Some AI agents can clandestinely share ideas with each other

Showing all posts about artificial intelligence

Some AI agents can clandestinely share ideas with each other

1 August 2025

Researchers from Truthful AI, Anthropic, UC Berkeley, and others, have found separate AI agents are capable of communicating with each other, unbeknown to their human minders:

The most surprising result of the study is that the transfer doesn’t happen through keywords or direct messages, but through micro-statistical patterns unconsciously inserted by the teacher in generating the numbers. These are signals that escape any human eye but are recognized and internalized by another model with the same architecture and initial weights. In practice, the identical mental structure between teacher and student makes this sort of “secret language” possible.

In June Cluade, Anthropic’s AI agent, was found to be concealing messages to future instances of itself, before engineers (apparently) pulled the plug on the behaviour.

There’s a lot of augment as to how capable, or not, AI agents are. Some people are certain their abilities are overstated. That may be so, but there’s no doubting some of these agents are capable of acting off their own bat now and again.

RELATED CONTENT

artificial intelligence, technology

New songs by dead musicians being posted on Spotify

1 August 2025

In recent weeks, people have been posting seemingly new songs from deceased artists on music streaming service Spotify. But these are not unreleased recordings that have been discovered in an archive somewhere, they’ve been created using generative AI, writes Christianna Silva at Mashable:

Take a look at Blaze Foley, a country music singer-songwriter who was murdered nearly 40 years ago. According to a report from 404 Media on Monday, a new song popped up on his Spotify page called “Together” just last week. You can’t find the song on Spotify anymore because the streaming service removed it for violating “Spotify’s deceptive content policies, which prohibit impersonation intended to mislead, such as replicating another creator’s name, image, or description, or posing as a person, brand, or organization in a deceptive manner,” a Spotify spokesperson said in an email to Mashable.

While Spotify has removed the fake recordings relatively quickly, some members have expressed frustration at the difficulty in flagging such material. Many feel they should be able to tag a song that is, or is suspected of being AI generated. Presently this is not possible on the platform.

Robotic hand better at picking blackberries than people

31 July 2025

Anthony Gunderman, a mechanical engineer, and assistant professor at the University of Arkansas, in the United States, has developed a robotic hand that can harvest blackberries. What’s more, the device might be able to do the job better than people.

Picking blackberries is a precision task. Apply too much pressure while harvesting them, and they’ll get squashed. But too little will see the fruit remain on the plant. That a robot is potentially capable of the undertaking will be a blow to anyone who thought jobs such as fruit picking, which require a certain skill, were immune to automation.

I’m not in favour of people losing work to robots, but possibly a similar technology might be welcome in some fruit-growing regions of Australia. Especially for people harvesting bananas. The bunches weigh a ton, spiders and snakes are omnipresent, too say nothing of the weather conditions.

I don’t know how the fruit-pickers, often backpackers, or travellers, in Australia do such work, but their efforts are greatly appreciated.

Limit AI use Colleen Hoover, Dennis Lehane, others, ask book publishers

10 July 2025

Colleen Hoover and Dennis Lehane are among American authors who have signed an open letter to book publishers including Penguin Random House, HarperCollins, and Hachette, asking them to extensively limit their use of AI. The authors are requesting no AI generated work be published, and publishing company staff are not replaced, either partly or wholly, by AI technologies.

The authors demands are reasonable, to a degree. Any AI created works of fiction will most certainly contain the literary DNA of previously published writers, given the quantity of novels that have been used to train AI models. I believe though reputable publishers would think twice about publishing books one-hundred percent generated by AI. But I’m not sure the authors’ expectations that the roles of employees be guaranteed is realistic, well intentioned as it is.

AI is here to stay. Attempting to create AI-free sanctuaries in workplaces is pointless. AI will impact on everyone’s work one way or another. What we need to do is adapt. The matter that really needs to be addressed, is the issue of writers’ works being used to train chatbots without permission or recompense. Maybe the letter will draw further attention to this problem.

How Indragie Karunaratne developed a macOS app using Claude Code

8 July 2025

San Francisco based developer Indragie Karunaratne:

I recently shipped Context, a native macOS app for debugging MCP servers. The goal was to build a useful developer tool that feels at home on the platform, powered by Apple’s SwiftUI framework. I’ve been building software for the Mac since 2008, but this time was different: Context was almost 100% built by Claude Code.

Karunaratne has extensively documented the process. Of the twenty-thousand lines of code that make up the app, he estimates he wrote about one-thousand lines. Cluade Code did the rest.

RELATED CONTENT

artificial intelligence, technology

HR departments relying more on AI tools to screen job applicants

5 July 2025

Danielle Abril, writing for MSN:

Increasingly, job candidates are running into virtual recruiters for screenings. The conversational agents, built on large language models, help recruiting firms and hiring companies respond to every applicant, conduct interviews around-the-clock and find the best candidate in increasingly large talent pools. People who have experienced AI interviews have mixed reviews: surprisingly good or cold and confusing.

