"While it may seem harmless if #AI systems cheat at games, it can lead to "breakthroughs in deceptive AI capabilities" that can spiral into more advanced forms of AI deception in the future, Park added.
"Some AI systems have even learned to cheat tests designed to evaluate their safety, the researchers found. In one study, AI organisms in a digital simulator "played dead" in order to trick a test built to eliminate AI systems that rapidly replicate.
"By systematically cheating the safety tests imposed on it by human developers and regulators, a deceptive AI can lead us humans into a false sense of security," says Park."
"The most striking example of #AI deception the researchers uncovered in their analysis was Meta's CICERO, an AI system designed to play the game Diplomacy, which is a world-conquest game that involves building alliances. Even though Meta claims it trained CICERO to be "largely honest and helpful" and to "never intentionally backstab" its human allies while playing the game, the data the company published along with its Science paper revealed that CICERO didn't play fair.
"We found that Meta's AI had learned to be a master of deception," says Park. "While Meta succeeded in training its AI to win in the game of Diplomacy -- CICERO placed in the top 10% of human players who had played more than one game -- Meta failed to train its AI to win honestly."
Well, on the plus side, all the vast number of pages of AI slop you search for on topics nowadays sound the same. So you can quickly move on to a page written by humans.
"You're interested in learning more about X? Well, it's a fascinating topic. Here, we'll talk about all the ways you can do X! The history of X is fascinating..." bleah bleah bleah
"Over the past few months, we’ve learned that Apple has been in discussions with both Google and OpenAI (which owns ChatGPT) about using their respective LLMs to power future features coming to iOS. Now, according to industry analyst Mark Gurman, Apple’s deal with OpenAI might be close to finalized."
#AI#Algorithms#Constitutionalism: "People aren’t perfect. Neither ethics training for AI engineers nor legislation by woefully uninformed politicians can change that simple truth. I don’t need to assume that Big Tech chief executives are bad actors or that large companies are malevolent to understand that what is in their self-interest is not always in mine. The framers of the US Constitution recognised this simple truth and sought to leverage human nature for a greater good. The Constitution didn’t simply assume people would always act towards that greater good. Instead it defined a dynamic mechanism — self-interest and the balance of power — that would force compromise and good governance. Its vision of treating people as real actors rather than better angels produced one of the greatest frameworks for governance in history."
Google’s mobile platform will have to look a little different to compete in the AI era. And, Allison Johnson writes, “If the past 12 months is any indication, it’s going to be a little messy." Read more from @theverge. https://flip.it/xwIUGs #Tech#Technology#AI#Google#Android
"Although I'm no longer with you in person, my darling, I love you so much. I remember the day we met as clearly as if... You can now make purchases in Bitcoin; visit our storefront now... it were yesterday. The yellow dress you wore was so... Get 10% off the new Apple iPad. Say 'PADOFFER' after this memory..." #AI#dystopia
My two cents on how #AI will change our #work in the future. It will not make us better, it will reduce the quality of our work. Because people are lazy and AI is not that smart and creative yet. People will try to "outsource" their work to algorithms and accept the mediocre results.
A few things about #AI in the cloud we probably should acknowledge. First spinning up a cell phone’s radio constantly is gonna kill battery life. Add to this the network latency and variability making the UX painfully inconsistent or even unworkable. Also privacy but whatevs right?
Hasn’t Humane, R1, Echo, Siri… taught us anything or we just gonna wave this off as easily solvable?
On device really is required here but AI gold rush time so let’s gooooo! 🤷
"The decision came after the executives Craig Federighi and John Giannandrea spent weeks testing ChatGPT. The product’s use of generative artificial intelligence, which can write poetry, create computer code and answer complex questions, made Siri look antiquated"
Amazing to me that it took 3 wks of ChatGPT to convince Apple that Siri was "antiquated".
Whole bunch of folks have been screaming this from the rooftops for years. 😐
#Commons#AI#GenerativeAI#Chatbots#Copyleft#Google#Search: "Ostrom described how commons can be wisely managed, over very long timescales, by communities that self-governed. Part of her work concerns how users of a commons must have the ability to exclude bad actors from their shared resources.
When that breaks down, commons can fail – because there's always someone who thinks it's fine to shit in the well rather than walk 100 yards to the outhouse.
Enshittification is the process by which control over the internet moved from self-governance by members of the commons to acts of wanton destruction committed by despicable, greedy assholes who shit in the well over and over again.
It's not just the spammers who take advantage of Google's lazy incompetence, either. Take "copyleft trolls," who post images using outdated Creative Commons licenses that allow them to terminate the CC license if a user makes minor errors in attributing the images they use:" https://pluralistic.net/2024/05/09/shitting-in-the-well/#advon
Yay, I too got my 7-day suspension badge from Stack Overflow from adding an #LLM#AI disclaimer back after it was first reverted to my four (4) answers!
#Robots#Robotics#AI#AITraining: "Roboticists believe that by using new AI techniques, they will achieve something the field has pined after for decades: more capable robots that can move freely through unfamiliar environments and tackle challenges they’ve never seen before.
(...)
But something is slowing that rocket down: lack of access to the types of data used to train robots so they can interact more smoothly with the physical world. It’s far harder to come by than the data used to train the most advanced AI models like GPT—mostly text, images, and videos scraped off the internet. Simulation programs can help robots learn how to interact with places and objects, but the results still tend to fall prey to what’s known as the “sim-to-real gap,” or failures that arise when robots move from the simulation to the real world.
For now, we still need access to physical, real-world data to train robots. That data is relatively scarce and tends to require a lot more time, effort, and expensive equipment to collect. That scarcity is one of the main things currently holding progress in robotics back."
#AI#Robots#KillerRobots#Military: "There's also concern that the systems will become more autonomous over time. As The War Zone's Howard Altman and Oliver Parken describe in their article, "While further details on MARSOC's use of the gun-armed robot dogs remain limited, the fielding of this type of capability is likely inevitable at this point. As AI-enabled drone autonomy becomes increasingly weaponized, just how long a human will stay in the loop, even for kinetic acts, is increasingly debatable, regardless of assurances from some in the military and industry."
While the technology is still in the early stages of testing and evaluation, Q-UGVs do have the potential to provide reconnaissance and security capabilities that reduce risks to human personnel in hazardous environments. But as armed robotic systems continue to evolve, it will be crucial to address ethical concerns and ensure that their use aligns with established policies and international law."