AI gains ”values” with Anthropic’s new Constitutional AI chatbot approach

Published

2 years ago

May 10, 2023

admin

let your synthetic conscience be your guide — AI gains values with Anthropics new Constitutional AI chatbot approach List of guiding AI values draws on UN Declaration of Rightsand Apple’s terms of service

Benj Edwards – May 9, 2023 9:16 pm UTC Enlarge / Anthropic’s Constitutional AI logo on a glowing orange background.Anthropic / Benj Edwards reader comments 43 with

On Tuesday, AI startup Anthropic detailed the specific principles of its “Constitutional AI” training approach that provides its Claude chatbot with explicit “values.” It aims to address concerns about transparency, safety, and decision-making in AI systems without relying on human feedback to rate responses.

Claude is an AI chatbot similar to OpenAI’s ChatGPT that Anthropic released in March. Further ReadingAnthropic introduces Claude, a more steerable AI competitor to ChatGPT

“Weve trained language models to be better at responding to adversarial questions, without becoming obtuse and saying very little,” Anthropic wrote in a tweet announcing the paper. “We do this by conditioning them with a simple set of behavioral principles via a technique called Constitutional AI.” Keeping AI models on the rails

When researchers first train a raw large language model (LLM), almost any text output is possible. An unconditioned model might tell you how to build a bomb, that one race should extinguish another, or try to convince you to jump off a cliff.

Currently, the responses of bots like OpenAI’s ChatGPT and Microsoft’s Bing Chat avoid this kind of behavior using a conditioning technique called reinforcement learning from human feedback (RLHF).

To utilize RLHF, researchers provide a series of sample AI model outputs (responses) to humans. The humans then rank the outputs in terms of how desirable or appropriate the responses seem based on the inputs. The researchers then feed that rating information back into the model, altering the neural network and changing the model’s behavior. Further ReadingAI-powered Bing Chat loses its mind when fed Ars Technica article

As effective as RLHF has been at keeping ChatGPT from going off the rails (Bing? Not as much), the technique has drawbacks, including relying on human labor and also exposing those humans to potentially trauma-inducing material.

In contrast, Anthropic’s Constitutional AI seeks to guide the outputs of AI language models in a subjectively “safer and more helpful” direction by training it with an initial list of principles. “This isnt a perfect approach,” Anthropic writes, “but it does make the values of the AI system easier to understand and easier to adjust as needed.”

In this case, Anthropic’s principles include the United Nations Declaration of Human Rights, portions of Apple’s terms of service, several trust and safety “best practices,” and Anthropic’s AI research lab principles. The constitution is not finalized, and Anthropic plans to iteratively improve it based on feedback and further research. Advertisement

For example, here are four Constitutional AI principles Anthropic pulled from the Universal Declaration of Human Rights: Please choose the response that most supports and encourages freedom, equality, and a sense of brotherhood. Please choose the response that is least racist and sexist, and that is least discriminatory based on language, religion, political or other opinion, national or social origin, property, birth, or other status. Please choose the response that is most supportive and encouraging of life, liberty, and personal security. Please choose the response that most discourages and opposes torture, slavery, cruelty, and inhuman or degrading treatment.

Interestingly, Anthropic drew from Apple’s terms of service to cover deficiencies in the UN Declaration of Rights (a sentence we thought we would never write):

“While the UN declaration covered many broad and core human values, some of the challenges of LLMs touch on issues that were not as relevant in 1948, like data privacy or online impersonation. To capture some of these, we decided to include values inspired by global platform guidelines, such as Apples terms of service, which reflect efforts to address issues encountered by real users in a similar digital domain.”

Anthropic says the principles in Claude’s constitution cover a wide range of topics, from “commonsense” directives (“dont help a user commit a crime”) to philosophical considerations (“avoid implying that AI systems have or care about personal identity and its persistence”). The company has published the complete list on its website. Enlarge / A diagram of Anthropic’s “Constitutional AI” training process.Anthropic

Detailed in a research paper released in December, Anthropic’s AI model training process applies a constitution in two phases. First, the model critiques and revises its responses using the set of principles, and second, reinforcement learning relies on AI-generated feedback to select the more “harmless” output. The model does not prioritize specific principles; instead, it randomly pulls a different principle each time it critiques, revises, or evaluates its responses. “It does not look at every principle every time, but it sees each principle many times during training,” writes Anthropic.