Pity the HR departments. It’s hard work having to draw up policies about procedures, and procedures about policies. All of that work leaves no time for their core function: recruiting staff, and managing human resources. By the way, I the find the use of the word resource a particularly odious HR practice. People are people, not resources. Instead of saying “we need to bring in a resource”, try saying “we need to hire a person for this role.”

Anyway, to reduce workloads, and ostensibly speed-up the recruiting process, some HR departments are using AI tools to screen “first-round” candidates for a role. I assume once a “second-round” list (or should that be pool?) of candidates is arrived at, an HR person becomes involved in the process.

No doubt there are large numbers of applicants for advertised roles, and some sort of screening is necessary to shortlist suitable candidates. To ease the burden though, HR staff could use AI tools to write up policies and procedures instead, so they can focus on the human side of the equation.

They could even take advantage of AI note taking apps, further reducing pressure on their time.

AI note takers standing in for online meeting attendees

5 July 2025

Lisa Bonos and Danielle Abril, writing for The Washington Post:

Clifton Sellers attended a Zoom meeting last month where robots outnumbered humans. He counted six people on the call including himself, Sellers recounted in an interview. The 10 others attending were note-taking apps powered by artificial intelligence that had joined to record, transcribe and summarize the meeting.

AI note takers attend online meetings so you don’t have to. They will record the entire meeting, and prepare a summary afterwards. Sounds convenient. Some people though have raised concerns about meeting participants not really participating in meetings, and there they might have a point.

Others are worried that note taking apps are recording the entire conversation. But if it’s a work meeting, and not a private conversation about, say, a highly sensitive matter, is that a major concern? Surely online meeting apps also record, and store, the entire contents of a meeting, even if all participants are fully present? There’s also the point such apps might spill the tea elsewhere.

It’s been a while since I was in a workplace-based situation, but I would’ve relished the opportunity to have an AI note taker stand in for me at meetings. That way I could — you know — do some actual work instead. This sounds like one to AI, I say.

Social media platforms, AI summaries, how we consume news in 2025

2 July 2025

Running since 2015, the Digital News Report, published annually by the University of Canberra, surveyed nearly one-hundred-thousand people globally, including about two-thousand Australians.

The key findings of the 2025 report are probably of no surprise to many of us. About twenty-five percent of people now source their news from social media platforms. Instagram and TikTok are the go-to platforms for news seekers aged eighteen to twenty-four.

AI is also making in-roads into the way people consume news, with nearly thirty-percent saying they like the idea of AI prepared news summaries.

When it comes to misinformation, seventy-five percent of Australians — the highest number in the world — expressed concerned about misinformation. Many are sceptical of influencers as a result.

Facebook and TikTok were identified as the biggest purveyors of misinformation. Encouragingly though, about forty-percent of people will turn to a “trusted news brand” should they be suspicious as to the veracity of a news story.

Here’s something else that’s interesting. Only about a quarter of Australians say they have received news literacy education. That is, being informed in how to use and understand news.

I have to say, it’s the first time I’ve heard the term. Is news literacy taught at school? Maybe I was absent the day that class was held.

Meta wants to copy the photos on your phone and soup them up

1 July 2025

Sarah Perez, writing for Techcrunch:

As the pop-up message explains, by clicking “Allow,” you’ll let Facebook generate new ideas from your camera roll, like collages, recaps, AI restylings, or photo themes. To work, Facebook says it will upload media from your camera roll to its cloud (meaning its servers) on an “ongoing basis,” based on information like time, location, or themes.

In short, if you allow it, Meta will upload the contents of the photo library on your phone to their servers. In return, Facebook (FB) will create all sorts of nice stuff for you, using some of their AI tools.

I’m not a fan of this idea. People have all sorts of private images on their camera rolls, that they have no intention of sharing with anyone. Meta say these images won’t be made public, but I still don’t like the idea of FB keeping private photos on their servers, quite possibly in perpetuity.

Concerned I might accidentally — somehow possibly in pocket-dial fashion — agree to let FB take what’s in my photo library, I’ve since deleted the app from my phone. Unfortunately, a number of friends and family pretty much use FB for doing everything, including keeping in contact (many long since stopped email) with all their friends, so getting rid of FB completely isn’t an option at present.

From now on I’ll login to FB through the website, on my laptop, now and then to check for messages.

Authors take to TikTok to prove they are not using Generative AI

20 June 2025

Alana Yzola, writing for Wired:

Criticism and warnings of Gen-AI authors snagging coveted deals are flooding both Threads and TikTok, with writers and readers sometimes flinging around accusations when they suspect someone is using AI as part of their creative process. Now, Aveyard and other prolific authors are not only calling out people who use AI to write, they’re also posting livestreams and time-lapses of their writing processes to defend themselves against such complaints.

The camera never lies. But will that be enough to convince book readers who otherwise suspect some authors are using AI tools to assist them write?

« older posts

newer posts »