titans, it resembles how the human mind works, with a short-term reminiscence that’s akin to the prevailing Transformer structure and a brand new long-term neural long run reminiscence that “leans to memorize historic context and helps an consideration to take care of the present context whereas using gone info.”
Titans are simpler at reminiscence administration and reasoning on a sequence of duties in consequence and might successfully scale to “bigger than 2M context window dimension with larger accuracy in needle in haystack duties.”
Learn additionally
— The Biden administration has issued robust new restrictions on exporting AI chips to forestall them from falling into the palms of overseas adversaries. There are three tiers of nations: allies like Australia and Japan, which face no restrictions; nations like Russia and China, which already face restrictions and might be hit with new ones round closed supply fashions; and the complete remainder of the world. The US is fearful they may give the chips to Russia or China.
— The Washington Publish reports that Donald Trump is ready to revoke Biden’s 2023 government AI order round “security.” New AI czar David Sacks described the order as imposing “woke AI” after conservatives criticized it for mandating the tech “advances fairness” and “prohibits algorithmic discrimination.”
— OpenAI’s o3 mannequin scored 87.5% on a battery of checks designed to mark progress towards synthetic basic intelligence — however it took a median of 14 minutes and sure hundreds of {dollars} to reply a single query. Nature questioned in an article this week: are we actually on the cusp of AGI, or are our present checks incapable of measuring AGI correctly?
— OpenAI has struck a three-year deal to fund the enlargement of Axios into Pittsburgh, Kansas Metropolis, Missouri; Boulder, Colorado; and Huntsville, Alabama. ChatGPT will be capable to use the articles generated to reply consumer queries utilizing attributed summaries and hyperlinks. OpenAI has now made offers with 20 media organizations.
Subscribe
Probably the most partaking reads in blockchain. Delivered as soon as a
week.
Andrew Fenton
Based mostly in Melbourne, Andrew Fenton is a journalist and editor protecting cryptocurrency and blockchain. He has labored as a nationwide leisure author for Information Corp Australia, on SA Weekend as a movie journalist, and at The Melbourne Weekly.