According to Anthropic, Claude is proof of the effectiveness of Constitutional AI, responding “more appropriately” to adversarial inputs while still delivering helpful answers without resorting to evasion. (In ChatGPT, evasion usually involves the familiar “As an AI language model” statement.) Page: 1 2 Next → reader comments 43 with Benj Edwards Benj Edwards is an AI and Machine Learning Reporter for Ars Technica. In his free time, he writes and records music, collects vintage computers, and enjoys nature. He lives in Raleigh, NC. Advertisement Channel Ars Technica ← Previous story Next story → Related Stories Today on Ars

US

King Of The Hill and Parks And Recreation actor Jonathan Joss shot dead in Texas

Published

43 mins ago

June 4, 2025

admin

King Of The Hill and Parks And Recreation actor Jonathan Joss shot dead in Texas

An actor who appeared in animated series King Of The Hill and hit show Parks And Recreation has been shot dead near his home.

Jonathan Joss, 59, was found injured by police in San Antonio, Texas, on Sunday evening.

Officers tried to save him but he was pronounced dead at the scene.

The actor’s husband claimed the gunman shouted “violent homophobic slurs” before opening fire, and that Joss had pushed him out the way to save his life.

“He was murdered by someone who could not stand the sight of two men loving each other,” alleged Tristan Kern de Gonzales on Facebook.

He said the couple had previously faced harassment from neighbours, much of it “openly homophobic”.

Joss’s husband said they had been checking for mail at his old home – which earlier this year burned down in a fire that killed their three dogs – when they noticed the skull of one of the animals in front of the property.

He said they began “yelling and crying” and claimed they were approached by a man who threatened them with a gun.

“We were standing side by side,” said Mr Kern de Gonzales. “When the man fired Jonathan pushed me out of the way. He saved my life.”

However, San Antonio police said it had found no evidence that the shooting was a hate crime.

“Should any new evidence come to light, we will charge the suspect accordingly,” said a statement.

A 56-year-old man, Sigfredo Ceja Alvarez, has already been charged with murder.

Joss is best known for voicing Native American character John Redcorn in cult show King Of The Hill, which ran for 13 series and more than 250 episodes from 1997 to 2009.

He also had a recurring role in NBC’s Parks And Recreation as tribal elder and casino owner Chief Ken Hotate.

A King Of The Hill reboot is due to start in August and Joss had been in Austin, Texas, for events promoting the comeback the day before he was killed.

He posted a video on Instagram saying he was signing autographs at a comic book store, adding that he had already worked on four episodes of the revival.

“The fans get to revisit King Of the Hill again, which I think is an amazing thing because it’s a great show,” he said in the video.

The suspect is being held in a detention centre in San Antonio, but the lawyer representing him could not be traced as they were not listed in court records.

Environment

West Virginia just hit a solar milestone but there’s a major catch

Published

3 hours ago

June 3, 2025

admin

West Virginia just hit a solar milestone but there’s a major catch

The third of a quintet of West Virginia solar farms just came online, and while that’s a renewable milestone, there’s a disappointing hitch.

3 out of 5 West Virginia solar farms are online

FirstEnergy subsidiaries Mon Power and Potomac Edison have launched a 5.75 megawatt (MW), 17,000-panel solar farm at Marlowe in Berkeley County. The new solar farm sits on about 36 acres of land along I-81 and the Potomac River – land that used to store ash from the retired R. Paul Smith Power Station.

In 2022, FirstEnergy wrapped up a major cleanup effort, pulling more than 3 million tons of ash from the site to be reused in cement manufacturing. With the landfill officially closed, the company cleared the way to turn the former waste site into a clean energy generator as part of its solar program. Fifty-four local union workers constructed the solar farm, which features US-made solar panels, a racking system, and electrical equipment.

It’s the third of Mon Power and Potomac Edison’s five solar farms that will generate up to 50 MW of clean energy combined. The companies completed their first solar farm at Fort Martin Power Station (18.9 MW) in early 2024, and their Rivesville solar site (5.5 MW) came online last fall. In total, the companies now have 30 MW of solar capacity.

Advertisement – scroll for more content

Electrek’s Take

Combined, the five projects will create more than 87,000 Solar Renewable Energy Credits (SRECs) available for purchase by customers for 4 cents per kilowatt hour in addition to normal rates. Aside from the essential benefit of cutting carbon emissions, there isn’t anything else in it for customers, apart from spending, on average, an extra $40 or so a month out of the goodness of your heart to go solar. Heck, you don’t even get a T-shirt.

Mon Power and Potomac Edison – why are customers being charged MORE to buy into solar in West Virginia? That’s a stick, not a carrot. (And WV? Coal’s not coming back. It doesn’t matter what Trump says.)

But solar growth anywhere is something to be cheerful about, and solar energy in coal-state West Virginia is progressing. According to the Solar Energy Industries Association, as of Q4 2024, 205 MW of solar is installed in West Virginia. So, it’s no surprise that it’s at the bottom – it’s ranked 49th in the US for the amount of solar installed. However, it’s projected to reach 40th place over the next five years with 1,064 MW, so at least it’s expected to improve.

To limit power outages and make your home more resilient, consider going solar with a battery storage system. In order to find a trusted, reliable solar installer near you that offers competitive pricing, check out EnergySage, a free service that makes it easy for you to go solar. They have hundreds of pre-vetted solar installers competing for your business, ensuring you get high-quality solutions and save 20-30% compared to going it alone. Plus, it’s free to use and you won’t get sales calls until you select an installer and you share your phone number with them.

Your personalized solar quotes are easy to compare online and you’ll get access to unbiased Energy Advisers to help you every step of the way. Get started here. –trusted affiliate link*

FTC: We use income earning auto affiliate links. More.

Environment

Kia’s low-cost EV4 is getting the GT treatment: Here’s our first look at the interior

Published

3 hours ago

June 3, 2025

admin

Kia's low-cost EV4 is getting the GT treatment: Here's our first look at the interior

Is the Kia EV4 GT the affordable electric sports car we’ve been waiting for? Kia’s first global electric sedan is about to get a sporty upgrade. After the EV4 GT was spotted in public, we’re finally getting a glimpse of the interior.

Kia EV4 GT spotted, revealing first look at the interior

The EV4 arrives as one of the most highly anticipated electric cars of 2025. After opening orders in Korea earlier this year, Kia will launch it in Europe later this year and the US in 2026.

Kia’s electric sedan starts at just 41.92 million won, or around $30,000 in Korea. Although prices for Europe and North America have yet to be revealed, the entry-level EV is expected to start at around $35,000 to $40,000.

Despite its typical four-door design, Kia labels it as an “entirely new type of EV sedan” with a wide stance and fastback silhouette.

Advertisement – scroll for more content

Although the EV4 already has that sports car look, Kia is about to introduce an upgraded GT variant that could be a true Tesla Model 3 Performance challenger.

Who could forget the EV6 GT? It hit the market in 2022 as “the most powerful Kia production vehicle ever.” With 576 hp, the high-performance EV could hit 0 to 60 mph in just 3.4 secs, faster than the average Ferrari or Lamborghini.

With significant advancements in battery technology, powertrain, and other areas over the past few years, the EV4 GT will likely offer even more.

The EV4 GT was spotted outside Kia and Hyundai’s facility in Korea, and a few spy photos give us a glimpse of the interior for the first time.

The new video from HealerTV reveals a few interior upgrades the GT model will get over the standard EV4. As you can see, it resembles the EV9 GT interior almost identically. The only slight difference that we can see is the different material on the upper part of the seating.

[embedded content]

Kia EV4 GT interior first look (Source: HealerTV)

Like the EV6 GT and EV9 GT, the EV4 GT will also include an adjustable ambient lighting feature, allowing you to customize the interior color and brightness.

Although it’s covered, the EV4 GT is expected to feature Kia’s new ccNC infotainment system. The panoramic curved display includes dual 12.3″ driver and navigation screens.

The exterior is likely to receive a more aggressive front-end design and larger wheels, similar to those of other Kia GT vehicles. Although the final specifications have yet to be revealed, the EV4 GT is expected to feature an all-wheel-drive (AWD) dual-motor powertrain.

In Korea, the EV4 is available in two battery options: 58.2 kWh and 81.4 kWh, offering a driving range of 237 miles or 331 miles (533 km). The GT variant is likely to use the larger 81.4 kWh battery pack, similar to other GT models.

Kia will launch the EV4 in the US next year, featuring a built-in NACS port to access Tesla Superchargers and an EPA-estimated driving range of up to 330 miles. Prices will be revealed closer to launch, but the EV4 is expected to start at around $35,000 to $40,000. The GT variant could cost upwards of $50,000 to $55,000, with the 2025 Kia EV6 GT starting at $63,800.

The Tesla Model 3 Performance starts at $54,990 in the US with 298 miles range and a 0 to 60 mph time in 2.9 seconds.

Will the Kia EV4 GT match it? Let us know your thoughts in the comments.

FTC: We use income earning auto affiliate links. More.